refactor: 移除独立 hub/proxy/executor/gateway crate,统一为 gateway tunnel 架构

- 删除 aether-hub、aether-proxy 独立项目及其 Dockerfile/配置
- 删除 crates/aether-executor 和 crates/aether-gateway 全部模块
- 新增 apps/ 目录作为应用入口
- 将 hub 概念重构为 gateway tunnel transport
- 将 executor 重构为 execution runtime
- 新增 tunnel.rs 合约定义和 testkit tunnel/execution_runtime 模块
- 更新 Python 服务层和测试适配新架构命名
This commit is contained in:
fawney19
2026-04-03 14:59:58 +08:00
parent ddf18fed9a
commit 8f26e1a31f
983 changed files with 103097 additions and 105836 deletions
@@ -0,0 +1,294 @@
use tracing::warn;
use crate::gateway::provider_transport::resolve_transport_proxy_snapshot;
use crate::gateway::scheduler::GatewayMinimalCandidateSelectionCandidate;
use crate::gateway::AppState;
#[derive(Debug, Clone, Copy, PartialEq, Eq, PartialOrd, Ord)]
enum TunnelOwnerAffinityBucket {
LocalTunnel = 0,
Neutral = 1,
RemoteTunnel = 2,
}
pub(crate) async fn prefer_local_tunnel_owner_candidates(
state: &AppState,
candidates: Vec<GatewayMinimalCandidateSelectionCandidate>,
) -> Vec<GatewayMinimalCandidateSelectionCandidate> {
let mut ranked = Vec::with_capacity(candidates.len());
for (original_index, candidate) in candidates.into_iter().enumerate() {
let bucket = resolve_candidate_tunnel_owner_affinity(state, &candidate).await;
ranked.push((bucket, original_index, candidate));
}
ranked.sort_by(|left, right| left.0.cmp(&right.0).then(left.1.cmp(&right.1)));
ranked
.into_iter()
.map(|(_, _, candidate)| candidate)
.collect()
}
async fn resolve_candidate_tunnel_owner_affinity(
state: &AppState,
candidate: &GatewayMinimalCandidateSelectionCandidate,
) -> TunnelOwnerAffinityBucket {
let transport = match state
.read_provider_transport_snapshot(
&candidate.provider_id,
&candidate.endpoint_id,
&candidate.key_id,
)
.await
{
Ok(Some(transport)) => transport,
Ok(None) => return TunnelOwnerAffinityBucket::Neutral,
Err(error) => {
warn!(
provider_id = %candidate.provider_id,
endpoint_id = %candidate.endpoint_id,
key_id = %candidate.key_id,
error = ?error,
"failed to load provider transport while evaluating tunnel owner affinity"
);
return TunnelOwnerAffinityBucket::Neutral;
}
};
let Some(proxy) = resolve_transport_proxy_snapshot(&transport) else {
return TunnelOwnerAffinityBucket::Neutral;
};
if proxy.enabled == Some(false) {
return TunnelOwnerAffinityBucket::Neutral;
}
let Some(node_id) = proxy
.node_id
.as_deref()
.map(str::trim)
.filter(|value| !value.is_empty())
else {
return TunnelOwnerAffinityBucket::Neutral;
};
if state.tunnel.has_local_proxy(node_id) {
return TunnelOwnerAffinityBucket::LocalTunnel;
}
match state
.tunnel
.lookup_attachment_owner(state.data.as_ref(), node_id)
.await
{
Ok(Some(owner)) if owner.gateway_instance_id == state.tunnel.local_instance_id() => {
TunnelOwnerAffinityBucket::LocalTunnel
}
Ok(Some(_)) => TunnelOwnerAffinityBucket::RemoteTunnel,
Ok(None) => TunnelOwnerAffinityBucket::Neutral,
Err(error) => {
warn!(
node_id = node_id,
error = %error,
"failed to load tunnel attachment owner while evaluating scheduler affinity"
);
TunnelOwnerAffinityBucket::Neutral
}
}
}
#[cfg(test)]
mod tests {
use std::time::{SystemTime, UNIX_EPOCH};
use aether_data::repository::provider_catalog::{
InMemoryProviderCatalogReadRepository, StoredProviderCatalogEndpoint,
StoredProviderCatalogKey, StoredProviderCatalogProvider,
};
use serde_json::json;
use super::*;
use crate::gateway::tunnel::TunnelAttachmentRecord;
use crate::gateway::GatewayDataState;
fn sample_candidate(
endpoint_id: &str,
key_id: &str,
) -> GatewayMinimalCandidateSelectionCandidate {
GatewayMinimalCandidateSelectionCandidate {
provider_id: "provider-1".to_string(),
provider_name: "provider-1".to_string(),
provider_type: "custom".to_string(),
provider_priority: 0,
endpoint_id: endpoint_id.to_string(),
endpoint_api_format: "openai:chat".to_string(),
key_id: key_id.to_string(),
key_name: key_id.to_string(),
key_auth_type: "api_key".to_string(),
key_internal_priority: 0,
key_global_priority_for_format: Some(0),
key_capabilities: None,
model_id: "model-1".to_string(),
global_model_id: "global-model-1".to_string(),
global_model_name: "gpt-4.1".to_string(),
selected_provider_model_name: "gpt-4.1".to_string(),
mapping_matched_model: None,
}
}
fn sample_provider() -> StoredProviderCatalogProvider {
StoredProviderCatalogProvider::new(
"provider-1".to_string(),
"provider-1".to_string(),
Some("https://provider.example".to_string()),
"custom".to_string(),
)
.expect("provider should build")
.with_transport_fields(true, false, false, None, None, None, None, None, None)
}
fn sample_endpoint(id: &str) -> StoredProviderCatalogEndpoint {
StoredProviderCatalogEndpoint::new(
id.to_string(),
"provider-1".to_string(),
"openai:chat".to_string(),
Some("openai".to_string()),
Some("chat".to_string()),
true,
)
.expect("endpoint should build")
.with_transport_fields(
"https://api.provider.example".to_string(),
None,
None,
None,
None,
None,
None,
None,
)
.expect("endpoint transport should build")
}
fn sample_key(id: &str, node_id: &str) -> StoredProviderCatalogKey {
StoredProviderCatalogKey::new(
id.to_string(),
"provider-1".to_string(),
id.to_string(),
"api_key".to_string(),
None,
true,
)
.expect("key should build")
.with_transport_fields(
Some(json!(["openai:chat"])),
"plain-upstream-key".to_string(),
None,
None,
Some(json!({"openai:chat": 1})),
None,
None,
Some(json!({
"enabled": true,
"mode": "tunnel",
"node_id": node_id,
})),
None,
)
.expect("key transport should build")
}
fn tunnel_attachment_key(node_id: &str) -> String {
format!("tunnel.attachments.{node_id}")
}
fn current_unix_secs() -> u64 {
SystemTime::now()
.duration_since(UNIX_EPOCH)
.unwrap_or_default()
.as_secs()
}
#[tokio::test]
async fn prefers_local_tunnel_owner_candidates_before_remote_tunnel_candidates() {
let provider_catalog = InMemoryProviderCatalogReadRepository::seed(
vec![sample_provider()],
vec![
sample_endpoint("endpoint-remote"),
sample_endpoint("endpoint-local"),
],
vec![
sample_key("key-remote", "node-remote"),
sample_key("key-local", "node-local"),
],
);
let observed_at_unix_secs = current_unix_secs();
let data_state = GatewayDataState::with_provider_transport_reader_for_tests(
std::sync::Arc::new(provider_catalog),
"development-key",
)
.with_system_config_values_for_tests(vec![
(
tunnel_attachment_key("node-remote"),
serde_json::to_value(TunnelAttachmentRecord {
gateway_instance_id: "gateway-b".to_string(),
relay_base_url: "http://gateway-b:8080".to_string(),
conn_count: 1,
observed_at_unix_secs,
})
.expect("remote attachment should serialize"),
),
(
tunnel_attachment_key("node-local"),
serde_json::to_value(TunnelAttachmentRecord {
gateway_instance_id: "gateway-a".to_string(),
relay_base_url: "http://gateway-a:8080".to_string(),
conn_count: 1,
observed_at_unix_secs,
})
.expect("local attachment should serialize"),
),
]);
let state = AppState::new("http://127.0.0.1:1")
.expect("state should build")
.with_data_state_for_tests(data_state)
.with_tunnel_identity_for_tests("gateway-a", Some("http://gateway-a:8080"));
let reordered = prefer_local_tunnel_owner_candidates(
&state,
vec![
sample_candidate("endpoint-remote", "key-remote"),
sample_candidate("endpoint-local", "key-local"),
],
)
.await;
assert_eq!(reordered[0].endpoint_id, "endpoint-local");
assert_eq!(reordered[1].endpoint_id, "endpoint-remote");
}
#[tokio::test]
async fn leaves_candidate_order_unchanged_when_transport_has_no_tunnel_proxy() {
let provider_catalog = InMemoryProviderCatalogReadRepository::seed(
vec![sample_provider()],
vec![sample_endpoint("endpoint-a"), sample_endpoint("endpoint-b")],
vec![sample_key("key-a", ""), sample_key("key-b", "")],
);
let data_state = GatewayDataState::with_provider_transport_reader_for_tests(
std::sync::Arc::new(provider_catalog),
"development-key",
);
let state = AppState::new("http://127.0.0.1:1")
.expect("state should build")
.with_data_state_for_tests(data_state)
.with_tunnel_identity_for_tests("gateway-a", Some("http://gateway-a:8080"));
let reordered = prefer_local_tunnel_owner_candidates(
&state,
vec![
sample_candidate("endpoint-a", "key-a"),
sample_candidate("endpoint-b", "key-b"),
],
)
.await;
assert_eq!(reordered[0].endpoint_id, "endpoint-a");
assert_eq!(reordered[1].endpoint_id, "endpoint-b");
}
}
@@ -0,0 +1,57 @@
use axum::body::Bytes;
use base64::Engine as _;
use crate::gateway::headers::is_json_request;
pub(crate) const GEMINI_FILES_GET_PLAN_KIND: &str = "gemini_files_get";
pub(crate) const GEMINI_FILES_UPLOAD_PLAN_KIND: &str = "gemini_files_upload";
pub(crate) const GEMINI_FILES_LIST_PLAN_KIND: &str = "gemini_files_list";
pub(crate) const GEMINI_FILES_DELETE_PLAN_KIND: &str = "gemini_files_delete";
pub(crate) const GEMINI_FILES_DOWNLOAD_PLAN_KIND: &str = "gemini_files_download";
pub(crate) const OPENAI_VIDEO_CONTENT_PLAN_KIND: &str = "openai_video_content";
pub(crate) const OPENAI_VIDEO_CANCEL_SYNC_PLAN_KIND: &str = "openai_video_cancel_sync";
pub(crate) const OPENAI_VIDEO_REMIX_SYNC_PLAN_KIND: &str = "openai_video_remix_sync";
pub(crate) const OPENAI_VIDEO_DELETE_SYNC_PLAN_KIND: &str = "openai_video_delete_sync";
pub(crate) const GEMINI_VIDEO_CREATE_SYNC_PLAN_KIND: &str = "gemini_video_create_sync";
pub(crate) const GEMINI_VIDEO_CANCEL_SYNC_PLAN_KIND: &str = "gemini_video_cancel_sync";
pub(crate) const OPENAI_CHAT_STREAM_PLAN_KIND: &str = "openai_chat_stream";
pub(crate) const CLAUDE_CHAT_STREAM_PLAN_KIND: &str = "claude_chat_stream";
pub(crate) const GEMINI_CHAT_STREAM_PLAN_KIND: &str = "gemini_chat_stream";
pub(crate) const OPENAI_CLI_STREAM_PLAN_KIND: &str = "openai_cli_stream";
pub(crate) const OPENAI_COMPACT_STREAM_PLAN_KIND: &str = "openai_compact_stream";
pub(crate) const CLAUDE_CLI_STREAM_PLAN_KIND: &str = "claude_cli_stream";
pub(crate) const GEMINI_CLI_STREAM_PLAN_KIND: &str = "gemini_cli_stream";
pub(crate) const OPENAI_VIDEO_CREATE_SYNC_PLAN_KIND: &str = "openai_video_create_sync";
pub(crate) const OPENAI_CHAT_SYNC_PLAN_KIND: &str = "openai_chat_sync";
pub(crate) const OPENAI_CLI_SYNC_PLAN_KIND: &str = "openai_cli_sync";
pub(crate) const OPENAI_COMPACT_SYNC_PLAN_KIND: &str = "openai_compact_sync";
pub(crate) const CLAUDE_CHAT_SYNC_PLAN_KIND: &str = "claude_chat_sync";
pub(crate) const GEMINI_CHAT_SYNC_PLAN_KIND: &str = "gemini_chat_sync";
pub(crate) const CLAUDE_CLI_SYNC_PLAN_KIND: &str = "claude_cli_sync";
pub(crate) const GEMINI_CLI_SYNC_PLAN_KIND: &str = "gemini_cli_sync";
pub(crate) const EXECUTION_RUNTIME_SYNC_ACTION: &str = "execution_runtime_sync";
pub(crate) const EXECUTION_RUNTIME_SYNC_DECISION_ACTION: &str = "execution_runtime_sync_decision";
pub(crate) const EXECUTION_RUNTIME_STREAM_ACTION: &str = "execution_runtime_stream";
pub(crate) const EXECUTION_RUNTIME_STREAM_DECISION_ACTION: &str =
"execution_runtime_stream_decision";
pub(crate) fn parse_direct_request_body(
parts: &http::request::Parts,
body_bytes: &Bytes,
) -> Option<(serde_json::Value, Option<String>)> {
if is_json_request(&parts.headers) {
if body_bytes.is_empty() {
Some((serde_json::json!({}), None))
} else {
serde_json::from_slice::<serde_json::Value>(body_bytes)
.ok()
.map(|value| (value, None))
}
} else {
Some((
serde_json::json!({}),
(!body_bytes.is_empty())
.then(|| base64::engine::general_purpose::STANDARD.encode(body_bytes)),
))
}
}
@@ -0,0 +1,162 @@
use std::collections::BTreeMap;
use aether_contracts::{ExecutionPlan, ExecutionTimeouts, ProxySnapshot};
use serde::{Deserialize, Serialize};
use tracing::warn;
use crate::gateway::headers::collect_control_headers;
use crate::gateway::{AppState, GatewayControlAuthContext, GatewayControlDecision, GatewayError};
#[derive(Debug, Serialize)]
pub(crate) struct GatewayControlPlanRequest {
pub(crate) trace_id: String,
pub(crate) method: String,
pub(crate) path: String,
pub(crate) query_string: Option<String>,
pub(crate) headers: BTreeMap<String, String>,
pub(crate) body_json: serde_json::Value,
#[serde(skip_serializing_if = "Option::is_none")]
pub(crate) body_base64: Option<String>,
pub(crate) auth_context: Option<GatewayControlAuthContext>,
}
#[derive(Debug, Deserialize, Serialize)]
pub(crate) struct GatewayControlPlanResponse {
pub(crate) action: String,
#[serde(default)]
pub(crate) plan_kind: Option<String>,
#[serde(default)]
pub(crate) plan: Option<ExecutionPlan>,
#[serde(default)]
pub(crate) report_kind: Option<String>,
#[serde(default)]
pub(crate) report_context: Option<serde_json::Value>,
#[serde(default)]
pub(crate) auth_context: Option<GatewayControlAuthContext>,
}
#[derive(Debug, Deserialize, Serialize)]
pub(crate) struct GatewayControlSyncDecisionResponse {
pub(crate) action: String,
#[serde(default)]
pub(crate) decision_kind: Option<String>,
#[serde(default)]
pub(crate) execution_strategy: Option<String>,
#[serde(default)]
pub(crate) conversion_mode: Option<String>,
#[serde(default)]
pub(crate) request_id: Option<String>,
#[serde(default)]
pub(crate) candidate_id: Option<String>,
#[serde(default)]
pub(crate) provider_name: Option<String>,
#[serde(default)]
pub(crate) provider_id: Option<String>,
#[serde(default)]
pub(crate) endpoint_id: Option<String>,
#[serde(default)]
pub(crate) key_id: Option<String>,
#[serde(default)]
pub(crate) upstream_base_url: Option<String>,
#[serde(default)]
pub(crate) upstream_url: Option<String>,
#[serde(default)]
pub(crate) provider_request_method: Option<String>,
#[serde(default)]
pub(crate) auth_header: Option<String>,
#[serde(default)]
pub(crate) auth_value: Option<String>,
#[serde(default)]
pub(crate) provider_api_format: Option<String>,
#[serde(default)]
pub(crate) client_api_format: Option<String>,
#[serde(default)]
pub(crate) provider_contract: Option<String>,
#[serde(default)]
pub(crate) client_contract: Option<String>,
#[serde(default)]
pub(crate) model_name: Option<String>,
#[serde(default)]
pub(crate) mapped_model: Option<String>,
#[serde(default)]
pub(crate) prompt_cache_key: Option<String>,
#[serde(default)]
pub(crate) extra_headers: BTreeMap<String, String>,
#[serde(default)]
pub(crate) provider_request_headers: BTreeMap<String, String>,
#[serde(default)]
pub(crate) provider_request_body: Option<serde_json::Value>,
#[serde(default)]
pub(crate) provider_request_body_base64: Option<String>,
#[serde(default)]
pub(crate) content_type: Option<String>,
#[serde(default)]
pub(crate) proxy: Option<ProxySnapshot>,
#[serde(default)]
pub(crate) tls_profile: Option<String>,
#[serde(default)]
pub(crate) timeouts: Option<ExecutionTimeouts>,
#[serde(default)]
pub(crate) upstream_is_stream: bool,
#[serde(default)]
pub(crate) report_kind: Option<String>,
#[serde(default)]
pub(crate) report_context: Option<serde_json::Value>,
#[serde(default)]
pub(crate) auth_context: Option<GatewayControlAuthContext>,
}
fn decision_has_exact_provider_request(payload: &GatewayControlSyncDecisionResponse) -> bool {
!payload.provider_request_headers.is_empty()
&& (payload.provider_request_body.is_some()
|| payload
.provider_request_body_base64
.as_ref()
.map(|value| !value.trim().is_empty())
.unwrap_or(false))
}
pub(crate) fn generic_decision_missing_exact_provider_request(
payload: &GatewayControlSyncDecisionResponse,
) -> bool {
if decision_has_exact_provider_request(payload) {
return false;
}
warn!(
decision_kind = payload.decision_kind.as_deref().unwrap_or_default(),
provider_api_format = payload.provider_api_format.as_deref().unwrap_or_default(),
client_api_format = payload.client_api_format.as_deref().unwrap_or_default(),
"gateway generic decision missing exact provider request; falling back to plan"
);
true
}
pub(crate) async fn build_gateway_plan_request(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: serde_json::Value,
body_base64: Option<String>,
) -> Result<GatewayControlPlanRequest, GatewayError> {
let auth_context = crate::gateway::resolve_execution_runtime_auth_context(
state,
decision,
&parts.headers,
&parts.uri,
trace_id,
)
.await?;
Ok(GatewayControlPlanRequest {
trace_id: trace_id.to_string(),
method: parts.method.to_string(),
path: parts.uri.path().to_string(),
query_string: parts.uri.query().map(ToOwned::to_owned),
headers: collect_control_headers(&parts.headers),
body_json,
body_base64,
auth_context,
})
}
@@ -0,0 +1,175 @@
use crate::gateway::ai_pipeline::planner::common::{
CLAUDE_CHAT_STREAM_PLAN_KIND, CLAUDE_CHAT_SYNC_PLAN_KIND, CLAUDE_CLI_STREAM_PLAN_KIND,
CLAUDE_CLI_SYNC_PLAN_KIND, EXECUTION_RUNTIME_STREAM_ACTION, EXECUTION_RUNTIME_SYNC_ACTION,
GEMINI_CHAT_STREAM_PLAN_KIND, GEMINI_CHAT_SYNC_PLAN_KIND, GEMINI_CLI_STREAM_PLAN_KIND,
GEMINI_CLI_SYNC_PLAN_KIND, GEMINI_FILES_DELETE_PLAN_KIND, GEMINI_FILES_DOWNLOAD_PLAN_KIND,
GEMINI_FILES_GET_PLAN_KIND, GEMINI_FILES_LIST_PLAN_KIND, GEMINI_VIDEO_CANCEL_SYNC_PLAN_KIND,
GEMINI_VIDEO_CREATE_SYNC_PLAN_KIND, OPENAI_CHAT_STREAM_PLAN_KIND, OPENAI_CHAT_SYNC_PLAN_KIND,
OPENAI_CLI_STREAM_PLAN_KIND, OPENAI_CLI_SYNC_PLAN_KIND, OPENAI_COMPACT_STREAM_PLAN_KIND,
OPENAI_COMPACT_SYNC_PLAN_KIND, OPENAI_VIDEO_CANCEL_SYNC_PLAN_KIND,
OPENAI_VIDEO_CONTENT_PLAN_KIND, OPENAI_VIDEO_CREATE_SYNC_PLAN_KIND,
OPENAI_VIDEO_DELETE_SYNC_PLAN_KIND, OPENAI_VIDEO_REMIX_SYNC_PLAN_KIND,
};
use crate::gateway::ai_pipeline::planner::plan_builders::{
build_gemini_stream_plan_from_decision, build_gemini_sync_plan_from_decision,
build_openai_chat_stream_plan_from_decision, build_openai_chat_sync_plan_from_decision,
build_openai_cli_stream_plan_from_decision, build_openai_cli_sync_plan_from_decision,
build_passthrough_stream_plan_from_decision, build_passthrough_sync_plan_from_decision,
build_standard_stream_plan_from_decision, build_standard_sync_plan_from_decision,
LocalStreamPlanAndReport, LocalSyncPlanAndReport,
};
use crate::gateway::{
AppState, GatewayControlAuthContext, GatewayControlDecision, GatewayControlPlanResponse,
GatewayControlSyncDecisionResponse, GatewayError,
};
pub(crate) async fn maybe_build_sync_plan_payload_impl(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
body_base64: Option<&str>,
body_is_empty: bool,
) -> Result<Option<GatewayControlPlanResponse>, GatewayError> {
let Some(plan_kind) = super::resolve_sync_plan_kind(parts, decision) else {
return Ok(None);
};
let Some(payload) = super::maybe_build_sync_decision_payload(
state,
parts,
trace_id,
decision,
body_json,
body_base64,
body_is_empty,
)
.await?
else {
return Ok(None);
};
build_sync_plan_payload_from_decision(parts, body_json, plan_kind, payload)
}
pub(crate) async fn maybe_build_stream_plan_payload_impl(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
) -> Result<Option<GatewayControlPlanResponse>, GatewayError> {
let Some(plan_kind) = super::resolve_stream_plan_kind(parts, decision) else {
return Ok(None);
};
let Some(payload) =
super::maybe_build_stream_decision_payload(state, parts, trace_id, decision, body_json)
.await?
else {
return Ok(None);
};
build_stream_plan_payload_from_decision(parts, body_json, plan_kind, payload)
}
fn build_sync_plan_payload_from_decision(
parts: &http::request::Parts,
body_json: &serde_json::Value,
plan_kind: &str,
payload: GatewayControlSyncDecisionResponse,
) -> Result<Option<GatewayControlPlanResponse>, GatewayError> {
let auth_context = payload.auth_context.clone();
let plan_and_report = match plan_kind {
OPENAI_CHAT_SYNC_PLAN_KIND => {
build_openai_chat_sync_plan_from_decision(parts, body_json, payload)?
}
OPENAI_CLI_SYNC_PLAN_KIND => {
build_openai_cli_sync_plan_from_decision(parts, body_json, payload, false)?
}
OPENAI_COMPACT_SYNC_PLAN_KIND => {
build_openai_cli_sync_plan_from_decision(parts, body_json, payload, true)?
}
CLAUDE_CHAT_SYNC_PLAN_KIND | CLAUDE_CLI_SYNC_PLAN_KIND => {
build_standard_sync_plan_from_decision(parts, body_json, payload)?
}
GEMINI_CHAT_SYNC_PLAN_KIND | GEMINI_CLI_SYNC_PLAN_KIND => {
build_gemini_sync_plan_from_decision(parts, body_json, payload)?
}
OPENAI_VIDEO_CREATE_SYNC_PLAN_KIND
| OPENAI_VIDEO_REMIX_SYNC_PLAN_KIND
| OPENAI_VIDEO_CANCEL_SYNC_PLAN_KIND
| OPENAI_VIDEO_DELETE_SYNC_PLAN_KIND
| GEMINI_VIDEO_CREATE_SYNC_PLAN_KIND
| GEMINI_VIDEO_CANCEL_SYNC_PLAN_KIND
| GEMINI_FILES_LIST_PLAN_KIND
| GEMINI_FILES_GET_PLAN_KIND
| GEMINI_FILES_DELETE_PLAN_KIND => {
build_passthrough_sync_plan_from_decision(parts, payload)?
}
_ => None,
};
Ok(plan_and_report.map(|value| build_sync_plan_response(plan_kind, value, auth_context)))
}
fn build_stream_plan_payload_from_decision(
parts: &http::request::Parts,
body_json: &serde_json::Value,
plan_kind: &str,
payload: GatewayControlSyncDecisionResponse,
) -> Result<Option<GatewayControlPlanResponse>, GatewayError> {
let auth_context = payload.auth_context.clone();
let plan_and_report = match plan_kind {
OPENAI_CHAT_STREAM_PLAN_KIND => {
build_openai_chat_stream_plan_from_decision(parts, body_json, payload)?
}
OPENAI_CLI_STREAM_PLAN_KIND => {
build_openai_cli_stream_plan_from_decision(parts, body_json, payload, false)?
}
OPENAI_COMPACT_STREAM_PLAN_KIND => {
build_openai_cli_stream_plan_from_decision(parts, body_json, payload, true)?
}
CLAUDE_CHAT_STREAM_PLAN_KIND | CLAUDE_CLI_STREAM_PLAN_KIND => {
build_standard_stream_plan_from_decision(parts, body_json, payload, true)?
}
GEMINI_CHAT_STREAM_PLAN_KIND | GEMINI_CLI_STREAM_PLAN_KIND => {
build_gemini_stream_plan_from_decision(parts, body_json, payload)?
}
OPENAI_VIDEO_CONTENT_PLAN_KIND | GEMINI_FILES_DOWNLOAD_PLAN_KIND => {
build_passthrough_stream_plan_from_decision(parts, payload)?
}
_ => None,
};
Ok(plan_and_report.map(|value| build_stream_plan_response(plan_kind, value, auth_context)))
}
fn build_sync_plan_response(
plan_kind: &str,
value: LocalSyncPlanAndReport,
auth_context: Option<GatewayControlAuthContext>,
) -> GatewayControlPlanResponse {
GatewayControlPlanResponse {
action: EXECUTION_RUNTIME_SYNC_ACTION.to_string(),
plan_kind: Some(plan_kind.to_string()),
plan: Some(value.plan),
report_kind: value.report_kind,
report_context: value.report_context,
auth_context,
}
}
fn build_stream_plan_response(
plan_kind: &str,
value: LocalStreamPlanAndReport,
auth_context: Option<GatewayControlAuthContext>,
) -> GatewayControlPlanResponse {
GatewayControlPlanResponse {
action: EXECUTION_RUNTIME_STREAM_ACTION.to_string(),
plan_kind: Some(plan_kind.to_string()),
plan: Some(value.plan),
report_kind: value.report_kind,
report_context: value.report_context,
auth_context,
}
}
@@ -0,0 +1,22 @@
mod control_plan;
mod stream;
mod sync;
pub(crate) use self::control_plan::{
maybe_build_stream_plan_payload_impl, maybe_build_sync_plan_payload_impl,
};
pub(crate) use self::stream::maybe_build_stream_decision_payload_impl as maybe_build_stream_decision_payload;
pub(crate) use self::sync::maybe_build_sync_decision_payload_impl as maybe_build_sync_decision_payload;
pub(crate) use super::{
maybe_build_stream_local_decision_payload,
maybe_build_stream_local_gemini_files_decision_payload,
maybe_build_stream_local_openai_cli_decision_payload,
maybe_build_stream_local_same_format_provider_decision_payload,
maybe_build_stream_local_standard_decision_payload, maybe_build_sync_local_decision_payload,
maybe_build_sync_local_gemini_files_decision_payload,
maybe_build_sync_local_openai_cli_decision_payload,
maybe_build_sync_local_same_format_provider_decision_payload,
maybe_build_sync_local_standard_decision_payload,
maybe_build_sync_local_video_decision_payload, resolve_stream_plan_kind,
resolve_sync_plan_kind,
};
@@ -0,0 +1,156 @@
use std::collections::BTreeMap;
use crate::gateway::ai_pipeline::planner::common::{
EXECUTION_RUNTIME_STREAM_DECISION_ACTION, OPENAI_VIDEO_CONTENT_PLAN_KIND,
};
use crate::gateway::scheduler::{
is_matching_stream_request, resolve_execution_runtime_stream_plan_kind,
};
use crate::gateway::{
AppState, GatewayControlDecision, GatewayControlSyncDecisionResponse, GatewayError,
};
pub(crate) async fn maybe_build_stream_decision_payload_impl(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
let Some(plan_kind) = resolve_execution_runtime_stream_plan_kind(parts, decision) else {
return Ok(None);
};
if !is_matching_stream_request(plan_kind, parts, body_json) {
return Ok(None);
}
if let Some(payload) = maybe_build_local_video_task_content_stream_decision_payload(
state, parts, trace_id, decision, plan_kind,
)
.await?
{
return Ok(Some(payload));
}
if let Some(payload) = super::maybe_build_stream_local_decision_payload(
state, parts, trace_id, decision, body_json, plan_kind,
)
.await?
{
return Ok(Some(payload));
}
if let Some(payload) = super::maybe_build_stream_local_openai_cli_decision_payload(
state, parts, trace_id, decision, body_json, plan_kind,
)
.await?
{
return Ok(Some(payload));
}
if let Some(payload) = super::maybe_build_stream_local_standard_decision_payload(
state, parts, trace_id, decision, body_json, plan_kind,
)
.await?
{
return Ok(Some(payload));
}
if let Some(payload) = super::maybe_build_stream_local_same_format_provider_decision_payload(
state, parts, trace_id, decision, body_json, plan_kind,
)
.await?
{
return Ok(Some(payload));
}
if let Some(payload) = super::maybe_build_stream_local_gemini_files_decision_payload(
state, parts, trace_id, decision, plan_kind,
)
.await?
{
return Ok(Some(payload));
}
Ok(None)
}
async fn maybe_build_local_video_task_content_stream_decision_payload(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
plan_kind: &str,
) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
if plan_kind != OPENAI_VIDEO_CONTENT_PLAN_KIND
|| decision.route_family.as_deref() != Some("openai")
{
return Ok(None);
}
let _ = state
.hydrate_video_task_for_route(decision.route_family.as_deref(), parts.uri.path())
.await?;
let Some(action) = state.video_tasks.prepare_openai_content_stream_action(
parts.uri.path(),
parts.uri.query(),
trace_id,
) else {
return Ok(None);
};
let crate::gateway::video_tasks::LocalVideoTaskContentAction::StreamPlan(plan) = action else {
return Ok(None);
};
let provider_contract = plan.provider_api_format.clone();
let client_contract = plan.client_api_format.clone();
let execution_strategy = if plan.provider_api_format == plan.client_api_format {
crate::gateway::ExecutionStrategy::LocalSameFormat
} else {
crate::gateway::ExecutionStrategy::LocalCrossFormat
};
let conversion_mode = if plan.provider_api_format == plan.client_api_format {
crate::gateway::ConversionMode::None
} else {
crate::gateway::ConversionMode::Bidirectional
};
Ok(Some(GatewayControlSyncDecisionResponse {
action: EXECUTION_RUNTIME_STREAM_DECISION_ACTION.to_string(),
decision_kind: Some(plan_kind.to_string()),
execution_strategy: Some(execution_strategy.as_str().to_string()),
conversion_mode: Some(conversion_mode.as_str().to_string()),
request_id: Some(plan.request_id),
candidate_id: plan.candidate_id,
provider_name: plan.provider_name,
provider_id: Some(plan.provider_id),
endpoint_id: Some(plan.endpoint_id),
key_id: Some(plan.key_id),
upstream_base_url: None,
upstream_url: Some(plan.url),
provider_request_method: Some(plan.method),
auth_header: None,
auth_value: None,
provider_api_format: Some(plan.provider_api_format),
client_api_format: Some(plan.client_api_format),
provider_contract: Some(provider_contract),
client_contract: Some(client_contract),
model_name: plan.model_name,
mapped_model: None,
prompt_cache_key: None,
extra_headers: BTreeMap::new(),
provider_request_headers: plan.headers,
provider_request_body: None,
provider_request_body_base64: None,
content_type: plan.content_type,
proxy: plan.proxy,
tls_profile: plan.tls_profile,
timeouts: plan.timeouts,
upstream_is_stream: true,
report_kind: None,
report_context: None,
auth_context: decision.auth_context.clone(),
}))
}
@@ -0,0 +1,222 @@
use std::collections::BTreeMap;
use url::Url;
use crate::gateway::ai_pipeline::planner::common::{
EXECUTION_RUNTIME_SYNC_DECISION_ACTION, GEMINI_FILES_DELETE_PLAN_KIND,
GEMINI_FILES_GET_PLAN_KIND, GEMINI_FILES_LIST_PLAN_KIND, GEMINI_VIDEO_CANCEL_SYNC_PLAN_KIND,
OPENAI_VIDEO_CANCEL_SYNC_PLAN_KIND, OPENAI_VIDEO_DELETE_SYNC_PLAN_KIND,
OPENAI_VIDEO_REMIX_SYNC_PLAN_KIND,
};
use crate::gateway::scheduler::resolve_execution_runtime_sync_plan_kind;
use crate::gateway::{
resolve_execution_runtime_auth_context, AppState, GatewayControlDecision,
GatewayControlSyncDecisionResponse, GatewayError,
};
pub(crate) async fn maybe_build_sync_decision_payload_impl(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
body_base64: Option<&str>,
body_is_empty: bool,
) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
let Some(plan_kind) = resolve_execution_runtime_sync_plan_kind(parts, decision) else {
return Ok(None);
};
if let Some(payload) = maybe_build_local_video_task_follow_up_sync_decision_payload(
state, parts, body_json, trace_id, decision, plan_kind,
)
.await?
{
return Ok(Some(payload));
}
if let Some(payload) = super::maybe_build_sync_local_video_decision_payload(
state, parts, body_json, trace_id, decision, plan_kind,
)
.await?
{
return Ok(Some(payload));
}
if let Some(payload) = super::maybe_build_sync_local_decision_payload(
state, parts, trace_id, decision, body_json, plan_kind,
)
.await?
{
return Ok(Some(payload));
}
if let Some(payload) = super::maybe_build_sync_local_openai_cli_decision_payload(
state, parts, trace_id, decision, body_json, plan_kind,
)
.await?
{
return Ok(Some(payload));
}
if let Some(payload) = super::maybe_build_sync_local_standard_decision_payload(
state, parts, trace_id, decision, body_json, plan_kind,
)
.await?
{
return Ok(Some(payload));
}
if let Some(payload) = super::maybe_build_sync_local_same_format_provider_decision_payload(
state, parts, trace_id, decision, body_json, plan_kind,
)
.await?
{
return Ok(Some(payload));
}
if matches!(
plan_kind,
GEMINI_FILES_LIST_PLAN_KIND | GEMINI_FILES_GET_PLAN_KIND | GEMINI_FILES_DELETE_PLAN_KIND
) {
if let Some(payload) = super::maybe_build_sync_local_gemini_files_decision_payload(
state,
parts,
body_json,
body_base64,
body_is_empty,
trace_id,
decision,
plan_kind,
)
.await?
{
return Ok(Some(payload));
}
}
Ok(None)
}
async fn maybe_build_local_video_task_follow_up_sync_decision_payload(
state: &AppState,
parts: &http::request::Parts,
body_json: &serde_json::Value,
trace_id: &str,
decision: &GatewayControlDecision,
plan_kind: &str,
) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
if !matches!(
plan_kind,
OPENAI_VIDEO_REMIX_SYNC_PLAN_KIND
| OPENAI_VIDEO_CANCEL_SYNC_PLAN_KIND
| OPENAI_VIDEO_DELETE_SYNC_PLAN_KIND
| GEMINI_VIDEO_CANCEL_SYNC_PLAN_KIND
) {
return Ok(None);
}
let _ = state
.hydrate_video_task_for_route(decision.route_family.as_deref(), parts.uri.path())
.await?;
let auth_context = resolve_execution_runtime_auth_context(
state,
decision,
&parts.headers,
&parts.uri,
trace_id,
)
.await?;
let Some(auth_context) = auth_context else {
return Ok(None);
};
let Some(follow_up) = state.video_tasks.prepare_follow_up_sync_plan(
plan_kind,
parts.uri.path(),
Some(body_json),
Some(&auth_context),
trace_id,
) else {
return Ok(None);
};
let auth_pair = extract_auth_header_pair(&follow_up.plan.headers);
let execution_strategy =
if follow_up.plan.provider_api_format == follow_up.plan.client_api_format {
crate::gateway::ExecutionStrategy::LocalSameFormat
} else {
crate::gateway::ExecutionStrategy::LocalCrossFormat
};
let conversion_mode = if follow_up.plan.provider_api_format == follow_up.plan.client_api_format
{
crate::gateway::ConversionMode::None
} else {
crate::gateway::ConversionMode::Bidirectional
};
Ok(Some(GatewayControlSyncDecisionResponse {
action: EXECUTION_RUNTIME_SYNC_DECISION_ACTION.to_string(),
decision_kind: Some(plan_kind.to_string()),
execution_strategy: Some(execution_strategy.as_str().to_string()),
conversion_mode: Some(conversion_mode.as_str().to_string()),
request_id: Some(trace_id.to_string()),
candidate_id: follow_up.plan.candidate_id.clone(),
provider_name: follow_up.plan.provider_name.clone(),
provider_id: Some(follow_up.plan.provider_id.clone()),
endpoint_id: Some(follow_up.plan.endpoint_id.clone()),
key_id: Some(follow_up.plan.key_id.clone()),
upstream_base_url: infer_upstream_base_url(&follow_up.plan.url),
upstream_url: Some(follow_up.plan.url.clone()),
provider_request_method: Some(follow_up.plan.method.clone()),
auth_header: auth_pair.as_ref().map(|(name, _)| name.clone()),
auth_value: auth_pair.as_ref().map(|(_, value)| value.clone()),
provider_api_format: Some(follow_up.plan.provider_api_format.clone()),
client_api_format: Some(follow_up.plan.client_api_format.clone()),
provider_contract: Some(follow_up.plan.provider_api_format.clone()),
client_contract: Some(follow_up.plan.client_api_format.clone()),
model_name: follow_up.plan.model_name.clone(),
mapped_model: None,
prompt_cache_key: None,
extra_headers: BTreeMap::new(),
provider_request_headers: follow_up.plan.headers.clone(),
provider_request_body: follow_up.plan.body.json_body.clone(),
provider_request_body_base64: follow_up.plan.body.body_bytes_b64.clone(),
content_type: follow_up.plan.content_type.clone(),
proxy: follow_up.plan.proxy.clone(),
tls_profile: follow_up.plan.tls_profile.clone(),
timeouts: follow_up.plan.timeouts.clone(),
upstream_is_stream: false,
report_kind: follow_up.report_kind,
report_context: follow_up.report_context,
auth_context: Some(auth_context),
}))
}
fn extract_auth_header_pair(headers: &BTreeMap<String, String>) -> Option<(String, String)> {
[
"authorization",
"x-api-key",
"api-key",
"x-goog-api-key",
"proxy-authorization",
]
.into_iter()
.find_map(|name| {
headers
.iter()
.find(|(header_name, _)| header_name.eq_ignore_ascii_case(name))
.map(|(header_name, value)| (header_name.clone(), value.clone()))
})
}
fn infer_upstream_base_url(upstream_url: &str) -> Option<String> {
let parsed = Url::parse(upstream_url).ok()?;
let host = parsed.host_str()?;
let mut base = format!("{}://{}", parsed.scheme(), host);
if let Some(port) = parsed.port() {
base.push(':');
base.push_str(port.to_string().as_str());
}
Some(base)
}
@@ -0,0 +1,132 @@
use axum::body::Body;
use axum::http::Response;
use crate::gateway::ai_pipeline::planner::plan_builders::{
LocalStreamPlanAndReport, LocalSyncPlanAndReport,
};
use crate::gateway::request_candidates::record_local_request_candidate_status;
use crate::gateway::{
execute_execution_runtime_stream, execute_execution_runtime_sync, AppState,
GatewayControlDecision, GatewayError,
};
pub(crate) trait LocalPlanAndReport {
fn plan(&self) -> &aether_contracts::ExecutionPlan;
fn report_kind(&self) -> Option<String>;
fn report_context(&self) -> Option<serde_json::Value>;
}
impl LocalPlanAndReport for LocalSyncPlanAndReport {
fn plan(&self) -> &aether_contracts::ExecutionPlan {
&self.plan
}
fn report_kind(&self) -> Option<String> {
self.report_kind.clone()
}
fn report_context(&self) -> Option<serde_json::Value> {
self.report_context.clone()
}
}
impl LocalPlanAndReport for LocalStreamPlanAndReport {
fn plan(&self) -> &aether_contracts::ExecutionPlan {
&self.plan
}
fn report_kind(&self) -> Option<String> {
self.report_kind.clone()
}
fn report_context(&self) -> Option<serde_json::Value> {
self.report_context.clone()
}
}
pub(crate) async fn execute_sync_plan_and_reports<T>(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
plan_kind: &str,
plan_and_reports: Vec<T>,
) -> Result<Option<Response<Body>>, GatewayError>
where
T: LocalPlanAndReport,
{
let mut remaining = plan_and_reports.into_iter();
while let Some(plan_and_report) = remaining.next() {
if let Some(response) = execute_execution_runtime_sync(
state,
parts.uri.path(),
plan_and_report.plan().clone(),
trace_id,
decision,
plan_kind,
plan_and_report.report_kind(),
plan_and_report.report_context(),
)
.await?
{
mark_unused_local_candidates(state, remaining.collect()).await;
return Ok(Some(response));
}
}
Ok(None)
}
pub(crate) async fn execute_stream_plan_and_reports<T>(
state: &AppState,
trace_id: &str,
decision: &GatewayControlDecision,
plan_kind: &str,
plan_and_reports: Vec<T>,
) -> Result<Option<Response<Body>>, GatewayError>
where
T: LocalPlanAndReport,
{
let mut remaining = plan_and_reports.into_iter();
while let Some(plan_and_report) = remaining.next() {
if let Some(response) = execute_execution_runtime_stream(
state,
plan_and_report.plan().clone(),
trace_id,
decision,
plan_kind,
plan_and_report.report_kind(),
plan_and_report.report_context(),
)
.await?
{
mark_unused_local_candidates(state, remaining.collect()).await;
return Ok(Some(response));
}
}
Ok(None)
}
pub(crate) async fn mark_unused_local_candidates<T>(state: &AppState, remaining: Vec<T>)
where
T: LocalPlanAndReport,
{
for plan_and_report in remaining {
record_local_request_candidate_status(
state,
plan_and_report.plan(),
plan_and_report.report_context().as_ref(),
aether_data::repository::candidates::RequestCandidateStatus::Unused,
None,
None,
None,
None,
None,
None,
)
.await;
}
}
@@ -0,0 +1,25 @@
use axum::body::{Body, Bytes};
use axum::http::Response;
use crate::gateway::intent;
use crate::gateway::{AppState, GatewayControlDecision, GatewayError};
pub(crate) async fn maybe_execute_sync_local_path(
state: &AppState,
parts: &http::request::Parts,
body_bytes: &Bytes,
trace_id: &str,
decision: &GatewayControlDecision,
) -> Result<Option<Response<Body>>, GatewayError> {
intent::maybe_execute_via_sync_intent_path(state, parts, body_bytes, trace_id, decision).await
}
pub(crate) async fn maybe_execute_stream_local_path(
state: &AppState,
parts: &http::request::Parts,
body_bytes: &Bytes,
trace_id: &str,
decision: &GatewayControlDecision,
) -> Result<Option<Response<Body>>, GatewayError> {
intent::maybe_execute_via_stream_intent_path(state, parts, body_bytes, trace_id, decision).await
}
@@ -0,0 +1,134 @@
use crate::gateway::{AppState, GatewayControlDecision, GatewayError};
pub(crate) mod candidate_affinity;
pub(crate) mod common;
pub(crate) mod contracts;
mod decision;
pub(crate) mod family_core;
pub(crate) mod local_path;
pub(crate) mod passthrough;
pub(crate) mod plan_builders;
pub(crate) mod specialized;
pub(crate) mod standard;
pub(crate) use self::candidate_affinity::prefer_local_tunnel_owner_candidates;
pub(crate) use self::common::{
parse_direct_request_body, CLAUDE_CHAT_STREAM_PLAN_KIND, CLAUDE_CHAT_SYNC_PLAN_KIND,
CLAUDE_CLI_STREAM_PLAN_KIND, CLAUDE_CLI_SYNC_PLAN_KIND, EXECUTION_RUNTIME_STREAM_ACTION,
EXECUTION_RUNTIME_STREAM_DECISION_ACTION, EXECUTION_RUNTIME_SYNC_ACTION,
EXECUTION_RUNTIME_SYNC_DECISION_ACTION, GEMINI_CHAT_STREAM_PLAN_KIND,
GEMINI_CHAT_SYNC_PLAN_KIND, GEMINI_CLI_STREAM_PLAN_KIND, GEMINI_CLI_SYNC_PLAN_KIND,
GEMINI_FILES_DELETE_PLAN_KIND, GEMINI_FILES_DOWNLOAD_PLAN_KIND, GEMINI_FILES_GET_PLAN_KIND,
GEMINI_FILES_LIST_PLAN_KIND, GEMINI_FILES_UPLOAD_PLAN_KIND, GEMINI_VIDEO_CANCEL_SYNC_PLAN_KIND,
GEMINI_VIDEO_CREATE_SYNC_PLAN_KIND, OPENAI_CHAT_STREAM_PLAN_KIND, OPENAI_CHAT_SYNC_PLAN_KIND,
OPENAI_CLI_STREAM_PLAN_KIND, OPENAI_CLI_SYNC_PLAN_KIND, OPENAI_COMPACT_STREAM_PLAN_KIND,
OPENAI_COMPACT_SYNC_PLAN_KIND, OPENAI_VIDEO_CANCEL_SYNC_PLAN_KIND,
OPENAI_VIDEO_CONTENT_PLAN_KIND, OPENAI_VIDEO_CREATE_SYNC_PLAN_KIND,
OPENAI_VIDEO_DELETE_SYNC_PLAN_KIND, OPENAI_VIDEO_REMIX_SYNC_PLAN_KIND,
};
pub(crate) use self::contracts::{
build_gateway_plan_request, generic_decision_missing_exact_provider_request,
GatewayControlPlanRequest, GatewayControlPlanResponse, GatewayControlSyncDecisionResponse,
};
pub(crate) use crate::gateway::ai_pipeline::conversion::request::{
convert_openai_chat_request_to_claude_request, convert_openai_chat_request_to_gemini_request,
convert_openai_chat_request_to_openai_cli_request, extract_openai_text_content,
normalize_openai_cli_request_to_openai_chat_request, parse_openai_tool_result_content,
};
pub(crate) use crate::gateway::scheduler::{
is_matching_stream_request,
resolve_execution_runtime_stream_plan_kind as resolve_stream_plan_kind,
resolve_execution_runtime_sync_plan_kind as resolve_sync_plan_kind,
};
pub(crate) use passthrough::{
maybe_build_stream_local_same_format_provider_decision_payload,
maybe_build_sync_local_same_format_provider_decision_payload,
maybe_execute_stream_via_local_same_format_provider_decision,
maybe_execute_sync_via_local_same_format_provider_decision,
};
pub(crate) use specialized::{
maybe_build_stream_local_gemini_files_decision_payload,
maybe_build_sync_local_gemini_files_decision_payload,
maybe_build_sync_local_video_decision_payload,
maybe_execute_stream_via_local_gemini_files_decision,
maybe_execute_sync_via_local_gemini_files_decision,
maybe_execute_sync_via_local_video_decision,
};
pub(crate) use local_path::{maybe_execute_stream_local_path, maybe_execute_sync_local_path};
pub(crate) use standard::{
copy_request_number_field, copy_request_number_field_as,
map_openai_reasoning_effort_to_claude_output, map_openai_reasoning_effort_to_gemini_budget,
maybe_build_stream_local_decision_payload,
maybe_build_stream_local_openai_cli_decision_payload,
maybe_build_stream_local_standard_decision_payload, maybe_build_sync_local_decision_payload,
maybe_build_sync_local_openai_cli_decision_payload,
maybe_build_sync_local_standard_decision_payload, maybe_execute_stream_via_local_decision,
maybe_execute_stream_via_local_openai_cli_decision,
maybe_execute_stream_via_local_standard_decision, maybe_execute_sync_via_local_decision,
maybe_execute_sync_via_local_openai_cli_decision,
maybe_execute_sync_via_local_standard_decision, parse_openai_stop_sequences,
resolve_openai_chat_max_tokens, value_as_u64,
};
pub(crate) async fn maybe_build_sync_decision_payload(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
body_base64: Option<&str>,
body_is_empty: bool,
) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
decision::maybe_build_sync_decision_payload(
state,
parts,
trace_id,
decision,
body_json,
body_base64,
body_is_empty,
)
.await
}
pub(crate) async fn maybe_build_stream_decision_payload(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
decision::maybe_build_stream_decision_payload(state, parts, trace_id, decision, body_json).await
}
pub(crate) async fn maybe_build_sync_plan_payload(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
body_base64: Option<&str>,
body_is_empty: bool,
) -> Result<Option<GatewayControlPlanResponse>, GatewayError> {
decision::maybe_build_sync_plan_payload_impl(
state,
parts,
trace_id,
decision,
body_json,
body_base64,
body_is_empty,
)
.await
}
pub(crate) async fn maybe_build_stream_plan_payload(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
) -> Result<Option<GatewayControlPlanResponse>, GatewayError> {
decision::maybe_build_stream_plan_payload_impl(state, parts, trace_id, decision, body_json)
.await
}
@@ -0,0 +1,11 @@
//! Requests that can stay in the same public/provider contract family.
mod provider;
pub(crate) use self::provider::{
maybe_build_stream_local_same_format_provider_decision_payload,
maybe_build_sync_local_same_format_provider_decision_payload,
maybe_execute_stream_via_local_same_format_provider_decision,
maybe_execute_sync_via_local_same_format_provider_decision,
};
pub(crate) use crate::gateway::provider_transport::provider_type_supports_local_same_format_transport;
@@ -0,0 +1,233 @@
use aether_contracts::{ExecutionPlan, RequestBody};
use super::{augment_sync_report_context, LocalStreamPlanAndReport, LocalSyncPlanAndReport};
use crate::gateway::{GatewayControlSyncDecisionResponse, GatewayError};
pub(crate) fn build_passthrough_sync_plan_from_decision(
parts: &http::request::Parts,
payload: GatewayControlSyncDecisionResponse,
) -> Result<Option<LocalSyncPlanAndReport>, GatewayError> {
let Some(request_id) = payload
.request_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(provider_id) = payload
.provider_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(endpoint_id) = payload
.endpoint_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(key_id) = payload
.key_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(provider_api_format) = payload
.provider_api_format
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(client_api_format) = payload
.client_api_format
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(upstream_url) = payload
.upstream_url
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let (request_body, provider_request_body_for_report) = resolve_passthrough_sync_request_body(
payload.provider_request_body.clone(),
payload.provider_request_body_base64.clone(),
);
let plan = ExecutionPlan {
request_id,
candidate_id: payload.candidate_id.clone(),
provider_name: payload.provider_name.clone(),
provider_id,
endpoint_id,
key_id,
method: payload
.provider_request_method
.clone()
.filter(|value| !value.trim().is_empty())
.unwrap_or_else(|| parts.method.to_string()),
url: upstream_url,
headers: payload.provider_request_headers.clone(),
content_type: payload.content_type.clone().or_else(|| {
payload
.provider_request_headers
.get("content-type")
.cloned()
}),
content_encoding: None,
body: request_body,
stream: false,
client_api_format,
provider_api_format,
model_name: payload.model_name.clone(),
proxy: payload.proxy.clone(),
tls_profile: payload.tls_profile.clone(),
timeouts: payload.timeouts.clone(),
};
let report_context = augment_sync_report_context(
payload.report_context,
&plan.headers,
&provider_request_body_for_report,
)?;
Ok(Some(LocalSyncPlanAndReport {
plan,
report_kind: payload.report_kind,
report_context,
}))
}
pub(crate) fn build_passthrough_stream_plan_from_decision(
parts: &http::request::Parts,
payload: GatewayControlSyncDecisionResponse,
) -> Result<Option<LocalStreamPlanAndReport>, GatewayError> {
let Some(request_id) = payload
.request_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(provider_id) = payload
.provider_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(endpoint_id) = payload
.endpoint_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(key_id) = payload
.key_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(provider_api_format) = payload
.provider_api_format
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(client_api_format) = payload
.client_api_format
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(upstream_url) = payload
.upstream_url
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let plan = ExecutionPlan {
request_id,
candidate_id: payload.candidate_id.clone(),
provider_name: payload.provider_name.clone(),
provider_id,
endpoint_id,
key_id,
method: parts.method.to_string(),
url: upstream_url,
headers: payload.provider_request_headers.clone(),
content_type: payload.content_type.clone().or_else(|| {
payload
.provider_request_headers
.get("content-type")
.cloned()
}),
content_encoding: None,
body: RequestBody {
json_body: None,
body_bytes_b64: None,
body_ref: None,
},
stream: true,
client_api_format,
provider_api_format,
model_name: payload.model_name.clone(),
proxy: payload.proxy.clone(),
tls_profile: payload.tls_profile.clone(),
timeouts: payload.timeouts.clone(),
};
Ok(Some(LocalStreamPlanAndReport {
plan,
report_kind: payload.report_kind,
report_context: payload.report_context,
}))
}
fn resolve_passthrough_sync_request_body(
provider_request_body: Option<serde_json::Value>,
provider_request_body_base64: Option<String>,
) -> (RequestBody, serde_json::Value) {
if let Some(body_bytes_b64) = provider_request_body_base64
.as_ref()
.map(|value| value.trim())
.filter(|value| !value.is_empty())
.map(ToOwned::to_owned)
{
return (
RequestBody {
json_body: None,
body_bytes_b64: Some(body_bytes_b64.clone()),
body_ref: None,
},
serde_json::json!({"body_bytes_b64": body_bytes_b64}),
);
}
match provider_request_body.unwrap_or(serde_json::Value::Null) {
serde_json::Value::Null => (
RequestBody {
json_body: None,
body_bytes_b64: None,
body_ref: None,
},
serde_json::Value::Null,
),
other => {
let report_body = other.clone();
(RequestBody::from_json(other), report_body)
}
}
}
@@ -0,0 +1,76 @@
use axum::body::Body;
use axum::http::Response;
use std::collections::BTreeMap;
use url::form_urlencoded;
use aether_data::repository::candidates::{RequestCandidateStatus, UpsertRequestCandidateRecord};
use serde_json::{json, Value};
use tracing::warn;
use uuid::Uuid;
use crate::gateway::headers::collect_control_headers;
use crate::gateway::provider_transport::{
apply_local_body_rules, apply_local_header_rules, build_antigravity_safe_v1internal_request,
build_antigravity_static_identity_headers, build_antigravity_v1internal_url,
build_claude_code_messages_url, build_claude_code_passthrough_headers,
build_claude_messages_url, build_gemini_content_url,
build_kiro_generate_assistant_response_url, build_kiro_provider_headers,
build_kiro_provider_request_body, build_openai_passthrough_headers, build_passthrough_headers,
build_passthrough_path_url, build_vertex_api_key_gemini_content_url,
classify_local_antigravity_request_support, ensure_upstream_auth_header,
resolve_local_gemini_auth, resolve_local_standard_auth,
resolve_local_vertex_api_key_query_auth, resolve_transport_execution_timeouts,
resolve_transport_proxy_snapshot_with_tunnel_affinity, resolve_transport_tls_profile,
sanitize_claude_code_request_body, supports_local_claude_code_transport_with_network,
supports_local_gemini_transport_with_network,
supports_local_kiro_request_transport_with_network,
supports_local_standard_transport_with_network,
supports_local_vertex_api_key_gemini_transport_with_network, AntigravityEnvelopeRequestType,
AntigravityRequestEnvelopeSupport, AntigravityRequestSideSupport, AntigravityRequestUrlAction,
LocalResolvedOAuthRequestAuth, KIRO_ENVELOPE_NAME,
};
use crate::gateway::request_candidates::{
current_unix_secs, record_local_request_candidate_status,
};
use crate::gateway::ai_pipeline::planner::plan_builders::{
LocalStreamPlanAndReport, LocalSyncPlanAndReport,
};
use crate::gateway::ai_pipeline::planner::prefer_local_tunnel_owner_candidates;
use crate::gateway::scheduler::{
list_selectable_candidates, GatewayMinimalCandidateSelectionCandidate,
};
use crate::gateway::{
append_execution_contract_fields_to_value, execute_execution_runtime_stream,
execute_execution_runtime_sync, AppState, ConversionMode, ExecutionStrategy,
GatewayControlDecision, GatewayControlSyncDecisionResponse, GatewayError,
};
use crate::gateway::ai_pipeline::planner::{
EXECUTION_RUNTIME_STREAM_DECISION_ACTION, EXECUTION_RUNTIME_SYNC_DECISION_ACTION,
};
mod family;
mod plans;
mod request;
pub(super) use self::family::{
materialize_local_same_format_provider_candidate_attempts,
maybe_build_local_same_format_provider_decision_payload_for_candidate,
resolve_local_same_format_provider_decision_input, LocalSameFormatProviderFamily,
LocalSameFormatProviderSpec,
};
pub(crate) use self::family::{
maybe_build_stream_local_same_format_provider_decision_payload,
maybe_build_sync_local_same_format_provider_decision_payload,
maybe_execute_stream_via_local_same_format_provider_decision,
maybe_execute_sync_via_local_same_format_provider_decision,
};
use self::plans::{
build_local_stream_plan_and_reports, build_local_sync_plan_and_reports, resolve_stream_spec,
resolve_sync_spec,
};
use self::request::{
build_same_format_provider_request_body, build_same_format_upstream_url,
extract_gemini_model_from_path,
};
const ANTIGRAVITY_ENVELOPE_NAME: &str = "antigravity:v1internal";
@@ -0,0 +1,163 @@
use aether_data::repository::candidates::{RequestCandidateStatus, UpsertRequestCandidateRecord};
use serde_json::json;
use tracing::warn;
use uuid::Uuid;
use crate::gateway::request_candidates::current_unix_secs;
use crate::gateway::scheduler::list_selectable_candidates;
use crate::gateway::{
append_execution_contract_fields_to_value, AppState, ConversionMode, ExecutionStrategy,
GatewayControlDecision, GatewayError,
};
use super::types::{
LocalSameFormatProviderCandidateAttempt, LocalSameFormatProviderDecisionInput,
LocalSameFormatProviderFamily, LocalSameFormatProviderSpec,
};
pub(crate) async fn resolve_local_same_format_provider_decision_input(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
spec: LocalSameFormatProviderSpec,
) -> Option<LocalSameFormatProviderDecisionInput> {
let Some(auth_context) = decision.auth_context.clone().filter(|auth_context| {
!auth_context.user_id.trim().is_empty() && !auth_context.api_key_id.trim().is_empty()
}) else {
return None;
};
let requested_model = match spec.family {
LocalSameFormatProviderFamily::Standard => body_json
.get("model")
.and_then(|value| value.as_str())
.map(str::trim)
.filter(|value| !value.is_empty())
.map(ToOwned::to_owned)?,
LocalSameFormatProviderFamily::Gemini => {
super::super::request::extract_gemini_model_from_path(parts.uri.path())?
}
};
let auth_snapshot = match state
.read_auth_api_key_snapshot(
&auth_context.user_id,
&auth_context.api_key_id,
current_unix_secs(),
)
.await
{
Ok(Some(snapshot)) => snapshot,
Ok(None) => return None,
Err(err) => {
warn!(
trace_id = %trace_id,
api_format = spec.api_format,
error = ?err,
"gateway local same-format decision auth snapshot read failed"
);
return None;
}
};
Some(LocalSameFormatProviderDecisionInput {
auth_context,
requested_model,
auth_snapshot,
})
}
pub(crate) async fn materialize_local_same_format_provider_candidate_attempts(
state: &AppState,
trace_id: &str,
input: &LocalSameFormatProviderDecisionInput,
spec: LocalSameFormatProviderSpec,
) -> Result<Vec<LocalSameFormatProviderCandidateAttempt>, GatewayError> {
let candidates = list_selectable_candidates(
state,
spec.api_format,
&input.requested_model,
spec.require_streaming,
Some(&input.auth_snapshot),
current_unix_secs(),
)
.await?;
let candidates =
crate::gateway::ai_pipeline::planner::prefer_local_tunnel_owner_candidates(state, candidates)
.await;
let created_at_unix_secs = current_unix_secs();
let mut attempts = Vec::with_capacity(candidates.len());
for (candidate_index, candidate) in candidates.into_iter().enumerate() {
let generated_candidate_id = Uuid::new_v4().to_string();
let extra_data = append_execution_contract_fields_to_value(
json!({
"provider_api_format": spec.api_format,
"client_api_format": spec.api_format,
"global_model_id": candidate.global_model_id.clone(),
"global_model_name": candidate.global_model_name.clone(),
"model_id": candidate.model_id.clone(),
"selected_provider_model_name": candidate.selected_provider_model_name.clone(),
"mapping_matched_model": candidate.mapping_matched_model.clone(),
"provider_name": candidate.provider_name.clone(),
"key_name": candidate.key_name.clone(),
}),
ExecutionStrategy::LocalSameFormat,
ConversionMode::None,
spec.api_format,
spec.api_format,
);
let candidate_id = match state
.upsert_request_candidate(UpsertRequestCandidateRecord {
id: generated_candidate_id.clone(),
request_id: trace_id.to_string(),
user_id: Some(input.auth_context.user_id.clone()),
api_key_id: Some(input.auth_context.api_key_id.clone()),
username: None,
api_key_name: None,
candidate_index: candidate_index as u32,
retry_index: 0,
provider_id: Some(candidate.provider_id.clone()),
endpoint_id: Some(candidate.endpoint_id.clone()),
key_id: Some(candidate.key_id.clone()),
status: RequestCandidateStatus::Available,
skip_reason: None,
is_cached: Some(false),
status_code: None,
error_type: None,
error_message: None,
latency_ms: None,
concurrent_requests: None,
extra_data: Some(extra_data),
required_capabilities: candidate.key_capabilities.clone(),
created_at_unix_secs: Some(created_at_unix_secs),
started_at_unix_secs: None,
finished_at_unix_secs: None,
})
.await
{
Ok(Some(stored)) => stored.id,
Ok(None) => generated_candidate_id.clone(),
Err(err) => {
warn!(
trace_id = %trace_id,
api_format = spec.api_format,
error = ?err,
"gateway local same-format decision request candidate upsert failed"
);
generated_candidate_id.clone()
}
};
attempts.push(LocalSameFormatProviderCandidateAttempt {
candidate,
candidate_index: candidate_index as u32,
candidate_id,
});
}
Ok(attempts)
}
@@ -0,0 +1,141 @@
use axum::body::Body;
use axum::http::Response;
use crate::gateway::ai_pipeline::planner::family_core::{
execute_stream_plan_and_reports, execute_sync_plan_and_reports,
};
use crate::gateway::ai_pipeline::planner::plan_builders::{
LocalStreamPlanAndReport, LocalSyncPlanAndReport,
};
use crate::gateway::{AppState, GatewayControlDecision, GatewayControlSyncDecisionResponse, GatewayError};
use super::super::plans::{
build_local_stream_plan_and_reports, build_local_sync_plan_and_reports, resolve_stream_spec,
resolve_sync_spec,
};
use super::candidates::{
materialize_local_same_format_provider_candidate_attempts,
resolve_local_same_format_provider_decision_input,
};
use super::payload::maybe_build_local_same_format_provider_decision_payload_for_candidate;
pub(crate) async fn maybe_execute_sync_via_local_same_format_provider_decision(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Option<Response<Body>>, GatewayError> {
let Some(spec) = resolve_sync_spec(plan_kind) else {
return Ok(None);
};
let plan_and_reports =
build_local_sync_plan_and_reports(state, parts, trace_id, decision, body_json, spec)
.await?;
if plan_and_reports.is_empty() {
return Ok(None);
}
execute_sync_plan_and_reports(state, parts, trace_id, decision, plan_kind, plan_and_reports)
.await
}
pub(crate) async fn maybe_execute_stream_via_local_same_format_provider_decision(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Option<Response<Body>>, GatewayError> {
let Some(spec) = resolve_stream_spec(plan_kind) else {
return Ok(None);
};
let plan_and_reports =
build_local_stream_plan_and_reports(state, parts, trace_id, decision, body_json, spec)
.await?;
if plan_and_reports.is_empty() {
return Ok(None);
}
execute_stream_plan_and_reports(state, trace_id, decision, plan_kind, plan_and_reports).await
}
pub(crate) async fn maybe_build_sync_local_same_format_provider_decision_payload(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
let Some(spec) = resolve_sync_spec(plan_kind) else {
return Ok(None);
};
let Some(input) = resolve_local_same_format_provider_decision_input(
state, parts, trace_id, decision, body_json, spec,
)
.await
else {
return Ok(None);
};
let attempts =
materialize_local_same_format_provider_candidate_attempts(state, trace_id, &input, spec)
.await?;
for attempt in attempts {
if let Some(payload) =
maybe_build_local_same_format_provider_decision_payload_for_candidate(
state, parts, trace_id, body_json, &input, attempt, spec,
)
.await
{
return Ok(Some(payload));
}
}
Ok(None)
}
pub(crate) async fn maybe_build_stream_local_same_format_provider_decision_payload(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
let Some(spec) = resolve_stream_spec(plan_kind) else {
return Ok(None);
};
let Some(input) = resolve_local_same_format_provider_decision_input(
state, parts, trace_id, decision, body_json, spec,
)
.await
else {
return Ok(None);
};
let attempts =
materialize_local_same_format_provider_candidate_attempts(state, trace_id, &input, spec)
.await?;
for attempt in attempts {
if let Some(payload) =
maybe_build_local_same_format_provider_decision_payload_for_candidate(
state, parts, trace_id, body_json, &input, attempt, spec,
)
.await
{
return Ok(Some(payload));
}
}
Ok(None)
}
@@ -0,0 +1,17 @@
mod candidates;
mod execute;
mod payload;
mod types;
pub(crate) use self::candidates::{
materialize_local_same_format_provider_candidate_attempts,
resolve_local_same_format_provider_decision_input,
};
pub(crate) use self::execute::{
maybe_build_stream_local_same_format_provider_decision_payload,
maybe_build_sync_local_same_format_provider_decision_payload,
maybe_execute_stream_via_local_same_format_provider_decision,
maybe_execute_sync_via_local_same_format_provider_decision,
};
pub(crate) use self::payload::maybe_build_local_same_format_provider_decision_payload_for_candidate;
pub(crate) use self::types::{LocalSameFormatProviderFamily, LocalSameFormatProviderSpec};
@@ -0,0 +1,562 @@
use std::collections::BTreeMap;
use aether_data::repository::candidates::{RequestCandidateStatus, UpsertRequestCandidateRecord};
use serde_json::json;
use tracing::warn;
use crate::gateway::headers::collect_control_headers;
use crate::gateway::provider_transport::{
apply_local_header_rules, build_antigravity_safe_v1internal_request,
build_antigravity_static_identity_headers, build_claude_code_passthrough_headers,
build_openai_passthrough_headers, build_passthrough_headers,
classify_local_antigravity_request_support, ensure_upstream_auth_header,
resolve_local_gemini_auth, resolve_local_standard_auth,
resolve_local_vertex_api_key_query_auth, resolve_transport_execution_timeouts,
resolve_transport_proxy_snapshot_with_tunnel_affinity, resolve_transport_tls_profile,
supports_local_claude_code_transport_with_network,
supports_local_gemini_transport_with_network,
supports_local_kiro_request_transport_with_network,
supports_local_standard_transport_with_network,
supports_local_vertex_api_key_gemini_transport_with_network, AntigravityEnvelopeRequestType,
AntigravityRequestEnvelopeSupport, AntigravityRequestSideSupport,
LocalResolvedOAuthRequestAuth, KIRO_ENVELOPE_NAME,
};
use crate::gateway::request_candidates::current_unix_secs;
use crate::gateway::scheduler::GatewayMinimalCandidateSelectionCandidate;
use crate::gateway::{
append_execution_contract_fields_to_value, AppState, ConversionMode, ExecutionStrategy,
GatewayControlSyncDecisionResponse, EXECUTION_RUNTIME_STREAM_DECISION_ACTION,
EXECUTION_RUNTIME_SYNC_DECISION_ACTION,
};
use super::types::{
LocalSameFormatProviderCandidateAttempt, LocalSameFormatProviderDecisionInput,
LocalSameFormatProviderFamily, LocalSameFormatProviderSpec,
};
pub(crate) async fn maybe_build_local_same_format_provider_decision_payload_for_candidate(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
body_json: &serde_json::Value,
input: &LocalSameFormatProviderDecisionInput,
attempt: LocalSameFormatProviderCandidateAttempt,
spec: LocalSameFormatProviderSpec,
) -> Option<GatewayControlSyncDecisionResponse> {
let LocalSameFormatProviderCandidateAttempt {
candidate,
candidate_index,
candidate_id,
} = attempt;
let transport = match state
.read_provider_transport_snapshot(
&candidate.provider_id,
&candidate.endpoint_id,
&candidate.key_id,
)
.await
{
Ok(Some(snapshot)) => snapshot,
Ok(None) => {
mark_skipped_local_same_format_provider_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_snapshot_missing",
)
.await;
return None;
}
Err(err) => {
warn!(
trace_id = %trace_id,
api_format = spec.api_format,
error = ?err,
"gateway local same-format decision provider transport read failed"
);
mark_skipped_local_same_format_provider_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_snapshot_read_failed",
)
.await;
return None;
}
};
let is_antigravity = transport
.provider
.provider_type
.trim()
.eq_ignore_ascii_case("antigravity");
let is_claude_code = transport
.provider
.provider_type
.trim()
.eq_ignore_ascii_case("claude_code");
let is_vertex = transport
.provider
.provider_type
.trim()
.eq_ignore_ascii_case("vertex_ai");
let transport_supported = match spec.family {
_ if transport
.provider
.provider_type
.trim()
.eq_ignore_ascii_case("kiro") =>
{
supports_local_kiro_request_transport_with_network(&transport)
}
_ if is_antigravity => true,
_ if is_claude_code => {
supports_local_claude_code_transport_with_network(&transport, spec.api_format)
}
_ if is_vertex => supports_local_vertex_api_key_gemini_transport_with_network(&transport),
LocalSameFormatProviderFamily::Standard => {
supports_local_standard_transport_with_network(&transport, spec.api_format)
}
LocalSameFormatProviderFamily::Gemini => {
supports_local_gemini_transport_with_network(&transport, spec.api_format)
}
};
if !transport_supported {
mark_skipped_local_same_format_provider_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_unsupported",
)
.await;
return None;
}
let is_kiro = transport
.provider
.provider_type
.trim()
.eq_ignore_ascii_case("kiro");
let vertex_query_auth = if is_vertex {
resolve_local_vertex_api_key_query_auth(&transport)
} else {
None
};
let should_try_oauth_auth = is_kiro
|| matches!(spec.family, LocalSameFormatProviderFamily::Standard)
&& resolve_local_standard_auth(&transport).is_none()
|| matches!(spec.family, LocalSameFormatProviderFamily::Gemini)
&& !is_vertex
&& resolve_local_gemini_auth(&transport).is_none();
let oauth_auth = if should_try_oauth_auth {
match state.resolve_local_oauth_request_auth(&transport).await {
Ok(Some(LocalResolvedOAuthRequestAuth::Kiro(auth))) => {
Some(LocalResolvedOAuthRequestAuth::Kiro(auth))
}
Ok(Some(LocalResolvedOAuthRequestAuth::Header { name, value })) => {
Some(LocalResolvedOAuthRequestAuth::Header { name, value })
}
Ok(None) => None,
Err(err) => {
warn!(
trace_id = %trace_id,
api_format = spec.api_format,
provider_type = %transport.provider.provider_type,
error = ?err,
"gateway local same-format oauth auth resolution failed"
);
None
}
}
} else {
None
};
let kiro_auth = match oauth_auth.as_ref() {
Some(LocalResolvedOAuthRequestAuth::Kiro(auth)) => Some(auth),
_ => None,
};
let auth = if let Some(auth) = kiro_auth.as_ref() {
Some((auth.name.to_string(), auth.value.clone()))
} else if let Some(LocalResolvedOAuthRequestAuth::Header { name, value }) = oauth_auth.as_ref()
{
Some((name.clone(), value.clone()))
} else if is_vertex {
None
} else {
match spec.family {
LocalSameFormatProviderFamily::Standard => resolve_local_standard_auth(&transport),
LocalSameFormatProviderFamily::Gemini => resolve_local_gemini_auth(&transport),
}
};
let (auth_header, auth_value) = match auth {
Some((name, value)) => (Some(name), Some(value)),
None if is_vertex && vertex_query_auth.is_some() => (None, None),
None => {
mark_skipped_local_same_format_provider_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_auth_unavailable",
)
.await;
return None;
}
};
if is_vertex && vertex_query_auth.is_none() {
mark_skipped_local_same_format_provider_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_auth_unavailable",
)
.await;
return None;
}
let mapped_model = candidate.selected_provider_model_name.trim().to_string();
if mapped_model.is_empty() {
mark_skipped_local_same_format_provider_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"mapped_model_missing",
)
.await;
return None;
}
let Some(base_provider_request_body) =
super::super::request::build_same_format_provider_request_body(
body_json,
&mapped_model,
spec,
transport.endpoint.body_rules.as_ref(),
is_kiro || is_antigravity || spec.require_streaming,
kiro_auth,
is_claude_code,
)
else {
mark_skipped_local_same_format_provider_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"provider_request_body_missing",
)
.await;
return None;
};
let antigravity_auth = if is_antigravity {
match classify_local_antigravity_request_support(
&transport,
&base_provider_request_body,
AntigravityEnvelopeRequestType::Agent,
) {
AntigravityRequestSideSupport::Supported(spec) => Some(spec.auth),
AntigravityRequestSideSupport::Unsupported(_) => {
mark_skipped_local_same_format_provider_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_unsupported",
)
.await;
return None;
}
}
} else {
None
};
let provider_request_body = if let Some(antigravity_auth) = antigravity_auth.as_ref() {
match build_antigravity_safe_v1internal_request(
antigravity_auth,
trace_id,
&mapped_model,
&base_provider_request_body,
AntigravityEnvelopeRequestType::Agent,
) {
AntigravityRequestEnvelopeSupport::Supported(envelope) => envelope,
AntigravityRequestEnvelopeSupport::Unsupported(_) => {
mark_skipped_local_same_format_provider_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"provider_request_body_missing",
)
.await;
return None;
}
}
} else {
base_provider_request_body
};
let upstream_is_stream = is_kiro || is_antigravity || spec.require_streaming;
let report_kind = if is_kiro && !spec.require_streaming {
"claude_cli_sync_finalize"
} else if is_antigravity && !spec.require_streaming {
match spec.api_format {
"gemini:chat" => "gemini_chat_sync_finalize",
"gemini:cli" => "gemini_cli_sync_finalize",
_ => spec.report_kind,
}
} else {
spec.report_kind
};
let Some(upstream_url) = super::super::request::build_same_format_upstream_url(
parts,
&transport,
&mapped_model,
spec,
upstream_is_stream,
kiro_auth,
) else {
mark_skipped_local_same_format_provider_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"upstream_url_missing",
)
.await;
return None;
};
let Some(provider_request_headers) = (if let Some(kiro_auth) = kiro_auth.as_ref() {
crate::gateway::provider_transport::build_kiro_provider_headers(
&parts.headers,
&provider_request_body,
body_json,
transport.endpoint.header_rules.as_ref(),
auth_header.as_deref().unwrap_or_default(),
auth_value.as_deref().unwrap_or_default(),
&kiro_auth.auth_config,
kiro_auth.machine_id.as_str(),
)
} else {
let extra_headers = antigravity_auth
.as_ref()
.map(build_antigravity_static_identity_headers)
.unwrap_or_default();
let mut provider_request_headers = if is_claude_code {
build_claude_code_passthrough_headers(
&parts.headers,
auth_header.as_deref().unwrap_or_default(),
auth_value.as_deref().unwrap_or_default(),
&extra_headers,
upstream_is_stream,
transport.key.fingerprint.as_ref(),
)
} else if is_vertex {
build_passthrough_headers(&parts.headers, &extra_headers, Some("application/json"))
} else {
build_openai_passthrough_headers(
&parts.headers,
auth_header.as_deref().unwrap_or_default(),
auth_value.as_deref().unwrap_or_default(),
&extra_headers,
Some("application/json"),
)
};
let protected_headers = auth_header
.as_deref()
.filter(|value| !value.trim().is_empty())
.map(|value| vec![value, "content-type"])
.unwrap_or_else(|| vec!["content-type"]);
if !apply_local_header_rules(
&mut provider_request_headers,
transport.endpoint.header_rules.as_ref(),
&protected_headers,
&provider_request_body,
Some(body_json),
) {
None
} else {
if let (Some(auth_header), Some(auth_value)) =
(auth_header.as_deref(), auth_value.as_deref())
{
ensure_upstream_auth_header(&mut provider_request_headers, auth_header, auth_value);
}
if upstream_is_stream {
provider_request_headers
.insert("accept".to_string(), "text/event-stream".to_string());
}
Some(provider_request_headers)
}
}) else {
mark_skipped_local_same_format_provider_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_header_rules_apply_failed",
)
.await;
return None;
};
let prompt_cache_key = provider_request_body
.get("prompt_cache_key")
.and_then(|value| value.as_str())
.map(str::trim)
.filter(|value| !value.is_empty())
.map(ToOwned::to_owned);
let proxy = resolve_transport_proxy_snapshot_with_tunnel_affinity(state, &transport).await;
let tls_profile = resolve_transport_tls_profile(&transport);
let report_context = append_execution_contract_fields_to_value(
json!({
"user_id": input.auth_context.user_id,
"api_key_id": input.auth_context.api_key_id,
"request_id": trace_id,
"candidate_id": candidate_id,
"candidate_index": candidate_index,
"retry_index": 0,
"model": input.requested_model,
"provider_name": transport.provider.name,
"provider_id": candidate.provider_id,
"endpoint_id": candidate.endpoint_id,
"key_id": candidate.key_id,
"provider_api_format": spec.api_format,
"client_api_format": spec.api_format,
"mapped_model": mapped_model,
"upstream_url": upstream_url,
"provider_request_method": serde_json::Value::Null,
"provider_request_headers": provider_request_headers,
"provider_request_body": provider_request_body,
"original_headers": collect_control_headers(&parts.headers),
"original_request_body": body_json,
"has_envelope": is_kiro || is_antigravity,
"envelope_name": if is_kiro {
Some(KIRO_ENVELOPE_NAME)
} else if is_antigravity {
Some(super::super::ANTIGRAVITY_ENVELOPE_NAME)
} else {
None
},
"needs_conversion": false,
}),
ExecutionStrategy::LocalSameFormat,
ConversionMode::None,
spec.api_format,
spec.api_format,
);
Some(GatewayControlSyncDecisionResponse {
action: if spec.require_streaming {
EXECUTION_RUNTIME_STREAM_DECISION_ACTION.to_string()
} else {
EXECUTION_RUNTIME_SYNC_DECISION_ACTION.to_string()
},
decision_kind: Some(spec.decision_kind.to_string()),
execution_strategy: Some(ExecutionStrategy::LocalSameFormat.as_str().to_string()),
conversion_mode: Some(ConversionMode::None.as_str().to_string()),
request_id: Some(trace_id.to_string()),
candidate_id: Some(candidate_id.clone()),
provider_name: Some(transport.provider.name.clone()),
provider_id: Some(candidate.provider_id.clone()),
endpoint_id: Some(candidate.endpoint_id.clone()),
key_id: Some(candidate.key_id.clone()),
upstream_base_url: Some(transport.endpoint.base_url.clone()),
upstream_url: Some(upstream_url.clone()),
provider_request_method: None,
auth_header,
auth_value,
provider_api_format: Some(spec.api_format.to_string()),
client_api_format: Some(spec.api_format.to_string()),
provider_contract: Some(spec.api_format.to_string()),
client_contract: Some(spec.api_format.to_string()),
model_name: Some(input.requested_model.clone()),
mapped_model: Some(mapped_model.clone()),
prompt_cache_key,
extra_headers: BTreeMap::new(),
provider_request_headers: provider_request_headers.clone(),
provider_request_body: Some(provider_request_body.clone()),
provider_request_body_base64: None,
content_type: Some("application/json".to_string()),
proxy,
tls_profile,
timeouts: resolve_transport_execution_timeouts(&transport),
upstream_is_stream,
report_kind: Some(report_kind.to_string()),
report_context: Some(report_context),
auth_context: Some(input.auth_context.clone()),
})
}
pub(super) async fn mark_skipped_local_same_format_provider_candidate(
state: &AppState,
input: &LocalSameFormatProviderDecisionInput,
trace_id: &str,
candidate: &GatewayMinimalCandidateSelectionCandidate,
candidate_index: u32,
candidate_id: &str,
skip_reason: &'static str,
) {
if let Err(err) = state
.upsert_request_candidate(UpsertRequestCandidateRecord {
id: candidate_id.to_string(),
request_id: trace_id.to_string(),
user_id: Some(input.auth_context.user_id.clone()),
api_key_id: Some(input.auth_context.api_key_id.clone()),
username: None,
api_key_name: None,
candidate_index,
retry_index: 0,
provider_id: Some(candidate.provider_id.clone()),
endpoint_id: Some(candidate.endpoint_id.clone()),
key_id: Some(candidate.key_id.clone()),
status: RequestCandidateStatus::Skipped,
skip_reason: Some(skip_reason.to_string()),
is_cached: Some(false),
status_code: None,
error_type: None,
error_message: None,
latency_ms: None,
concurrent_requests: None,
extra_data: None,
required_capabilities: candidate.key_capabilities.clone(),
created_at_unix_secs: None,
started_at_unix_secs: None,
finished_at_unix_secs: Some(current_unix_secs()),
})
.await
{
warn!(
trace_id = %trace_id,
candidate_id = %candidate_id,
skip_reason,
error = ?err,
"gateway local same-format decision failed to persist skipped candidate"
);
}
}
@@ -0,0 +1,28 @@
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub(crate) enum LocalSameFormatProviderFamily {
Standard,
Gemini,
}
#[derive(Debug, Clone, Copy)]
pub(crate) struct LocalSameFormatProviderSpec {
pub(crate) api_format: &'static str,
pub(crate) decision_kind: &'static str,
pub(crate) report_kind: &'static str,
pub(crate) family: LocalSameFormatProviderFamily,
pub(crate) require_streaming: bool,
}
#[derive(Debug, Clone)]
pub(crate) struct LocalSameFormatProviderDecisionInput {
pub(crate) auth_context: crate::gateway::GatewayControlAuthContext,
pub(crate) requested_model: String,
pub(crate) auth_snapshot: crate::gateway::gateway_data::StoredGatewayAuthApiKeySnapshot,
}
#[derive(Debug, Clone)]
pub(crate) struct LocalSameFormatProviderCandidateAttempt {
pub(crate) candidate: crate::gateway::scheduler::GatewayMinimalCandidateSelectionCandidate,
pub(crate) candidate_index: u32,
pub(crate) candidate_id: String,
}
@@ -0,0 +1,198 @@
use tracing::warn;
use super::{
materialize_local_same_format_provider_candidate_attempts,
maybe_build_local_same_format_provider_decision_payload_for_candidate,
resolve_local_same_format_provider_decision_input, AppState, GatewayControlDecision,
GatewayError, LocalSameFormatProviderFamily, LocalSameFormatProviderSpec,
LocalStreamPlanAndReport, LocalSyncPlanAndReport,
};
use crate::gateway::ai_pipeline::planner::plan_builders::{
build_gemini_stream_plan_from_decision, build_gemini_sync_plan_from_decision,
build_standard_stream_plan_from_decision, build_standard_sync_plan_from_decision,
};
use crate::gateway::ai_pipeline::planner::{
CLAUDE_CHAT_STREAM_PLAN_KIND, CLAUDE_CHAT_SYNC_PLAN_KIND, CLAUDE_CLI_STREAM_PLAN_KIND,
CLAUDE_CLI_SYNC_PLAN_KIND, GEMINI_CHAT_STREAM_PLAN_KIND, GEMINI_CHAT_SYNC_PLAN_KIND,
GEMINI_CLI_STREAM_PLAN_KIND, GEMINI_CLI_SYNC_PLAN_KIND,
};
pub(super) fn resolve_sync_spec(plan_kind: &str) -> Option<LocalSameFormatProviderSpec> {
match plan_kind {
CLAUDE_CHAT_SYNC_PLAN_KIND => Some(LocalSameFormatProviderSpec {
api_format: "claude:chat",
decision_kind: CLAUDE_CHAT_SYNC_PLAN_KIND,
report_kind: "claude_chat_sync_success",
family: LocalSameFormatProviderFamily::Standard,
require_streaming: false,
}),
CLAUDE_CLI_SYNC_PLAN_KIND => Some(LocalSameFormatProviderSpec {
api_format: "claude:cli",
decision_kind: CLAUDE_CLI_SYNC_PLAN_KIND,
report_kind: "claude_cli_sync_success",
family: LocalSameFormatProviderFamily::Standard,
require_streaming: false,
}),
GEMINI_CHAT_SYNC_PLAN_KIND => Some(LocalSameFormatProviderSpec {
api_format: "gemini:chat",
decision_kind: GEMINI_CHAT_SYNC_PLAN_KIND,
report_kind: "gemini_chat_sync_success",
family: LocalSameFormatProviderFamily::Gemini,
require_streaming: false,
}),
GEMINI_CLI_SYNC_PLAN_KIND => Some(LocalSameFormatProviderSpec {
api_format: "gemini:cli",
decision_kind: GEMINI_CLI_SYNC_PLAN_KIND,
report_kind: "gemini_cli_sync_success",
family: LocalSameFormatProviderFamily::Gemini,
require_streaming: false,
}),
_ => None,
}
}
pub(super) fn resolve_stream_spec(plan_kind: &str) -> Option<LocalSameFormatProviderSpec> {
match plan_kind {
CLAUDE_CHAT_STREAM_PLAN_KIND => Some(LocalSameFormatProviderSpec {
api_format: "claude:chat",
decision_kind: CLAUDE_CHAT_STREAM_PLAN_KIND,
report_kind: "claude_chat_stream_success",
family: LocalSameFormatProviderFamily::Standard,
require_streaming: true,
}),
CLAUDE_CLI_STREAM_PLAN_KIND => Some(LocalSameFormatProviderSpec {
api_format: "claude:cli",
decision_kind: CLAUDE_CLI_STREAM_PLAN_KIND,
report_kind: "claude_cli_stream_success",
family: LocalSameFormatProviderFamily::Standard,
require_streaming: true,
}),
GEMINI_CHAT_STREAM_PLAN_KIND => Some(LocalSameFormatProviderSpec {
api_format: "gemini:chat",
decision_kind: GEMINI_CHAT_STREAM_PLAN_KIND,
report_kind: "gemini_chat_stream_success",
family: LocalSameFormatProviderFamily::Gemini,
require_streaming: true,
}),
GEMINI_CLI_STREAM_PLAN_KIND => Some(LocalSameFormatProviderSpec {
api_format: "gemini:cli",
decision_kind: GEMINI_CLI_STREAM_PLAN_KIND,
report_kind: "gemini_cli_stream_success",
family: LocalSameFormatProviderFamily::Gemini,
require_streaming: true,
}),
_ => None,
}
}
pub(super) async fn build_local_sync_plan_and_reports(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
spec: LocalSameFormatProviderSpec,
) -> Result<Vec<LocalSyncPlanAndReport>, GatewayError> {
let Some(input) = resolve_local_same_format_provider_decision_input(
state, parts, trace_id, decision, body_json, spec,
)
.await
else {
return Ok(Vec::new());
};
let attempts =
materialize_local_same_format_provider_candidate_attempts(state, trace_id, &input, spec)
.await?;
let mut plans = Vec::new();
for attempt in attempts {
let Some(payload) = maybe_build_local_same_format_provider_decision_payload_for_candidate(
state, parts, trace_id, body_json, &input, attempt, spec,
)
.await
else {
continue;
};
let built = match spec.family {
LocalSameFormatProviderFamily::Standard => {
build_standard_sync_plan_from_decision(parts, body_json, payload)
}
LocalSameFormatProviderFamily::Gemini => {
build_gemini_sync_plan_from_decision(parts, body_json, payload)
}
};
match built {
Ok(Some(value)) => plans.push(value),
Ok(None) => {}
Err(err) => {
warn!(
trace_id = %trace_id,
api_format = spec.api_format,
error = ?err,
"gateway local same-format sync decision plan build failed"
);
}
}
}
Ok(plans)
}
pub(super) async fn build_local_stream_plan_and_reports(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
spec: LocalSameFormatProviderSpec,
) -> Result<Vec<LocalStreamPlanAndReport>, GatewayError> {
let Some(input) = resolve_local_same_format_provider_decision_input(
state, parts, trace_id, decision, body_json, spec,
)
.await
else {
return Ok(Vec::new());
};
let attempts =
materialize_local_same_format_provider_candidate_attempts(state, trace_id, &input, spec)
.await?;
let mut plans = Vec::new();
for attempt in attempts {
let Some(payload) = maybe_build_local_same_format_provider_decision_payload_for_candidate(
state, parts, trace_id, body_json, &input, attempt, spec,
)
.await
else {
continue;
};
let built = match spec.family {
LocalSameFormatProviderFamily::Standard => {
build_standard_stream_plan_from_decision(parts, body_json, payload, false)
}
LocalSameFormatProviderFamily::Gemini => {
build_gemini_stream_plan_from_decision(parts, body_json, payload)
}
};
match built {
Ok(Some(value)) => plans.push(value),
Ok(None) => {}
Err(err) => {
warn!(
trace_id = %trace_id,
api_format = spec.api_format,
error = ?err,
"gateway local same-format stream decision plan build failed"
);
}
}
}
Ok(plans)
}
@@ -0,0 +1,186 @@
use super::*;
pub(super) fn build_same_format_provider_request_body(
body_json: &Value,
mapped_model: &str,
spec: LocalSameFormatProviderSpec,
body_rules: Option<&Value>,
upstream_is_stream: bool,
kiro_auth: Option<&crate::gateway::provider_transport::KiroRequestAuth>,
is_claude_code: bool,
) -> Option<Value> {
if let Some(kiro_auth) = kiro_auth {
return build_kiro_provider_request_body(
body_json,
mapped_model,
&kiro_auth.auth_config,
body_rules,
);
}
let request_body_object = body_json.as_object()?;
let mut provider_request_body = serde_json::Map::from_iter(
request_body_object
.iter()
.map(|(key, value)| (key.clone(), value.clone())),
);
match spec.family {
LocalSameFormatProviderFamily::Standard => {
provider_request_body
.insert("model".to_string(), Value::String(mapped_model.to_string()));
if upstream_is_stream {
provider_request_body.insert("stream".to_string(), Value::Bool(true));
}
}
LocalSameFormatProviderFamily::Gemini => {
provider_request_body.remove("model");
}
}
let mut provider_request_body = Value::Object(provider_request_body);
if is_claude_code {
sanitize_claude_code_request_body(&mut provider_request_body);
}
if !apply_local_body_rules(&mut provider_request_body, body_rules, Some(body_json)) {
return None;
}
Some(provider_request_body)
}
pub(super) fn build_same_format_upstream_url(
parts: &http::request::Parts,
transport: &crate::gateway::provider_transport::GatewayProviderTransportSnapshot,
mapped_model: &str,
spec: LocalSameFormatProviderSpec,
upstream_is_stream: bool,
kiro_auth: Option<&crate::gateway::provider_transport::KiroRequestAuth>,
) -> Option<String> {
if let Some(kiro_auth) = kiro_auth {
return build_kiro_generate_assistant_response_url(
&transport.endpoint.base_url,
parts.uri.query(),
Some(kiro_auth.auth_config.effective_api_region()),
);
}
if transport
.provider
.provider_type
.trim()
.eq_ignore_ascii_case("claude_code")
{
return Some(build_claude_code_messages_url(
&transport.endpoint.base_url,
parts.uri.query(),
));
}
if transport
.provider
.provider_type
.trim()
.eq_ignore_ascii_case("vertex_ai")
{
let auth = resolve_local_vertex_api_key_query_auth(transport)?;
return build_vertex_api_key_gemini_content_url(
mapped_model,
upstream_is_stream,
&auth.value,
parts.uri.query(),
);
}
if transport
.provider
.provider_type
.trim()
.eq_ignore_ascii_case("antigravity")
{
let query = parts.uri.query().map(|query| {
form_urlencoded::parse(query.as_bytes())
.into_owned()
.collect::<BTreeMap<String, String>>()
});
return build_antigravity_v1internal_url(
&transport.endpoint.base_url,
if upstream_is_stream {
AntigravityRequestUrlAction::StreamGenerateContent
} else {
AntigravityRequestUrlAction::GenerateContent
},
query.as_ref(),
);
}
let custom_path = transport
.endpoint
.custom_path
.as_deref()
.map(str::trim)
.filter(|value| !value.is_empty());
if let Some(path) = custom_path {
let blocked_keys = match spec.family {
LocalSameFormatProviderFamily::Standard => &[][..],
LocalSameFormatProviderFamily::Gemini => &["key"][..],
};
let url = build_passthrough_path_url(
&transport.endpoint.base_url,
path,
parts.uri.query(),
blocked_keys,
)?;
return Some(maybe_add_gemini_stream_alt_sse(url, spec));
}
let url = match spec.family {
LocalSameFormatProviderFamily::Standard => Some(build_claude_messages_url(
&transport.endpoint.base_url,
parts.uri.query(),
)),
LocalSameFormatProviderFamily::Gemini => build_gemini_content_url(
&transport.endpoint.base_url,
mapped_model,
spec.require_streaming,
parts.uri.query(),
),
}?;
Some(maybe_add_gemini_stream_alt_sse(url, spec))
}
pub(super) fn extract_gemini_model_from_path(path: &str) -> Option<String> {
let (_, suffix) = path.split_once("/models/")?;
let model = suffix
.split_once(':')
.map(|(value, _)| value)
.unwrap_or(suffix);
let model = model.trim();
if model.is_empty() {
None
} else {
Some(model.to_string())
}
}
fn maybe_add_gemini_stream_alt_sse(
upstream_url: String,
spec: LocalSameFormatProviderSpec,
) -> String {
if spec.family != LocalSameFormatProviderFamily::Gemini || !spec.require_streaming {
return upstream_url;
}
let has_alt = upstream_url
.split_once('?')
.map(|(_, query)| {
form_urlencoded::parse(query.as_bytes())
.any(|(key, _)| key.as_ref().eq_ignore_ascii_case("alt"))
})
.unwrap_or(false);
if has_alt {
return upstream_url;
}
if upstream_url.contains('?') {
format!("{upstream_url}&alt=sse")
} else {
format!("{upstream_url}?alt=sse")
}
}
@@ -0,0 +1,65 @@
use std::collections::BTreeMap;
use aether_contracts::ExecutionPlan;
pub(crate) use crate::gateway::ai_pipeline::planner::generic_decision_missing_exact_provider_request;
use crate::gateway::{GatewayControlSyncDecisionResponse, GatewayError};
pub(crate) struct LocalSyncPlanAndReport {
pub(crate) plan: ExecutionPlan,
pub(crate) report_kind: Option<String>,
pub(crate) report_context: Option<serde_json::Value>,
}
pub(crate) struct LocalStreamPlanAndReport {
pub(crate) plan: ExecutionPlan,
pub(crate) report_kind: Option<String>,
pub(crate) report_context: Option<serde_json::Value>,
}
#[path = "standard/gemini/plan_builders.rs"]
mod gemini_builders;
#[path = "standard/openai/plan_builders.rs"]
mod openai_builders;
#[path = "passthrough/plan_builders.rs"]
mod passthrough_builders;
#[path = "standard/plan_builders.rs"]
mod standard_builders;
pub(crate) use gemini_builders::{
build_gemini_stream_plan_from_decision, build_gemini_sync_plan_from_decision,
};
pub(crate) use openai_builders::{
build_openai_chat_stream_plan_from_decision, build_openai_chat_sync_plan_from_decision,
build_openai_cli_stream_plan_from_decision, build_openai_cli_sync_plan_from_decision,
};
pub(crate) use passthrough_builders::{
build_passthrough_stream_plan_from_decision, build_passthrough_sync_plan_from_decision,
};
pub(crate) use standard_builders::{
build_standard_stream_plan_from_decision, build_standard_sync_plan_from_decision,
};
pub(super) fn augment_sync_report_context(
report_context: Option<serde_json::Value>,
provider_request_headers: &BTreeMap<String, String>,
provider_request_body: &serde_json::Value,
) -> Result<Option<serde_json::Value>, GatewayError> {
let mut report_context = match report_context {
Some(serde_json::Value::Object(map)) => map,
Some(_) => serde_json::Map::new(),
None => serde_json::Map::new(),
};
report_context.insert(
"provider_request_headers".to_string(),
serde_json::to_value(provider_request_headers)
.map_err(|err| GatewayError::Internal(err.to_string()))?,
);
report_context.insert(
"provider_request_body".to_string(),
provider_request_body.clone(),
);
Ok(Some(serde_json::Value::Object(report_context)))
}
@@ -0,0 +1,877 @@
use std::collections::BTreeMap;
use aether_data::repository::candidates::{RequestCandidateStatus, UpsertRequestCandidateRecord};
use axum::body::Body;
use axum::http::Response;
use serde_json::json;
use tracing::warn;
use uuid::Uuid;
use crate::gateway::headers::collect_control_headers;
use crate::gateway::provider_transport::{
apply_local_body_rules, apply_local_header_rules, build_gemini_files_passthrough_url,
build_passthrough_headers_with_auth, resolve_local_gemini_auth,
resolve_transport_execution_timeouts, resolve_transport_proxy_snapshot_with_tunnel_affinity,
resolve_transport_tls_profile, supports_local_gemini_transport_with_network,
};
use crate::gateway::request_candidates::{
current_unix_secs, record_local_request_candidate_status,
};
use crate::gateway::ai_pipeline::planner::plan_builders::{
build_passthrough_stream_plan_from_decision, build_passthrough_sync_plan_from_decision,
LocalStreamPlanAndReport, LocalSyncPlanAndReport,
};
use crate::gateway::ai_pipeline::planner::prefer_local_tunnel_owner_candidates;
use crate::gateway::scheduler::{
list_selectable_candidates_for_required_capability_without_requested_model,
GatewayMinimalCandidateSelectionCandidate,
};
use crate::gateway::{
execute_execution_runtime_stream, execute_execution_runtime_sync, AppState,
GatewayControlDecision, GatewayControlSyncDecisionResponse, GatewayError,
};
use crate::gateway::ai_pipeline::planner::{
EXECUTION_RUNTIME_STREAM_DECISION_ACTION, EXECUTION_RUNTIME_SYNC_DECISION_ACTION,
GEMINI_FILES_DELETE_PLAN_KIND, GEMINI_FILES_DOWNLOAD_PLAN_KIND, GEMINI_FILES_GET_PLAN_KIND,
GEMINI_FILES_LIST_PLAN_KIND, GEMINI_FILES_UPLOAD_PLAN_KIND,
};
const GEMINI_FILES_CANDIDATE_API_FORMAT: &str = "gemini:chat";
const GEMINI_FILES_CLIENT_API_FORMAT: &str = "gemini:files";
const GEMINI_FILES_REQUIRED_CAPABILITY: &str = "gemini_files";
#[derive(Debug, Clone, Copy)]
struct LocalGeminiFilesSpec {
decision_kind: &'static str,
report_kind: Option<&'static str>,
require_streaming: bool,
}
#[derive(Debug, Clone)]
struct LocalGeminiFilesDecisionInput {
auth_context: crate::gateway::GatewayControlAuthContext,
auth_snapshot: crate::gateway::gateway_data::StoredGatewayAuthApiKeySnapshot,
}
#[derive(Debug, Clone)]
struct LocalGeminiFilesCandidateAttempt {
candidate: GatewayMinimalCandidateSelectionCandidate,
candidate_index: u32,
candidate_id: String,
}
pub(crate) async fn maybe_execute_sync_via_local_gemini_files_decision(
state: &AppState,
parts: &http::request::Parts,
body_json: &serde_json::Value,
body_base64: Option<&str>,
body_is_empty: bool,
trace_id: &str,
decision: &GatewayControlDecision,
plan_kind: &str,
) -> Result<Option<Response<Body>>, GatewayError> {
let Some(spec) = resolve_sync_spec(plan_kind) else {
return Ok(None);
};
let plan_and_reports = build_local_sync_plan_and_reports(
state,
parts,
body_json,
body_base64,
body_is_empty,
trace_id,
decision,
spec,
)
.await?;
if plan_and_reports.is_empty() {
return Ok(None);
}
let mut remaining = plan_and_reports.into_iter();
while let Some(plan_and_report) = remaining.next() {
if let Some(response) = execute_execution_runtime_sync(
state,
parts.uri.path(),
plan_and_report.plan,
trace_id,
decision,
plan_kind,
plan_and_report.report_kind,
plan_and_report.report_context,
)
.await?
{
mark_unused_local_files_candidates(state, remaining.collect()).await;
return Ok(Some(response));
}
}
Ok(None)
}
pub(crate) async fn maybe_execute_stream_via_local_gemini_files_decision(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
plan_kind: &str,
) -> Result<Option<Response<Body>>, GatewayError> {
let Some(spec) = resolve_stream_spec(plan_kind) else {
return Ok(None);
};
let plan_and_reports =
build_local_stream_plan_and_reports(state, parts, trace_id, decision, spec).await?;
if plan_and_reports.is_empty() {
return Ok(None);
}
let mut remaining = plan_and_reports.into_iter();
while let Some(plan_and_report) = remaining.next() {
if let Some(response) = execute_execution_runtime_stream(
state,
plan_and_report.plan,
trace_id,
decision,
plan_kind,
plan_and_report.report_kind,
plan_and_report.report_context,
)
.await?
{
mark_unused_local_files_candidates(state, remaining.collect()).await;
return Ok(Some(response));
}
}
Ok(None)
}
pub(crate) async fn maybe_build_sync_local_gemini_files_decision_payload(
state: &AppState,
parts: &http::request::Parts,
body_json: &serde_json::Value,
body_base64: Option<&str>,
body_is_empty: bool,
trace_id: &str,
decision: &GatewayControlDecision,
plan_kind: &str,
) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
let Some(spec) = resolve_sync_spec(plan_kind) else {
return Ok(None);
};
let Some(input) = resolve_local_gemini_files_decision_input(state, trace_id, decision).await
else {
return Ok(None);
};
let attempts =
materialize_local_gemini_files_candidate_attempts(state, trace_id, &input).await?;
for attempt in attempts {
if let Some(payload) = maybe_build_local_gemini_files_decision_payload_for_candidate(
state,
parts,
body_json,
body_base64,
body_is_empty,
trace_id,
&input,
attempt,
spec,
)
.await
{
return Ok(Some(payload));
}
}
Ok(None)
}
pub(crate) async fn maybe_build_stream_local_gemini_files_decision_payload(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
plan_kind: &str,
) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
let Some(spec) = resolve_stream_spec(plan_kind) else {
return Ok(None);
};
let Some(input) = resolve_local_gemini_files_decision_input(state, trace_id, decision).await
else {
return Ok(None);
};
let attempts =
materialize_local_gemini_files_candidate_attempts(state, trace_id, &input).await?;
let empty_body_json = serde_json::Value::Null;
for attempt in attempts {
if let Some(payload) = maybe_build_local_gemini_files_decision_payload_for_candidate(
state,
parts,
&empty_body_json,
None,
true,
trace_id,
&input,
attempt,
spec,
)
.await
{
return Ok(Some(payload));
}
}
Ok(None)
}
fn resolve_sync_spec(plan_kind: &str) -> Option<LocalGeminiFilesSpec> {
match plan_kind {
GEMINI_FILES_UPLOAD_PLAN_KIND => Some(LocalGeminiFilesSpec {
decision_kind: GEMINI_FILES_UPLOAD_PLAN_KIND,
report_kind: Some("gemini_files_store_mapping"),
require_streaming: false,
}),
GEMINI_FILES_LIST_PLAN_KIND => Some(LocalGeminiFilesSpec {
decision_kind: GEMINI_FILES_LIST_PLAN_KIND,
report_kind: Some("gemini_files_store_mapping"),
require_streaming: false,
}),
GEMINI_FILES_GET_PLAN_KIND => Some(LocalGeminiFilesSpec {
decision_kind: GEMINI_FILES_GET_PLAN_KIND,
report_kind: Some("gemini_files_store_mapping"),
require_streaming: false,
}),
GEMINI_FILES_DELETE_PLAN_KIND => Some(LocalGeminiFilesSpec {
decision_kind: GEMINI_FILES_DELETE_PLAN_KIND,
report_kind: Some("gemini_files_delete_mapping"),
require_streaming: false,
}),
_ => None,
}
}
fn resolve_stream_spec(plan_kind: &str) -> Option<LocalGeminiFilesSpec> {
match plan_kind {
GEMINI_FILES_DOWNLOAD_PLAN_KIND => Some(LocalGeminiFilesSpec {
decision_kind: GEMINI_FILES_DOWNLOAD_PLAN_KIND,
report_kind: None,
require_streaming: true,
}),
_ => None,
}
}
async fn build_local_sync_plan_and_reports(
state: &AppState,
parts: &http::request::Parts,
body_json: &serde_json::Value,
body_base64: Option<&str>,
body_is_empty: bool,
trace_id: &str,
decision: &GatewayControlDecision,
spec: LocalGeminiFilesSpec,
) -> Result<Vec<LocalSyncPlanAndReport>, GatewayError> {
let Some(input) = resolve_local_gemini_files_decision_input(state, trace_id, decision).await
else {
return Ok(Vec::new());
};
let attempts =
materialize_local_gemini_files_candidate_attempts(state, trace_id, &input).await?;
let mut plans = Vec::new();
for attempt in attempts {
let Some(payload) = maybe_build_local_gemini_files_decision_payload_for_candidate(
state,
parts,
body_json,
body_base64,
body_is_empty,
trace_id,
&input,
attempt,
spec,
)
.await
else {
continue;
};
match build_passthrough_sync_plan_from_decision(parts, payload) {
Ok(Some(value)) => plans.push(value),
Ok(None) => {}
Err(err) => {
warn!(
trace_id = %trace_id,
decision_kind = spec.decision_kind,
error = ?err,
"gateway local gemini files sync decision plan build failed"
);
}
}
}
Ok(plans)
}
async fn build_local_stream_plan_and_reports(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
spec: LocalGeminiFilesSpec,
) -> Result<Vec<LocalStreamPlanAndReport>, GatewayError> {
let Some(input) = resolve_local_gemini_files_decision_input(state, trace_id, decision).await
else {
return Ok(Vec::new());
};
let attempts =
materialize_local_gemini_files_candidate_attempts(state, trace_id, &input).await?;
let mut plans = Vec::new();
let empty_body_json = serde_json::Value::Null;
for attempt in attempts {
let Some(payload) = maybe_build_local_gemini_files_decision_payload_for_candidate(
state,
parts,
&empty_body_json,
None,
true,
trace_id,
&input,
attempt,
spec,
)
.await
else {
continue;
};
match build_passthrough_stream_plan_from_decision(parts, payload) {
Ok(Some(value)) => plans.push(value),
Ok(None) => {}
Err(err) => {
warn!(
trace_id = %trace_id,
decision_kind = spec.decision_kind,
error = ?err,
"gateway local gemini files stream decision plan build failed"
);
}
}
}
Ok(plans)
}
async fn resolve_local_gemini_files_decision_input(
state: &AppState,
trace_id: &str,
decision: &GatewayControlDecision,
) -> Option<LocalGeminiFilesDecisionInput> {
let Some(auth_context) = decision.auth_context.clone().filter(|auth_context| {
!auth_context.user_id.trim().is_empty() && !auth_context.api_key_id.trim().is_empty()
}) else {
return None;
};
let auth_snapshot = match state
.read_auth_api_key_snapshot(
&auth_context.user_id,
&auth_context.api_key_id,
current_unix_secs(),
)
.await
{
Ok(Some(snapshot)) => snapshot,
Ok(None) => return None,
Err(err) => {
warn!(
trace_id = %trace_id,
error = ?err,
"gateway local gemini files decision auth snapshot read failed"
);
return None;
}
};
Some(LocalGeminiFilesDecisionInput {
auth_context,
auth_snapshot,
})
}
async fn materialize_local_gemini_files_candidate_attempts(
state: &AppState,
trace_id: &str,
input: &LocalGeminiFilesDecisionInput,
) -> Result<Vec<LocalGeminiFilesCandidateAttempt>, GatewayError> {
let candidates = list_selectable_candidates_for_required_capability_without_requested_model(
state,
GEMINI_FILES_CANDIDATE_API_FORMAT,
GEMINI_FILES_REQUIRED_CAPABILITY,
false,
Some(&input.auth_snapshot),
current_unix_secs(),
)
.await?;
let candidates = prefer_local_tunnel_owner_candidates(state, candidates).await;
let created_at_unix_secs = current_unix_secs();
let mut attempts = Vec::with_capacity(candidates.len());
for (candidate_index, candidate) in candidates.into_iter().enumerate() {
let generated_candidate_id = Uuid::new_v4().to_string();
let extra_data = json!({
"provider_api_format": GEMINI_FILES_CLIENT_API_FORMAT,
"client_api_format": GEMINI_FILES_CLIENT_API_FORMAT,
"candidate_api_format": GEMINI_FILES_CANDIDATE_API_FORMAT,
"global_model_id": candidate.global_model_id.clone(),
"global_model_name": candidate.global_model_name.clone(),
"model_id": candidate.model_id.clone(),
"selected_provider_model_name": candidate.selected_provider_model_name.clone(),
"mapping_matched_model": candidate.mapping_matched_model.clone(),
"provider_name": candidate.provider_name.clone(),
"key_name": candidate.key_name.clone(),
});
let candidate_id = match state
.upsert_request_candidate(UpsertRequestCandidateRecord {
id: generated_candidate_id.clone(),
request_id: trace_id.to_string(),
user_id: Some(input.auth_context.user_id.clone()),
api_key_id: Some(input.auth_context.api_key_id.clone()),
username: None,
api_key_name: None,
candidate_index: candidate_index as u32,
retry_index: 0,
provider_id: Some(candidate.provider_id.clone()),
endpoint_id: Some(candidate.endpoint_id.clone()),
key_id: Some(candidate.key_id.clone()),
status: RequestCandidateStatus::Available,
skip_reason: None,
is_cached: Some(false),
status_code: None,
error_type: None,
error_message: None,
latency_ms: None,
concurrent_requests: None,
extra_data: Some(extra_data),
required_capabilities: candidate.key_capabilities.clone(),
created_at_unix_secs: Some(created_at_unix_secs),
started_at_unix_secs: None,
finished_at_unix_secs: None,
})
.await
{
Ok(Some(stored)) => stored.id,
Ok(None) => generated_candidate_id.clone(),
Err(err) => {
warn!(
trace_id = %trace_id,
error = ?err,
"gateway local gemini files request candidate upsert failed"
);
generated_candidate_id.clone()
}
};
attempts.push(LocalGeminiFilesCandidateAttempt {
candidate,
candidate_index: candidate_index as u32,
candidate_id,
});
}
Ok(attempts)
}
async fn maybe_build_local_gemini_files_decision_payload_for_candidate(
state: &AppState,
parts: &http::request::Parts,
body_json: &serde_json::Value,
body_base64: Option<&str>,
body_is_empty: bool,
trace_id: &str,
input: &LocalGeminiFilesDecisionInput,
attempt: LocalGeminiFilesCandidateAttempt,
spec: LocalGeminiFilesSpec,
) -> Option<GatewayControlSyncDecisionResponse> {
let LocalGeminiFilesCandidateAttempt {
candidate,
candidate_index,
candidate_id,
} = attempt;
let transport = match state
.read_provider_transport_snapshot(
&candidate.provider_id,
&candidate.endpoint_id,
&candidate.key_id,
)
.await
{
Ok(Some(snapshot)) => snapshot,
Ok(None) => {
mark_skipped_local_gemini_files_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_snapshot_missing",
)
.await;
return None;
}
Err(err) => {
warn!(
trace_id = %trace_id,
error = ?err,
"gateway local gemini files provider transport read failed"
);
mark_skipped_local_gemini_files_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_snapshot_read_failed",
)
.await;
return None;
}
};
if !supports_local_gemini_transport_with_network(&transport, GEMINI_FILES_CANDIDATE_API_FORMAT)
{
mark_skipped_local_gemini_files_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_unsupported",
)
.await;
return None;
}
let Some((auth_header, auth_value)) = resolve_local_gemini_auth(&transport) else {
mark_skipped_local_gemini_files_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_auth_unavailable",
)
.await;
return None;
};
let custom_path = transport
.endpoint
.custom_path
.as_deref()
.map(str::trim)
.filter(|value| !value.is_empty());
let passthrough_path = custom_path.unwrap_or(parts.uri.path());
let upstream_url =
if spec.decision_kind == GEMINI_FILES_UPLOAD_PLAN_KIND || custom_path.is_some() {
build_gemini_files_passthrough_url(
&transport.endpoint.base_url,
passthrough_path,
parts.uri.query(),
)
} else {
build_gemini_files_passthrough_url(
&transport.endpoint.base_url,
passthrough_path,
parts.uri.query(),
)
};
let Some(upstream_url) = upstream_url else {
mark_skipped_local_gemini_files_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"upstream_url_missing",
)
.await;
return None;
};
let mut provider_request_body = if spec.decision_kind == GEMINI_FILES_UPLOAD_PLAN_KIND
&& !body_is_empty
&& body_base64.is_none()
{
Some(body_json.clone())
} else {
None
};
let provider_request_body_base64 = if spec.decision_kind == GEMINI_FILES_UPLOAD_PLAN_KIND {
body_base64
.map(str::trim)
.filter(|value| !value.is_empty())
.map(ToOwned::to_owned)
} else {
None
};
let original_request_body = if let Some(body_bytes_b64) = provider_request_body_base64.clone() {
json!({"body_bytes_b64": body_bytes_b64})
} else if !body_is_empty {
body_json.clone()
} else {
serde_json::Value::Null
};
if provider_request_body_base64.is_some() && transport.endpoint.body_rules.is_some() {
mark_skipped_local_gemini_files_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_body_rules_unsupported_for_binary_upload",
)
.await;
return None;
}
if let Some(body) = provider_request_body.as_mut() {
if !apply_local_body_rules(
body,
transport.endpoint.body_rules.as_ref(),
Some(body_json),
) {
mark_skipped_local_gemini_files_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_body_rules_apply_failed",
)
.await;
return None;
}
}
let mut provider_request_headers = build_passthrough_headers_with_auth(
&parts.headers,
&auth_header,
&auth_value,
&BTreeMap::new(),
);
if !apply_local_header_rules(
&mut provider_request_headers,
transport.endpoint.header_rules.as_ref(),
&[&auth_header, "content-type"],
provider_request_body
.as_ref()
.unwrap_or(&original_request_body),
Some(&original_request_body),
) {
mark_skipped_local_gemini_files_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_header_rules_apply_failed",
)
.await;
return None;
}
let file_name = parts
.uri
.path()
.trim_start_matches("/v1beta/")
.trim()
.to_string();
let proxy = resolve_transport_proxy_snapshot_with_tunnel_affinity(state, &transport).await;
let tls_profile = resolve_transport_tls_profile(&transport);
Some(GatewayControlSyncDecisionResponse {
action: if spec.require_streaming {
EXECUTION_RUNTIME_STREAM_DECISION_ACTION.to_string()
} else {
EXECUTION_RUNTIME_SYNC_DECISION_ACTION.to_string()
},
decision_kind: Some(spec.decision_kind.to_string()),
execution_strategy: Some(
crate::gateway::ExecutionStrategy::LocalSameFormat
.as_str()
.to_string(),
),
conversion_mode: Some(crate::gateway::ConversionMode::None.as_str().to_string()),
request_id: Some(trace_id.to_string()),
candidate_id: Some(candidate_id.clone()),
provider_name: Some(transport.provider.name.clone()),
provider_id: Some(candidate.provider_id.clone()),
endpoint_id: Some(candidate.endpoint_id.clone()),
key_id: Some(candidate.key_id.clone()),
upstream_base_url: Some(transport.endpoint.base_url.clone()),
upstream_url: Some(upstream_url),
provider_request_method: Some(parts.method.to_string()),
auth_header: Some(auth_header),
auth_value: Some(auth_value),
provider_api_format: Some(GEMINI_FILES_CLIENT_API_FORMAT.to_string()),
client_api_format: Some(GEMINI_FILES_CLIENT_API_FORMAT.to_string()),
provider_contract: Some(GEMINI_FILES_CLIENT_API_FORMAT.to_string()),
client_contract: Some(GEMINI_FILES_CLIENT_API_FORMAT.to_string()),
model_name: Some("gemini-files".to_string()),
mapped_model: Some(candidate.selected_provider_model_name.clone()),
prompt_cache_key: None,
extra_headers: BTreeMap::new(),
provider_request_headers,
provider_request_body,
provider_request_body_base64,
content_type: parts
.headers
.get(http::header::CONTENT_TYPE)
.and_then(|value| value.to_str().ok())
.map(str::trim)
.filter(|value| !value.is_empty())
.map(ToOwned::to_owned),
proxy,
tls_profile,
timeouts: resolve_transport_execution_timeouts(&transport),
upstream_is_stream: spec.require_streaming,
report_kind: spec.report_kind.map(ToOwned::to_owned),
report_context: Some(json!({
"user_id": input.auth_context.user_id,
"api_key_id": input.auth_context.api_key_id,
"request_id": trace_id,
"candidate_id": candidate_id,
"candidate_index": candidate_index,
"retry_index": 0,
"model": "gemini-files",
"provider_name": transport.provider.name,
"provider_id": candidate.provider_id,
"endpoint_id": candidate.endpoint_id,
"key_id": candidate.key_id,
"file_key_id": candidate.key_id,
"file_name": file_name,
"provider_api_format": GEMINI_FILES_CLIENT_API_FORMAT,
"client_api_format": GEMINI_FILES_CLIENT_API_FORMAT,
"original_headers": collect_control_headers(&parts.headers),
"original_request_body": original_request_body,
"has_envelope": false,
"needs_conversion": false,
})),
auth_context: Some(input.auth_context.clone()),
})
}
async fn mark_skipped_local_gemini_files_candidate(
state: &AppState,
input: &LocalGeminiFilesDecisionInput,
trace_id: &str,
candidate: &GatewayMinimalCandidateSelectionCandidate,
candidate_index: u32,
candidate_id: &str,
skip_reason: &'static str,
) {
if let Err(err) = state
.upsert_request_candidate(UpsertRequestCandidateRecord {
id: candidate_id.to_string(),
request_id: trace_id.to_string(),
user_id: Some(input.auth_context.user_id.clone()),
api_key_id: Some(input.auth_context.api_key_id.clone()),
username: None,
api_key_name: None,
candidate_index,
retry_index: 0,
provider_id: Some(candidate.provider_id.clone()),
endpoint_id: Some(candidate.endpoint_id.clone()),
key_id: Some(candidate.key_id.clone()),
status: RequestCandidateStatus::Skipped,
skip_reason: Some(skip_reason.to_string()),
is_cached: Some(false),
status_code: None,
error_type: None,
error_message: None,
latency_ms: None,
concurrent_requests: None,
extra_data: None,
required_capabilities: candidate.key_capabilities.clone(),
created_at_unix_secs: None,
started_at_unix_secs: None,
finished_at_unix_secs: Some(current_unix_secs()),
})
.await
{
warn!(
trace_id = %trace_id,
candidate_id = %candidate_id,
skip_reason,
error = ?err,
"gateway local gemini files failed to persist skipped candidate"
);
}
}
async fn mark_unused_local_files_candidates<T>(state: &AppState, remaining: Vec<T>)
where
T: LocalGeminiFilesPlanAndReport,
{
for plan_and_report in remaining {
record_local_request_candidate_status(
state,
plan_and_report.plan(),
plan_and_report.report_context(),
RequestCandidateStatus::Unused,
None,
None,
None,
None,
None,
None,
)
.await;
}
}
trait LocalGeminiFilesPlanAndReport {
fn plan(&self) -> &aether_contracts::ExecutionPlan;
fn report_context(&self) -> Option<&serde_json::Value>;
}
impl LocalGeminiFilesPlanAndReport for LocalSyncPlanAndReport {
fn plan(&self) -> &aether_contracts::ExecutionPlan {
&self.plan
}
fn report_context(&self) -> Option<&serde_json::Value> {
self.report_context.as_ref()
}
}
impl LocalGeminiFilesPlanAndReport for LocalStreamPlanAndReport {
fn plan(&self) -> &aether_contracts::ExecutionPlan {
&self.plan
}
fn report_context(&self) -> Option<&serde_json::Value> {
self.report_context.as_ref()
}
}
@@ -0,0 +1,14 @@
//! Non-matrix AI surfaces such as files and video.
mod files;
mod video;
pub(crate) use self::files::{
maybe_build_stream_local_gemini_files_decision_payload,
maybe_build_sync_local_gemini_files_decision_payload,
maybe_execute_stream_via_local_gemini_files_decision,
maybe_execute_sync_via_local_gemini_files_decision,
};
pub(crate) use self::video::{
maybe_build_sync_local_video_decision_payload, maybe_execute_sync_via_local_video_decision,
};
@@ -0,0 +1,770 @@
use std::collections::BTreeMap;
use aether_data::repository::candidates::{RequestCandidateStatus, UpsertRequestCandidateRecord};
use axum::body::Body;
use axum::http::Response;
use serde_json::{json, Value};
use tracing::warn;
use uuid::Uuid;
use crate::gateway::headers::collect_control_headers;
use crate::gateway::provider_transport::{
apply_local_body_rules, apply_local_header_rules, build_gemini_video_predict_long_running_url,
build_passthrough_headers_with_auth, build_passthrough_path_url, resolve_local_gemini_auth,
resolve_local_openai_chat_auth, resolve_transport_execution_timeouts,
resolve_transport_proxy_snapshot_with_tunnel_affinity, resolve_transport_tls_profile,
supports_local_gemini_transport_with_network, supports_local_standard_transport_with_network,
};
use crate::gateway::request_candidates::{
current_unix_secs, record_local_request_candidate_status,
};
use crate::gateway::ai_pipeline::planner::plan_builders::{
build_passthrough_sync_plan_from_decision, LocalSyncPlanAndReport,
};
use crate::gateway::ai_pipeline::planner::prefer_local_tunnel_owner_candidates;
use crate::gateway::scheduler::{
list_selectable_candidates, GatewayMinimalCandidateSelectionCandidate,
};
use crate::gateway::{
execute_execution_runtime_sync, AppState, GatewayControlDecision,
GatewayControlSyncDecisionResponse, GatewayError,
};
use crate::gateway::ai_pipeline::planner::{
EXECUTION_RUNTIME_SYNC_DECISION_ACTION, GEMINI_VIDEO_CREATE_SYNC_PLAN_KIND,
OPENAI_VIDEO_CREATE_SYNC_PLAN_KIND,
};
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
enum LocalVideoCreateFamily {
OpenAi,
Gemini,
}
#[derive(Debug, Clone, Copy)]
struct LocalVideoCreateSpec {
api_format: &'static str,
decision_kind: &'static str,
report_kind: &'static str,
family: LocalVideoCreateFamily,
}
#[derive(Debug, Clone)]
struct LocalVideoCreateDecisionInput {
auth_context: crate::gateway::GatewayControlAuthContext,
requested_model: String,
auth_snapshot: crate::gateway::gateway_data::StoredGatewayAuthApiKeySnapshot,
}
#[derive(Debug, Clone)]
struct LocalVideoCreateCandidateAttempt {
candidate: GatewayMinimalCandidateSelectionCandidate,
candidate_index: u32,
candidate_id: String,
}
pub(crate) async fn maybe_execute_sync_via_local_video_decision(
state: &AppState,
parts: &http::request::Parts,
body_json: &serde_json::Value,
trace_id: &str,
decision: &GatewayControlDecision,
plan_kind: &str,
) -> Result<Option<Response<Body>>, GatewayError> {
let Some(spec) = resolve_sync_spec(plan_kind) else {
return Ok(None);
};
let plan_and_reports =
build_local_sync_plan_and_reports(state, parts, body_json, trace_id, decision, spec)
.await?;
if plan_and_reports.is_empty() {
return Ok(None);
}
let mut remaining = plan_and_reports.into_iter();
while let Some(plan_and_report) = remaining.next() {
if let Some(response) = execute_execution_runtime_sync(
state,
parts.uri.path(),
plan_and_report.plan,
trace_id,
decision,
plan_kind,
plan_and_report.report_kind,
plan_and_report.report_context,
)
.await?
{
mark_unused_local_video_candidates(state, remaining.collect()).await;
return Ok(Some(response));
}
}
Ok(None)
}
pub(crate) async fn maybe_build_sync_local_video_decision_payload(
state: &AppState,
parts: &http::request::Parts,
body_json: &serde_json::Value,
trace_id: &str,
decision: &GatewayControlDecision,
plan_kind: &str,
) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
let Some(spec) = resolve_sync_spec(plan_kind) else {
return Ok(None);
};
let Some(input) = resolve_local_video_create_decision_input(
state, parts, trace_id, decision, body_json, spec,
)
.await
else {
return Ok(None);
};
let candidates = match list_selectable_candidates(
state,
spec.api_format,
&input.requested_model,
false,
Some(&input.auth_snapshot),
current_unix_secs(),
)
.await
{
Ok(candidates) => candidates,
Err(err) => {
warn!(
trace_id = %trace_id,
decision_kind = spec.decision_kind,
error = ?err,
"gateway local video decision scheduler selection failed"
);
return Ok(None);
}
};
let attempts = materialize_local_video_create_candidate_attempts(
state,
trace_id,
&input,
candidates,
spec.api_format,
)
.await;
for attempt in attempts {
if let Some(payload) = maybe_build_local_video_create_decision_payload_for_candidate(
state, parts, body_json, trace_id, &input, attempt, spec,
)
.await
{
return Ok(Some(payload));
}
}
Ok(None)
}
fn resolve_sync_spec(plan_kind: &str) -> Option<LocalVideoCreateSpec> {
match plan_kind {
OPENAI_VIDEO_CREATE_SYNC_PLAN_KIND => Some(LocalVideoCreateSpec {
api_format: "openai:video",
decision_kind: OPENAI_VIDEO_CREATE_SYNC_PLAN_KIND,
report_kind: "openai_video_create_sync_finalize",
family: LocalVideoCreateFamily::OpenAi,
}),
GEMINI_VIDEO_CREATE_SYNC_PLAN_KIND => Some(LocalVideoCreateSpec {
api_format: "gemini:video",
decision_kind: GEMINI_VIDEO_CREATE_SYNC_PLAN_KIND,
report_kind: "gemini_video_create_sync_finalize",
family: LocalVideoCreateFamily::Gemini,
}),
_ => None,
}
}
async fn build_local_sync_plan_and_reports(
state: &AppState,
parts: &http::request::Parts,
body_json: &serde_json::Value,
trace_id: &str,
decision: &GatewayControlDecision,
spec: LocalVideoCreateSpec,
) -> Result<Vec<LocalSyncPlanAndReport>, GatewayError> {
let Some(input) = resolve_local_video_create_decision_input(
state, parts, trace_id, decision, body_json, spec,
)
.await
else {
return Ok(Vec::new());
};
let candidates = match list_selectable_candidates(
state,
spec.api_format,
&input.requested_model,
false,
Some(&input.auth_snapshot),
current_unix_secs(),
)
.await
{
Ok(candidates) => candidates,
Err(err) => {
warn!(
trace_id = %trace_id,
decision_kind = spec.decision_kind,
error = ?err,
"gateway local video decision scheduler selection failed"
);
return Ok(Vec::new());
}
};
let attempts = materialize_local_video_create_candidate_attempts(
state,
trace_id,
&input,
candidates,
spec.api_format,
)
.await;
let mut plans = Vec::new();
for attempt in attempts {
let Some(payload) = maybe_build_local_video_create_decision_payload_for_candidate(
state, parts, body_json, trace_id, &input, attempt, spec,
)
.await
else {
continue;
};
match build_passthrough_sync_plan_from_decision(parts, payload) {
Ok(Some(value)) => plans.push(value),
Ok(None) => {}
Err(err) => {
warn!(
trace_id = %trace_id,
decision_kind = spec.decision_kind,
error = ?err,
"gateway local video sync decision plan build failed"
);
}
}
}
Ok(plans)
}
async fn resolve_local_video_create_decision_input(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
spec: LocalVideoCreateSpec,
) -> Option<LocalVideoCreateDecisionInput> {
let Some(auth_context) = decision.auth_context.clone().filter(|auth_context| {
!auth_context.user_id.trim().is_empty() && !auth_context.api_key_id.trim().is_empty()
}) else {
return None;
};
let requested_model = match spec.family {
LocalVideoCreateFamily::OpenAi => body_json
.get("model")
.and_then(|value| value.as_str())
.map(str::trim)
.filter(|value| !value.is_empty())
.map(ToOwned::to_owned)?,
LocalVideoCreateFamily::Gemini => extract_gemini_video_model_from_path(parts.uri.path())?,
};
let auth_snapshot = match state
.read_auth_api_key_snapshot(
&auth_context.user_id,
&auth_context.api_key_id,
current_unix_secs(),
)
.await
{
Ok(Some(snapshot)) => snapshot,
Ok(None) => return None,
Err(err) => {
warn!(
trace_id = %trace_id,
decision_kind = spec.decision_kind,
error = ?err,
"gateway local video decision auth snapshot read failed"
);
return None;
}
};
Some(LocalVideoCreateDecisionInput {
auth_context,
requested_model,
auth_snapshot,
})
}
async fn maybe_build_local_video_create_decision_payload_for_candidate(
state: &AppState,
parts: &http::request::Parts,
body_json: &serde_json::Value,
trace_id: &str,
input: &LocalVideoCreateDecisionInput,
attempt: LocalVideoCreateCandidateAttempt,
spec: LocalVideoCreateSpec,
) -> Option<GatewayControlSyncDecisionResponse> {
let LocalVideoCreateCandidateAttempt {
candidate,
candidate_index,
candidate_id,
} = attempt;
let transport = match state
.read_provider_transport_snapshot(
&candidate.provider_id,
&candidate.endpoint_id,
&candidate.key_id,
)
.await
{
Ok(Some(snapshot)) => snapshot,
Ok(None) => {
mark_skipped_local_video_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_snapshot_missing",
)
.await;
return None;
}
Err(err) => {
warn!(
trace_id = %trace_id,
decision_kind = spec.decision_kind,
error = ?err,
"gateway local video decision provider transport read failed"
);
mark_skipped_local_video_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_snapshot_read_failed",
)
.await;
return None;
}
};
let transport_supported = match spec.family {
LocalVideoCreateFamily::OpenAi => {
supports_local_standard_transport_with_network(&transport, spec.api_format)
}
LocalVideoCreateFamily::Gemini => {
supports_local_gemini_transport_with_network(&transport, spec.api_format)
}
};
if !transport_supported {
mark_skipped_local_video_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_unsupported",
)
.await;
return None;
}
let auth = match spec.family {
LocalVideoCreateFamily::OpenAi => resolve_local_openai_chat_auth(&transport),
LocalVideoCreateFamily::Gemini => resolve_local_gemini_auth(&transport),
};
let Some((auth_header, auth_value)) = auth else {
mark_skipped_local_video_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_auth_unavailable",
)
.await;
return None;
};
let mapped_model = candidate.selected_provider_model_name.trim().to_string();
if mapped_model.is_empty() {
mark_skipped_local_video_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"mapped_model_missing",
)
.await;
return None;
}
let upstream_url = build_video_upstream_url(parts, &transport, &mapped_model, spec.family);
let Some(upstream_url) = upstream_url else {
mark_skipped_local_video_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"upstream_url_missing",
)
.await;
return None;
};
let Some(provider_request_body) = build_provider_request_body(
body_json,
spec.family,
&mapped_model,
transport.endpoint.body_rules.as_ref(),
) else {
mark_skipped_local_video_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"provider_request_body_missing",
)
.await;
return None;
};
let mut provider_request_headers = build_passthrough_headers_with_auth(
&parts.headers,
&auth_header,
&auth_value,
&BTreeMap::new(),
);
if !apply_local_header_rules(
&mut provider_request_headers,
transport.endpoint.header_rules.as_ref(),
&[&auth_header, "content-type"],
&provider_request_body,
Some(body_json),
) {
mark_skipped_local_video_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_header_rules_apply_failed",
)
.await;
return None;
}
let proxy = resolve_transport_proxy_snapshot_with_tunnel_affinity(state, &transport).await;
let tls_profile = resolve_transport_tls_profile(&transport);
Some(GatewayControlSyncDecisionResponse {
action: EXECUTION_RUNTIME_SYNC_DECISION_ACTION.to_string(),
decision_kind: Some(spec.decision_kind.to_string()),
execution_strategy: Some(
crate::gateway::ExecutionStrategy::LocalSameFormat
.as_str()
.to_string(),
),
conversion_mode: Some(crate::gateway::ConversionMode::None.as_str().to_string()),
request_id: Some(trace_id.to_string()),
candidate_id: Some(candidate_id.clone()),
provider_name: Some(transport.provider.name.clone()),
provider_id: Some(candidate.provider_id.clone()),
endpoint_id: Some(candidate.endpoint_id.clone()),
key_id: Some(candidate.key_id.clone()),
upstream_base_url: Some(transport.endpoint.base_url.clone()),
upstream_url: Some(upstream_url),
provider_request_method: Some(parts.method.to_string()),
auth_header: Some(auth_header),
auth_value: Some(auth_value),
provider_api_format: Some(spec.api_format.to_string()),
client_api_format: Some(spec.api_format.to_string()),
provider_contract: Some(spec.api_format.to_string()),
client_contract: Some(spec.api_format.to_string()),
model_name: Some(input.requested_model.clone()),
mapped_model: Some(mapped_model.clone()),
prompt_cache_key: None,
extra_headers: BTreeMap::new(),
provider_request_headers,
provider_request_body: Some(provider_request_body),
provider_request_body_base64: None,
content_type: parts
.headers
.get(http::header::CONTENT_TYPE)
.and_then(|value| value.to_str().ok())
.map(str::trim)
.filter(|value| !value.is_empty())
.map(ToOwned::to_owned),
proxy,
tls_profile,
timeouts: resolve_transport_execution_timeouts(&transport),
upstream_is_stream: false,
report_kind: Some(spec.report_kind.to_string()),
report_context: Some(json!({
"user_id": input.auth_context.user_id.clone(),
"api_key_id": input.auth_context.api_key_id.clone(),
"request_id": trace_id,
"candidate_id": candidate_id,
"candidate_index": candidate_index,
"retry_index": 0,
"model": input.requested_model.clone(),
"provider_name": transport.provider.name.clone(),
"provider_id": candidate.provider_id.clone(),
"endpoint_id": candidate.endpoint_id.clone(),
"key_id": candidate.key_id.clone(),
"provider_api_format": spec.api_format,
"client_api_format": spec.api_format,
"mapped_model": mapped_model,
"original_headers": collect_control_headers(&parts.headers),
"original_request_body": body_json,
"has_envelope": false,
"needs_conversion": false,
})),
auth_context: Some(input.auth_context.clone()),
})
}
fn build_provider_request_body(
body_json: &serde_json::Value,
family: LocalVideoCreateFamily,
mapped_model: &str,
body_rules: Option<&serde_json::Value>,
) -> Option<serde_json::Value> {
let mut provider_request_body = match family {
LocalVideoCreateFamily::OpenAi => {
let mut provider_request_body = body_json.as_object().cloned().unwrap_or_default();
provider_request_body
.insert("model".to_string(), Value::String(mapped_model.to_string()));
serde_json::Value::Object(provider_request_body)
}
LocalVideoCreateFamily::Gemini => body_json.clone(),
};
if !apply_local_body_rules(&mut provider_request_body, body_rules, Some(body_json)) {
return None;
}
Some(provider_request_body)
}
fn build_video_upstream_url(
parts: &http::request::Parts,
transport: &crate::gateway::provider_transport::GatewayProviderTransportSnapshot,
mapped_model: &str,
family: LocalVideoCreateFamily,
) -> Option<String> {
let custom_path = transport
.endpoint
.custom_path
.as_deref()
.map(str::trim)
.filter(|value| !value.is_empty());
if let Some(path) = custom_path {
let blocked_keys = match family {
LocalVideoCreateFamily::OpenAi => &[][..],
LocalVideoCreateFamily::Gemini => &["key"][..],
};
return build_passthrough_path_url(
&transport.endpoint.base_url,
path,
parts.uri.query(),
blocked_keys,
);
}
match family {
LocalVideoCreateFamily::OpenAi => build_passthrough_path_url(
&transport.endpoint.base_url,
parts.uri.path(),
parts.uri.query(),
&[],
),
LocalVideoCreateFamily::Gemini => build_gemini_video_predict_long_running_url(
&transport.endpoint.base_url,
mapped_model,
parts.uri.query(),
),
}
}
async fn materialize_local_video_create_candidate_attempts(
state: &AppState,
trace_id: &str,
input: &LocalVideoCreateDecisionInput,
candidates: Vec<GatewayMinimalCandidateSelectionCandidate>,
api_format: &str,
) -> Vec<LocalVideoCreateCandidateAttempt> {
let candidates = prefer_local_tunnel_owner_candidates(state, candidates).await;
let created_at_unix_secs = current_unix_secs();
let mut attempts = Vec::with_capacity(candidates.len());
for (candidate_index, candidate) in candidates.into_iter().enumerate() {
let generated_candidate_id = Uuid::new_v4().to_string();
let extra_data = json!({
"provider_api_format": api_format,
"client_api_format": api_format,
"global_model_id": candidate.global_model_id.clone(),
"global_model_name": candidate.global_model_name.clone(),
"model_id": candidate.model_id.clone(),
"selected_provider_model_name": candidate.selected_provider_model_name.clone(),
"mapping_matched_model": candidate.mapping_matched_model.clone(),
"provider_name": candidate.provider_name.clone(),
"key_name": candidate.key_name.clone(),
});
let candidate_id = match state
.upsert_request_candidate(UpsertRequestCandidateRecord {
id: generated_candidate_id.clone(),
request_id: trace_id.to_string(),
user_id: Some(input.auth_context.user_id.clone()),
api_key_id: Some(input.auth_context.api_key_id.clone()),
username: None,
api_key_name: None,
candidate_index: candidate_index as u32,
retry_index: 0,
provider_id: Some(candidate.provider_id.clone()),
endpoint_id: Some(candidate.endpoint_id.clone()),
key_id: Some(candidate.key_id.clone()),
status: RequestCandidateStatus::Available,
skip_reason: None,
is_cached: Some(false),
status_code: None,
error_type: None,
error_message: None,
latency_ms: None,
concurrent_requests: None,
extra_data: Some(extra_data),
required_capabilities: candidate.key_capabilities.clone(),
created_at_unix_secs: Some(created_at_unix_secs),
started_at_unix_secs: None,
finished_at_unix_secs: None,
})
.await
{
Ok(Some(stored)) => stored.id,
Ok(None) => generated_candidate_id.clone(),
Err(err) => {
warn!(
trace_id = %trace_id,
decision_api_format = api_format,
error = ?err,
"gateway local video decision request candidate upsert failed"
);
generated_candidate_id.clone()
}
};
attempts.push(LocalVideoCreateCandidateAttempt {
candidate,
candidate_index: candidate_index as u32,
candidate_id,
});
}
attempts
}
async fn mark_skipped_local_video_candidate(
state: &AppState,
input: &LocalVideoCreateDecisionInput,
trace_id: &str,
candidate: &GatewayMinimalCandidateSelectionCandidate,
candidate_index: u32,
candidate_id: &str,
skip_reason: &'static str,
) {
let terminal_unix_secs = current_unix_secs();
if let Err(err) = state
.upsert_request_candidate(UpsertRequestCandidateRecord {
id: candidate_id.to_string(),
request_id: trace_id.to_string(),
user_id: Some(input.auth_context.user_id.clone()),
api_key_id: Some(input.auth_context.api_key_id.clone()),
username: None,
api_key_name: None,
candidate_index,
retry_index: 0,
provider_id: Some(candidate.provider_id.clone()),
endpoint_id: Some(candidate.endpoint_id.clone()),
key_id: Some(candidate.key_id.clone()),
status: RequestCandidateStatus::Skipped,
skip_reason: Some(skip_reason.to_string()),
is_cached: Some(false),
status_code: None,
error_type: None,
error_message: None,
latency_ms: None,
concurrent_requests: None,
extra_data: None,
required_capabilities: candidate.key_capabilities.clone(),
created_at_unix_secs: None,
started_at_unix_secs: None,
finished_at_unix_secs: Some(terminal_unix_secs),
})
.await
{
warn!(
trace_id = %trace_id,
candidate_id = %candidate_id,
skip_reason,
error = ?err,
"gateway local video decision failed to persist skipped candidate"
);
}
}
async fn mark_unused_local_video_candidates(
state: &AppState,
remaining: Vec<LocalSyncPlanAndReport>,
) {
for plan_and_report in remaining {
record_local_request_candidate_status(
state,
&plan_and_report.plan,
plan_and_report.report_context.as_ref(),
RequestCandidateStatus::Unused,
None,
None,
None,
None,
None,
None,
)
.await;
}
}
fn extract_gemini_video_model_from_path(path: &str) -> Option<String> {
let suffix = path.strip_prefix("/v1beta/models/")?;
let model = suffix.split(':').next()?.trim();
if model.is_empty() {
return None;
}
Some(model.to_string())
}
@@ -0,0 +1,33 @@
use crate::gateway::ai_pipeline::planner::{
CLAUDE_CHAT_STREAM_PLAN_KIND, CLAUDE_CHAT_SYNC_PLAN_KIND,
};
use super::super::family::{LocalStandardSourceFamily, LocalStandardSourceMode, LocalStandardSpec};
pub(super) fn resolve_sync_spec(plan_kind: &str) -> Option<LocalStandardSpec> {
match plan_kind {
CLAUDE_CHAT_SYNC_PLAN_KIND => Some(LocalStandardSpec {
api_format: "claude:chat",
decision_kind: CLAUDE_CHAT_SYNC_PLAN_KIND,
report_kind: "claude_chat_sync_finalize",
family: LocalStandardSourceFamily::Standard,
mode: LocalStandardSourceMode::Chat,
require_streaming: false,
}),
_ => None,
}
}
pub(super) fn resolve_stream_spec(plan_kind: &str) -> Option<LocalStandardSpec> {
match plan_kind {
CLAUDE_CHAT_STREAM_PLAN_KIND => Some(LocalStandardSpec {
api_format: "claude:chat",
decision_kind: CLAUDE_CHAT_STREAM_PLAN_KIND,
report_kind: "claude_chat_stream_success",
family: LocalStandardSourceFamily::Standard,
mode: LocalStandardSourceMode::Chat,
require_streaming: true,
}),
_ => None,
}
}
@@ -0,0 +1,33 @@
use crate::gateway::ai_pipeline::planner::{
CLAUDE_CLI_STREAM_PLAN_KIND, CLAUDE_CLI_SYNC_PLAN_KIND,
};
use super::super::family::{LocalStandardSourceFamily, LocalStandardSourceMode, LocalStandardSpec};
pub(super) fn resolve_sync_spec(plan_kind: &str) -> Option<LocalStandardSpec> {
match plan_kind {
CLAUDE_CLI_SYNC_PLAN_KIND => Some(LocalStandardSpec {
api_format: "claude:cli",
decision_kind: CLAUDE_CLI_SYNC_PLAN_KIND,
report_kind: "claude_cli_sync_finalize",
family: LocalStandardSourceFamily::Standard,
mode: LocalStandardSourceMode::Cli,
require_streaming: false,
}),
_ => None,
}
}
pub(super) fn resolve_stream_spec(plan_kind: &str) -> Option<LocalStandardSpec> {
match plan_kind {
CLAUDE_CLI_STREAM_PLAN_KIND => Some(LocalStandardSpec {
api_format: "claude:cli",
decision_kind: CLAUDE_CLI_STREAM_PLAN_KIND,
report_kind: "claude_cli_stream_success",
family: LocalStandardSourceFamily::Standard,
mode: LocalStandardSourceMode::Cli,
require_streaming: true,
}),
_ => None,
}
}
@@ -0,0 +1,104 @@
use axum::body::Body;
use axum::http::Response;
use crate::gateway::{
AppState, GatewayControlDecision, GatewayControlSyncDecisionResponse, GatewayError,
};
use super::family::{
maybe_build_stream_via_standard_family_payload, maybe_build_sync_via_standard_family_payload,
maybe_execute_stream_via_standard_family_decision,
maybe_execute_sync_via_standard_family_decision,
};
pub(crate) use crate::gateway::ai_pipeline::conversion::request::normalize_claude_request_to_openai_chat_request;
mod chat;
mod cli;
pub(crate) async fn maybe_execute_sync_via_local_claude_decision(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Option<Response<Body>>, GatewayError> {
maybe_execute_sync_via_standard_family_decision(
state,
parts,
trace_id,
decision,
body_json,
plan_kind,
|plan_kind| {
chat::resolve_sync_spec(plan_kind).or_else(|| cli::resolve_sync_spec(plan_kind))
},
)
.await
}
pub(crate) async fn maybe_execute_stream_via_local_claude_decision(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Option<Response<Body>>, GatewayError> {
maybe_execute_stream_via_standard_family_decision(
state,
parts,
trace_id,
decision,
body_json,
plan_kind,
|plan_kind| {
chat::resolve_stream_spec(plan_kind).or_else(|| cli::resolve_stream_spec(plan_kind))
},
)
.await
}
pub(crate) async fn maybe_build_sync_local_claude_decision_payload(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
maybe_build_sync_via_standard_family_payload(
state,
parts,
trace_id,
decision,
body_json,
plan_kind,
|plan_kind| {
chat::resolve_sync_spec(plan_kind).or_else(|| cli::resolve_sync_spec(plan_kind))
},
)
.await
}
pub(crate) async fn maybe_build_stream_local_claude_decision_payload(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
maybe_build_stream_via_standard_family_payload(
state,
parts,
trace_id,
decision,
body_json,
plan_kind,
|plan_kind| {
chat::resolve_stream_spec(plan_kind).or_else(|| cli::resolve_stream_spec(plan_kind))
},
)
.await
}
@@ -0,0 +1,271 @@
use std::collections::BTreeSet;
use aether_data::repository::candidates::{RequestCandidateStatus, UpsertRequestCandidateRecord};
use serde_json::json;
use tracing::warn;
use uuid::Uuid;
use crate::gateway::request_candidates::current_unix_secs;
use crate::gateway::scheduler::{
list_selectable_candidates, GatewayMinimalCandidateSelectionCandidate,
};
use crate::gateway::{
append_execution_contract_fields_to_value, AppState, ConversionMode, ExecutionStrategy,
GatewayControlDecision, GatewayError,
};
use super::types::{
LocalStandardCandidateAttempt, LocalStandardDecisionInput, LocalStandardSourceFamily,
LocalStandardSourceMode, LocalStandardSpec,
};
pub(super) async fn resolve_local_standard_decision_input(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
spec: LocalStandardSpec,
) -> Option<LocalStandardDecisionInput> {
let Some(auth_context) = decision.auth_context.clone().filter(|auth_context| {
!auth_context.user_id.trim().is_empty() && !auth_context.api_key_id.trim().is_empty()
}) else {
return None;
};
let requested_model = match spec.family {
LocalStandardSourceFamily::Standard => body_json
.get("model")
.and_then(serde_json::Value::as_str)
.map(str::trim)
.filter(|value| !value.is_empty())
.map(ToOwned::to_owned)?,
LocalStandardSourceFamily::Gemini => extract_gemini_model_from_path(parts.uri.path())?,
};
let auth_snapshot = match state
.read_auth_api_key_snapshot(
&auth_context.user_id,
&auth_context.api_key_id,
current_unix_secs(),
)
.await
{
Ok(Some(snapshot)) => snapshot,
Ok(None) => return None,
Err(err) => {
warn!(
trace_id = %trace_id,
api_format = spec.api_format,
error = ?err,
"gateway local standard decision auth snapshot read failed"
);
return None;
}
};
Some(LocalStandardDecisionInput {
auth_context,
requested_model,
auth_snapshot,
})
}
pub(super) async fn materialize_local_standard_candidate_attempts(
state: &AppState,
trace_id: &str,
input: &LocalStandardDecisionInput,
spec: LocalStandardSpec,
) -> Result<Vec<LocalStandardCandidateAttempt>, GatewayError> {
let mut seen_candidates = BTreeSet::new();
let mut candidates = Vec::new();
for candidate_api_format in candidate_api_formats_for_spec(spec) {
let auth_snapshot = if *candidate_api_format == spec.api_format {
Some(&input.auth_snapshot)
} else {
None
};
let mut selected_candidates = list_selectable_candidates(
state,
candidate_api_format,
&input.requested_model,
spec.require_streaming,
auth_snapshot,
current_unix_secs(),
)
.await?;
if auth_snapshot.is_none() {
selected_candidates.retain(|candidate| {
auth_snapshot_allows_cross_format_candidate(
&input.auth_snapshot,
&input.requested_model,
candidate,
)
});
}
for candidate in selected_candidates {
let candidate_key = format!(
"{}:{}:{}:{}:{}:{}",
candidate.provider_id,
candidate.endpoint_id,
candidate.key_id,
candidate.model_id,
candidate.selected_provider_model_name,
candidate.endpoint_api_format,
);
if seen_candidates.insert(candidate_key) {
candidates.push(candidate);
}
}
}
let candidates =
crate::gateway::ai_pipeline::planner::prefer_local_tunnel_owner_candidates(state, candidates)
.await;
let created_at_unix_secs = current_unix_secs();
let mut attempts = Vec::with_capacity(candidates.len());
for (candidate_index, candidate) in candidates.into_iter().enumerate() {
let candidate_id = Uuid::new_v4().to_string();
let provider_api_format = candidate.endpoint_api_format.trim().to_ascii_lowercase();
let execution_strategy = if provider_api_format == spec.api_format {
ExecutionStrategy::LocalSameFormat
} else {
ExecutionStrategy::LocalCrossFormat
};
let conversion_mode = if crate::gateway::ai_pipeline::conversion::request_conversion_kind(
spec.api_format,
provider_api_format.as_str(),
)
.is_some()
{
ConversionMode::Bidirectional
} else {
ConversionMode::None
};
let extra_data = append_execution_contract_fields_to_value(
json!({
"provider_api_format": provider_api_format,
"client_api_format": spec.api_format,
"global_model_id": candidate.global_model_id.clone(),
"global_model_name": candidate.global_model_name.clone(),
"model_id": candidate.model_id.clone(),
"selected_provider_model_name": candidate.selected_provider_model_name.clone(),
"mapping_matched_model": candidate.mapping_matched_model.clone(),
"provider_name": candidate.provider_name.clone(),
"key_name": candidate.key_name.clone(),
}),
execution_strategy,
conversion_mode,
spec.api_format,
candidate.endpoint_api_format.as_str(),
);
let stored_candidate_id = match state
.upsert_request_candidate(UpsertRequestCandidateRecord {
id: candidate_id.clone(),
request_id: trace_id.to_string(),
user_id: Some(input.auth_context.user_id.clone()),
api_key_id: Some(input.auth_context.api_key_id.clone()),
username: None,
api_key_name: None,
candidate_index: candidate_index as u32,
retry_index: 0,
provider_id: Some(candidate.provider_id.clone()),
endpoint_id: Some(candidate.endpoint_id.clone()),
key_id: Some(candidate.key_id.clone()),
status: RequestCandidateStatus::Available,
skip_reason: None,
is_cached: Some(false),
status_code: None,
error_type: None,
error_message: None,
latency_ms: None,
concurrent_requests: None,
extra_data: Some(extra_data),
required_capabilities: candidate.key_capabilities.clone(),
created_at_unix_secs: Some(created_at_unix_secs),
started_at_unix_secs: None,
finished_at_unix_secs: None,
})
.await
{
Ok(Some(stored)) => stored.id,
Ok(None) => candidate_id.clone(),
Err(err) => {
warn!(
trace_id = %trace_id,
api_format = spec.api_format,
error = ?err,
"gateway local standard decision request candidate upsert failed"
);
candidate_id.clone()
}
};
attempts.push(LocalStandardCandidateAttempt {
candidate,
candidate_index: candidate_index as u32,
candidate_id: stored_candidate_id,
});
}
Ok(attempts)
}
fn auth_snapshot_allows_cross_format_candidate(
auth_snapshot: &crate::gateway::gateway_data::StoredGatewayAuthApiKeySnapshot,
requested_model: &str,
candidate: &GatewayMinimalCandidateSelectionCandidate,
) -> bool {
if let Some(allowed_providers) = auth_snapshot.effective_allowed_providers() {
let provider_allowed = allowed_providers.iter().any(|value| {
value
.trim()
.eq_ignore_ascii_case(candidate.provider_id.trim())
|| value
.trim()
.eq_ignore_ascii_case(candidate.provider_name.trim())
});
if !provider_allowed {
return false;
}
}
if let Some(allowed_models) = auth_snapshot.effective_allowed_models() {
let model_allowed = allowed_models
.iter()
.any(|value| value == requested_model || value == &candidate.global_model_name);
if !model_allowed {
return false;
}
}
true
}
fn candidate_api_formats_for_spec(spec: LocalStandardSpec) -> &'static [&'static str] {
match spec.mode {
LocalStandardSourceMode::Chat | LocalStandardSourceMode::Cli => &[
"openai:chat",
"openai:cli",
"openai:compact",
"claude:chat",
"claude:cli",
"gemini:chat",
"gemini:cli",
],
}
}
fn extract_gemini_model_from_path(path: &str) -> Option<String> {
let marker = "/models/";
let start = path.find(marker)? + marker.len();
let tail = &path[start..];
let end = tail.find(':').unwrap_or(tail.len());
let model = tail[..end].trim();
if model.is_empty() {
None
} else {
Some(model.to_string())
}
}
@@ -0,0 +1,236 @@
use axum::body::Body;
use axum::http::Response;
use tracing::warn;
use crate::gateway::ai_pipeline::planner::family_core::{
execute_stream_plan_and_reports, execute_sync_plan_and_reports,
};
use crate::gateway::ai_pipeline::planner::plan_builders::{
build_gemini_stream_plan_from_decision, build_gemini_sync_plan_from_decision,
build_standard_stream_plan_from_decision, build_standard_sync_plan_from_decision,
LocalStreamPlanAndReport, LocalSyncPlanAndReport,
};
use crate::gateway::{AppState, GatewayControlDecision, GatewayControlSyncDecisionResponse, GatewayError};
use super::candidates::{
materialize_local_standard_candidate_attempts, resolve_local_standard_decision_input,
};
use super::payload::maybe_build_local_standard_decision_payload_for_candidate;
use super::types::{LocalStandardSourceFamily, LocalStandardSpec};
pub(crate) async fn maybe_execute_sync_via_standard_family_decision(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
resolve_sync_spec: fn(&str) -> Option<LocalStandardSpec>,
) -> Result<Option<Response<Body>>, GatewayError> {
let Some(spec) = resolve_sync_spec(plan_kind) else {
return Ok(None);
};
let plan_and_reports =
build_local_sync_plan_and_reports(state, parts, trace_id, decision, body_json, spec)
.await?;
if plan_and_reports.is_empty() {
return Ok(None);
}
execute_sync_plan_and_reports(state, parts, trace_id, decision, plan_kind, plan_and_reports)
.await
}
pub(crate) async fn maybe_execute_stream_via_standard_family_decision(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
resolve_stream_spec: fn(&str) -> Option<LocalStandardSpec>,
) -> Result<Option<Response<Body>>, GatewayError> {
let Some(spec) = resolve_stream_spec(plan_kind) else {
return Ok(None);
};
let plan_and_reports =
build_local_stream_plan_and_reports(state, parts, trace_id, decision, body_json, spec)
.await?;
if plan_and_reports.is_empty() {
return Ok(None);
}
execute_stream_plan_and_reports(state, trace_id, decision, plan_kind, plan_and_reports).await
}
pub(crate) async fn maybe_build_sync_via_standard_family_payload(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
resolve_sync_spec: fn(&str) -> Option<LocalStandardSpec>,
) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
let Some(spec) = resolve_sync_spec(plan_kind) else {
return Ok(None);
};
let Some(input) =
resolve_local_standard_decision_input(state, parts, trace_id, decision, body_json, spec)
.await
else {
return Ok(None);
};
let attempts =
materialize_local_standard_candidate_attempts(state, trace_id, &input, spec).await?;
for attempt in attempts {
if let Some(payload) = maybe_build_local_standard_decision_payload_for_candidate(
state, parts, trace_id, body_json, &input, attempt, spec,
)
.await
{
return Ok(Some(payload));
}
}
Ok(None)
}
pub(crate) async fn maybe_build_stream_via_standard_family_payload(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
resolve_stream_spec: fn(&str) -> Option<LocalStandardSpec>,
) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
let Some(spec) = resolve_stream_spec(plan_kind) else {
return Ok(None);
};
let Some(input) =
resolve_local_standard_decision_input(state, parts, trace_id, decision, body_json, spec)
.await
else {
return Ok(None);
};
let attempts =
materialize_local_standard_candidate_attempts(state, trace_id, &input, spec).await?;
for attempt in attempts {
if let Some(payload) = maybe_build_local_standard_decision_payload_for_candidate(
state, parts, trace_id, body_json, &input, attempt, spec,
)
.await
{
return Ok(Some(payload));
}
}
Ok(None)
}
async fn build_local_sync_plan_and_reports(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
spec: LocalStandardSpec,
) -> Result<Vec<LocalSyncPlanAndReport>, GatewayError> {
let Some(input) =
resolve_local_standard_decision_input(state, parts, trace_id, decision, body_json, spec)
.await
else {
return Ok(Vec::new());
};
let attempts =
materialize_local_standard_candidate_attempts(state, trace_id, &input, spec).await?;
let mut plans = Vec::new();
for attempt in attempts {
let Some(payload) = maybe_build_local_standard_decision_payload_for_candidate(
state, parts, trace_id, body_json, &input, attempt, spec,
)
.await
else {
continue;
};
let built = match spec.family {
LocalStandardSourceFamily::Standard => {
build_standard_sync_plan_from_decision(parts, body_json, payload)
}
LocalStandardSourceFamily::Gemini => {
build_gemini_sync_plan_from_decision(parts, body_json, payload)
}
};
match built {
Ok(Some(value)) => plans.push(value),
Ok(None) => {}
Err(err) => {
warn!(
trace_id = %trace_id,
api_format = spec.api_format,
error = ?err,
"gateway local standard sync plan build failed"
);
}
}
}
Ok(plans)
}
async fn build_local_stream_plan_and_reports(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
spec: LocalStandardSpec,
) -> Result<Vec<LocalStreamPlanAndReport>, GatewayError> {
let Some(input) =
resolve_local_standard_decision_input(state, parts, trace_id, decision, body_json, spec)
.await
else {
return Ok(Vec::new());
};
let attempts =
materialize_local_standard_candidate_attempts(state, trace_id, &input, spec).await?;
let mut plans = Vec::new();
for attempt in attempts {
let Some(payload) = maybe_build_local_standard_decision_payload_for_candidate(
state, parts, trace_id, body_json, &input, attempt, spec,
)
.await
else {
continue;
};
let built = match spec.family {
LocalStandardSourceFamily::Standard => {
build_standard_stream_plan_from_decision(parts, body_json, payload, false)
}
LocalStandardSourceFamily::Gemini => {
build_gemini_stream_plan_from_decision(parts, body_json, payload)
}
};
match built {
Ok(Some(value)) => plans.push(value),
Ok(None) => {}
Err(err) => {
warn!(
trace_id = %trace_id,
api_format = spec.api_format,
error = ?err,
"gateway local standard stream plan build failed"
);
}
}
}
Ok(plans)
}
@@ -0,0 +1,13 @@
mod candidates;
mod execute;
mod payload;
mod types;
pub(crate) use self::execute::{
maybe_build_stream_via_standard_family_payload, maybe_build_sync_via_standard_family_payload,
maybe_execute_stream_via_standard_family_decision,
maybe_execute_sync_via_standard_family_decision,
};
pub(crate) use self::types::{
LocalStandardSourceFamily, LocalStandardSourceMode, LocalStandardSpec,
};
@@ -0,0 +1,360 @@
use std::collections::BTreeMap;
use aether_data::repository::candidates::{RequestCandidateStatus, UpsertRequestCandidateRecord};
use serde_json::json;
use tracing::warn;
use crate::gateway::headers::collect_control_headers;
use crate::gateway::provider_transport::{
apply_local_header_rules, build_openai_passthrough_headers, ensure_upstream_auth_header,
resolve_transport_execution_timeouts, resolve_transport_proxy_snapshot_with_tunnel_affinity,
resolve_transport_tls_profile, LocalResolvedOAuthRequestAuth,
};
use crate::gateway::request_candidates::current_unix_secs;
use crate::gateway::{
append_execution_contract_fields_to_value, AppState, ConversionMode, ExecutionStrategy,
GatewayControlSyncDecisionResponse, EXECUTION_RUNTIME_STREAM_DECISION_ACTION,
EXECUTION_RUNTIME_SYNC_DECISION_ACTION,
};
use super::types::{LocalStandardCandidateAttempt, LocalStandardDecisionInput, LocalStandardSpec};
pub(super) async fn maybe_build_local_standard_decision_payload_for_candidate(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
body_json: &serde_json::Value,
input: &LocalStandardDecisionInput,
attempt: LocalStandardCandidateAttempt,
spec: LocalStandardSpec,
) -> Option<GatewayControlSyncDecisionResponse> {
let LocalStandardCandidateAttempt {
candidate,
candidate_index,
candidate_id,
} = attempt;
let provider_api_format = candidate.endpoint_api_format.trim().to_ascii_lowercase();
let Some(conversion_kind) = crate::gateway::ai_pipeline::conversion::request_conversion_kind(
spec.api_format,
provider_api_format.as_str(),
) else {
if provider_api_format == spec.api_format {
return None;
}
return None;
};
let transport = match state
.read_provider_transport_snapshot(
&candidate.provider_id,
&candidate.endpoint_id,
&candidate.key_id,
)
.await
{
Ok(Some(snapshot)) => snapshot,
Ok(None) => {
mark_skipped_local_standard_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_snapshot_missing",
)
.await;
return None;
}
Err(err) => {
warn!(
trace_id = %trace_id,
api_format = spec.api_format,
error = ?err,
"gateway local standard decision provider transport read failed"
);
mark_skipped_local_standard_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_snapshot_read_failed",
)
.await;
return None;
}
};
if !crate::gateway::ai_pipeline::conversion::request_conversion_transport_supported(
&transport,
conversion_kind,
) {
mark_skipped_local_standard_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_unsupported",
)
.await;
return None;
}
let resolved_auth = crate::gateway::ai_pipeline::conversion::request_conversion_direct_auth(
&transport,
conversion_kind,
);
let oauth_auth = if resolved_auth.is_none() {
match state.resolve_local_oauth_request_auth(&transport).await {
Ok(Some(LocalResolvedOAuthRequestAuth::Header { name, value })) => Some((name, value)),
Ok(Some(LocalResolvedOAuthRequestAuth::Kiro(_))) => None,
Ok(None) => None,
Err(err) => {
warn!(
trace_id = %trace_id,
api_format = spec.api_format,
provider_type = %transport.provider.provider_type,
error = ?err,
"gateway local standard oauth auth resolution failed"
);
None
}
}
} else {
None
};
let Some((auth_header, auth_value)) = resolved_auth.or(oauth_auth) else {
mark_skipped_local_standard_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_auth_unavailable",
)
.await;
return None;
};
let mapped_model = candidate.selected_provider_model_name.trim().to_string();
if mapped_model.is_empty() {
mark_skipped_local_standard_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"mapped_model_missing",
)
.await;
return None;
}
let provider_request_body =
match crate::gateway::ai_pipeline::planner::standard::build_standard_request_body(
body_json,
spec.api_format,
&mapped_model,
provider_api_format.as_str(),
parts.uri.path(),
spec.require_streaming,
transport.endpoint.body_rules.as_ref(),
) {
Some(body) => body,
None => {
mark_skipped_local_standard_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"provider_request_body_missing",
)
.await;
return None;
}
};
let upstream_url =
match crate::gateway::ai_pipeline::planner::standard::build_standard_upstream_url(
parts,
&transport,
&mapped_model,
provider_api_format.as_str(),
spec.require_streaming,
) {
Some(url) => url,
None => {
mark_skipped_local_standard_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"upstream_url_missing",
)
.await;
return None;
}
};
let mut provider_request_headers = build_openai_passthrough_headers(
&parts.headers,
&auth_header,
&auth_value,
&BTreeMap::new(),
Some("application/json"),
);
if !apply_local_header_rules(
&mut provider_request_headers,
transport.endpoint.header_rules.as_ref(),
&[&auth_header, "content-type"],
&provider_request_body,
Some(body_json),
) {
mark_skipped_local_standard_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_header_rules_apply_failed",
)
.await;
return None;
}
ensure_upstream_auth_header(&mut provider_request_headers, &auth_header, &auth_value);
if spec.require_streaming {
provider_request_headers
.entry("accept".to_string())
.or_insert_with(|| "text/event-stream".to_string());
}
Some(GatewayControlSyncDecisionResponse {
action: if spec.require_streaming {
EXECUTION_RUNTIME_STREAM_DECISION_ACTION.to_string()
} else {
EXECUTION_RUNTIME_SYNC_DECISION_ACTION.to_string()
},
decision_kind: Some(spec.decision_kind.to_string()),
execution_strategy: Some(ExecutionStrategy::LocalCrossFormat.as_str().to_string()),
conversion_mode: Some(ConversionMode::Bidirectional.as_str().to_string()),
request_id: Some(trace_id.to_string()),
candidate_id: Some(candidate_id.clone()),
provider_name: Some(candidate.provider_name.clone()),
provider_id: Some(candidate.provider_id.clone()),
endpoint_id: Some(candidate.endpoint_id.clone()),
key_id: Some(candidate.key_id.clone()),
upstream_base_url: Some(transport.endpoint.base_url.clone()),
upstream_url: Some(upstream_url.clone()),
provider_request_method: None,
auth_header: Some(auth_header),
auth_value: Some(auth_value),
provider_api_format: Some(provider_api_format.clone()),
client_api_format: Some(spec.api_format.to_string()),
provider_contract: Some(provider_api_format.clone()),
client_contract: Some(spec.api_format.to_string()),
model_name: Some(input.requested_model.clone()),
mapped_model: Some(mapped_model.clone()),
prompt_cache_key: None,
extra_headers: BTreeMap::new(),
provider_request_headers: provider_request_headers.clone(),
provider_request_body: Some(provider_request_body.clone()),
provider_request_body_base64: None,
content_type: Some("application/json".to_string()),
proxy: resolve_transport_proxy_snapshot_with_tunnel_affinity(state, &transport).await,
tls_profile: resolve_transport_tls_profile(&transport),
timeouts: resolve_transport_execution_timeouts(&transport),
upstream_is_stream: spec.require_streaming,
report_kind: Some(spec.report_kind.to_string()),
report_context: Some(append_execution_contract_fields_to_value(
json!({
"user_id": input.auth_context.user_id,
"api_key_id": input.auth_context.api_key_id,
"request_id": trace_id,
"candidate_id": candidate_id,
"candidate_index": candidate_index,
"retry_index": 0,
"model": input.requested_model,
"provider_name": candidate.provider_name,
"provider_id": candidate.provider_id,
"endpoint_id": candidate.endpoint_id,
"key_id": candidate.key_id,
"provider_api_format": provider_api_format,
"client_api_format": spec.api_format,
"mapped_model": mapped_model,
"upstream_url": upstream_url,
"provider_request_method": serde_json::Value::Null,
"provider_request_headers": provider_request_headers,
"provider_request_body": provider_request_body,
"original_headers": collect_control_headers(&parts.headers),
"original_request_body": body_json,
"has_envelope": false,
"needs_conversion": true,
}),
ExecutionStrategy::LocalCrossFormat,
ConversionMode::Bidirectional,
spec.api_format,
candidate.endpoint_api_format.as_str(),
)),
auth_context: Some(input.auth_context.clone()),
})
}
pub(super) async fn mark_skipped_local_standard_candidate(
state: &AppState,
input: &LocalStandardDecisionInput,
trace_id: &str,
candidate: &crate::gateway::scheduler::GatewayMinimalCandidateSelectionCandidate,
candidate_index: u32,
candidate_id: &str,
skip_reason: &'static str,
) {
if let Err(err) = state
.upsert_request_candidate(UpsertRequestCandidateRecord {
id: candidate_id.to_string(),
request_id: trace_id.to_string(),
user_id: Some(input.auth_context.user_id.clone()),
api_key_id: Some(input.auth_context.api_key_id.clone()),
username: None,
api_key_name: None,
candidate_index,
retry_index: 0,
provider_id: Some(candidate.provider_id.clone()),
endpoint_id: Some(candidate.endpoint_id.clone()),
key_id: Some(candidate.key_id.clone()),
status: RequestCandidateStatus::Skipped,
skip_reason: Some(skip_reason.to_string()),
is_cached: Some(false),
status_code: None,
error_type: None,
error_message: None,
latency_ms: None,
concurrent_requests: None,
extra_data: None,
required_capabilities: candidate.key_capabilities.clone(),
created_at_unix_secs: None,
started_at_unix_secs: None,
finished_at_unix_secs: Some(current_unix_secs()),
})
.await
{
warn!(
trace_id = %trace_id,
candidate_id = %candidate_id,
skip_reason,
error = ?err,
"gateway local standard decision failed to persist skipped candidate"
);
}
}
@@ -0,0 +1,35 @@
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub(crate) enum LocalStandardSourceFamily {
Standard,
Gemini,
}
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub(crate) enum LocalStandardSourceMode {
Chat,
Cli,
}
#[derive(Debug, Clone, Copy)]
pub(crate) struct LocalStandardSpec {
pub(crate) api_format: &'static str,
pub(crate) decision_kind: &'static str,
pub(crate) report_kind: &'static str,
pub(crate) family: LocalStandardSourceFamily,
pub(crate) mode: LocalStandardSourceMode,
pub(crate) require_streaming: bool,
}
#[derive(Debug, Clone)]
pub(super) struct LocalStandardDecisionInput {
pub(super) auth_context: crate::gateway::GatewayControlAuthContext,
pub(super) requested_model: String,
pub(super) auth_snapshot: crate::gateway::gateway_data::StoredGatewayAuthApiKeySnapshot,
}
#[derive(Debug, Clone)]
pub(super) struct LocalStandardCandidateAttempt {
pub(super) candidate: crate::gateway::scheduler::GatewayMinimalCandidateSelectionCandidate,
pub(super) candidate_index: u32,
pub(super) candidate_id: String,
}
@@ -0,0 +1,33 @@
use crate::gateway::ai_pipeline::planner::{
GEMINI_CHAT_STREAM_PLAN_KIND, GEMINI_CHAT_SYNC_PLAN_KIND,
};
use super::super::family::{LocalStandardSourceFamily, LocalStandardSourceMode, LocalStandardSpec};
pub(super) fn resolve_sync_spec(plan_kind: &str) -> Option<LocalStandardSpec> {
match plan_kind {
GEMINI_CHAT_SYNC_PLAN_KIND => Some(LocalStandardSpec {
api_format: "gemini:chat",
decision_kind: GEMINI_CHAT_SYNC_PLAN_KIND,
report_kind: "gemini_chat_sync_finalize",
family: LocalStandardSourceFamily::Gemini,
mode: LocalStandardSourceMode::Chat,
require_streaming: false,
}),
_ => None,
}
}
pub(super) fn resolve_stream_spec(plan_kind: &str) -> Option<LocalStandardSpec> {
match plan_kind {
GEMINI_CHAT_STREAM_PLAN_KIND => Some(LocalStandardSpec {
api_format: "gemini:chat",
decision_kind: GEMINI_CHAT_STREAM_PLAN_KIND,
report_kind: "gemini_chat_stream_success",
family: LocalStandardSourceFamily::Gemini,
mode: LocalStandardSourceMode::Chat,
require_streaming: true,
}),
_ => None,
}
}
@@ -0,0 +1,33 @@
use crate::gateway::ai_pipeline::planner::{
GEMINI_CLI_STREAM_PLAN_KIND, GEMINI_CLI_SYNC_PLAN_KIND,
};
use super::super::family::{LocalStandardSourceFamily, LocalStandardSourceMode, LocalStandardSpec};
pub(super) fn resolve_sync_spec(plan_kind: &str) -> Option<LocalStandardSpec> {
match plan_kind {
GEMINI_CLI_SYNC_PLAN_KIND => Some(LocalStandardSpec {
api_format: "gemini:cli",
decision_kind: GEMINI_CLI_SYNC_PLAN_KIND,
report_kind: "gemini_cli_sync_finalize",
family: LocalStandardSourceFamily::Gemini,
mode: LocalStandardSourceMode::Cli,
require_streaming: false,
}),
_ => None,
}
}
pub(super) fn resolve_stream_spec(plan_kind: &str) -> Option<LocalStandardSpec> {
match plan_kind {
GEMINI_CLI_STREAM_PLAN_KIND => Some(LocalStandardSpec {
api_format: "gemini:cli",
decision_kind: GEMINI_CLI_STREAM_PLAN_KIND,
report_kind: "gemini_cli_stream_success",
family: LocalStandardSourceFamily::Gemini,
mode: LocalStandardSourceMode::Cli,
require_streaming: true,
}),
_ => None,
}
}
@@ -0,0 +1,104 @@
use axum::body::Body;
use axum::http::Response;
use crate::gateway::{
AppState, GatewayControlDecision, GatewayControlSyncDecisionResponse, GatewayError,
};
use super::family::{
maybe_build_stream_via_standard_family_payload, maybe_build_sync_via_standard_family_payload,
maybe_execute_stream_via_standard_family_decision,
maybe_execute_sync_via_standard_family_decision,
};
pub(crate) use crate::gateway::ai_pipeline::conversion::request::normalize_gemini_request_to_openai_chat_request;
mod chat;
mod cli;
pub(crate) async fn maybe_execute_sync_via_local_gemini_decision(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Option<Response<Body>>, GatewayError> {
maybe_execute_sync_via_standard_family_decision(
state,
parts,
trace_id,
decision,
body_json,
plan_kind,
|plan_kind| {
chat::resolve_sync_spec(plan_kind).or_else(|| cli::resolve_sync_spec(plan_kind))
},
)
.await
}
pub(crate) async fn maybe_execute_stream_via_local_gemini_decision(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Option<Response<Body>>, GatewayError> {
maybe_execute_stream_via_standard_family_decision(
state,
parts,
trace_id,
decision,
body_json,
plan_kind,
|plan_kind| {
chat::resolve_stream_spec(plan_kind).or_else(|| cli::resolve_stream_spec(plan_kind))
},
)
.await
}
pub(crate) async fn maybe_build_sync_local_gemini_decision_payload(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
maybe_build_sync_via_standard_family_payload(
state,
parts,
trace_id,
decision,
body_json,
plan_kind,
|plan_kind| {
chat::resolve_sync_spec(plan_kind).or_else(|| cli::resolve_sync_spec(plan_kind))
},
)
.await
}
pub(crate) async fn maybe_build_stream_local_gemini_decision_payload(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
maybe_build_stream_via_standard_family_payload(
state,
parts,
trace_id,
decision,
body_json,
plan_kind,
|plan_kind| {
chat::resolve_stream_spec(plan_kind).or_else(|| cli::resolve_stream_spec(plan_kind))
},
)
.await
}
@@ -0,0 +1,242 @@
use aether_contracts::{ExecutionPlan, RequestBody};
use super::{
augment_sync_report_context, generic_decision_missing_exact_provider_request,
LocalStreamPlanAndReport, LocalSyncPlanAndReport,
};
use crate::gateway::provider_transport::ensure_upstream_auth_header;
use crate::gateway::{GatewayControlSyncDecisionResponse, GatewayError};
pub(crate) fn build_gemini_sync_plan_from_decision(
_parts: &http::request::Parts,
_body_json: &serde_json::Value,
payload: GatewayControlSyncDecisionResponse,
) -> Result<Option<LocalSyncPlanAndReport>, GatewayError> {
if generic_decision_missing_exact_provider_request(&payload) {
return Ok(None);
}
let Some(request_id) = payload
.request_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(provider_id) = payload
.provider_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(endpoint_id) = payload
.endpoint_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(key_id) = payload
.key_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(url) = payload
.upstream_url
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let auth_header = payload
.auth_header
.clone()
.filter(|value| !value.trim().is_empty());
let auth_value = payload
.auth_value
.clone()
.filter(|value| !value.trim().is_empty());
if auth_header.is_some() != auth_value.is_some() {
return Ok(None);
}
let Some(provider_api_format) = payload
.provider_api_format
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(client_api_format) = payload
.client_api_format
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(provider_request_body_value) = payload.provider_request_body.clone() else {
return Ok(None);
};
let mut provider_request_headers = payload.provider_request_headers.clone();
if let (Some(auth_header), Some(auth_value)) = (auth_header.as_deref(), auth_value.as_deref()) {
ensure_upstream_auth_header(&mut provider_request_headers, auth_header, auth_value);
}
if payload.upstream_is_stream {
provider_request_headers
.entry("accept".to_string())
.or_insert_with(|| "text/event-stream".to_string());
}
let plan = ExecutionPlan {
request_id,
candidate_id: payload.candidate_id.clone(),
provider_name: payload.provider_name.clone(),
provider_id,
endpoint_id,
key_id,
method: "POST".to_string(),
url,
headers: std::mem::take(&mut provider_request_headers),
content_type: payload
.content_type
.clone()
.or_else(|| Some("application/json".to_string())),
content_encoding: None,
body: RequestBody::from_json(provider_request_body_value.clone()),
stream: payload.upstream_is_stream,
client_api_format,
provider_api_format,
model_name: payload.model_name.clone(),
proxy: payload.proxy.clone(),
tls_profile: payload.tls_profile.clone(),
timeouts: payload.timeouts.clone(),
};
let report_context = augment_sync_report_context(
payload.report_context,
&plan.headers,
&provider_request_body_value,
)?;
Ok(Some(LocalSyncPlanAndReport {
plan,
report_kind: payload.report_kind,
report_context,
}))
}
pub(crate) fn build_gemini_stream_plan_from_decision(
_parts: &http::request::Parts,
_body_json: &serde_json::Value,
payload: GatewayControlSyncDecisionResponse,
) -> Result<Option<LocalStreamPlanAndReport>, GatewayError> {
if generic_decision_missing_exact_provider_request(&payload) {
return Ok(None);
}
let Some(request_id) = payload
.request_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(provider_id) = payload
.provider_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(endpoint_id) = payload
.endpoint_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(key_id) = payload
.key_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(url) = payload
.upstream_url
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let auth_header = payload
.auth_header
.clone()
.filter(|value| !value.trim().is_empty());
let auth_value = payload
.auth_value
.clone()
.filter(|value| !value.trim().is_empty());
if auth_header.is_some() != auth_value.is_some() {
return Ok(None);
}
let Some(provider_api_format) = payload
.provider_api_format
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(client_api_format) = payload
.client_api_format
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(provider_request_body_value) = payload.provider_request_body.clone() else {
return Ok(None);
};
let mut provider_request_headers = payload.provider_request_headers.clone();
if let (Some(auth_header), Some(auth_value)) = (auth_header.as_deref(), auth_value.as_deref()) {
ensure_upstream_auth_header(&mut provider_request_headers, auth_header, auth_value);
}
provider_request_headers.insert("accept".to_string(), "text/event-stream".to_string());
let plan = ExecutionPlan {
request_id,
candidate_id: payload.candidate_id.clone(),
provider_name: payload.provider_name.clone(),
provider_id,
endpoint_id,
key_id,
method: "POST".to_string(),
url,
headers: std::mem::take(&mut provider_request_headers),
content_type: payload
.content_type
.clone()
.or_else(|| Some("application/json".to_string())),
content_encoding: None,
body: RequestBody::from_json(provider_request_body_value.clone()),
stream: true,
client_api_format,
provider_api_format,
model_name: payload.model_name.clone(),
proxy: payload.proxy.clone(),
tls_profile: payload.tls_profile.clone(),
timeouts: payload.timeouts.clone(),
};
let report_context = augment_sync_report_context(
payload.report_context,
&plan.headers,
&provider_request_body_value,
)?;
Ok(Some(LocalStreamPlanAndReport {
plan,
report_kind: payload.report_kind,
report_context,
}))
}
@@ -0,0 +1,258 @@
use std::collections::BTreeMap;
use serde_json::{json, Map, Value};
use super::{
claude::normalize_claude_request_to_openai_chat_request,
gemini::normalize_gemini_request_to_openai_chat_request,
};
use crate::gateway::ai_pipeline::conversion::request::{
convert_openai_chat_request_to_claude_request, convert_openai_chat_request_to_gemini_request,
convert_openai_chat_request_to_openai_cli_request,
normalize_openai_cli_request_to_openai_chat_request,
};
use crate::gateway::ai_pipeline::conversion::{request_conversion_kind, RequestConversionKind};
use crate::gateway::provider_transport::{
apply_local_body_rules, build_claude_messages_url, build_gemini_content_url,
build_openai_chat_url, build_openai_cli_url, build_passthrough_path_url,
};
pub(crate) fn build_standard_request_body(
body_json: &Value,
client_api_format: &str,
mapped_model: &str,
provider_api_format: &str,
request_path: &str,
upstream_is_stream: bool,
body_rules: Option<&Value>,
) -> Option<Value> {
let canonical_request = normalize_standard_request_to_openai_chat_request(
body_json,
client_api_format,
request_path,
)?;
let conversion_kind = request_conversion_kind(client_api_format, provider_api_format)?;
let mut provider_request_body = match conversion_kind {
RequestConversionKind::ToOpenAIChat => {
build_openai_chat_request_body(&canonical_request, mapped_model, upstream_is_stream)?
}
RequestConversionKind::ToOpenAIFamilyCli => {
convert_openai_chat_request_to_openai_cli_request(
&canonical_request,
mapped_model,
upstream_is_stream,
false,
)?
}
RequestConversionKind::ToOpenAICompact => {
convert_openai_chat_request_to_openai_cli_request(
&canonical_request,
mapped_model,
false,
true,
)?
}
RequestConversionKind::ToClaudeStandard => convert_openai_chat_request_to_claude_request(
&canonical_request,
mapped_model,
upstream_is_stream,
)?,
RequestConversionKind::ToGeminiStandard => convert_openai_chat_request_to_gemini_request(
&canonical_request,
mapped_model,
upstream_is_stream,
)?,
};
if !apply_local_body_rules(&mut provider_request_body, body_rules, Some(body_json)) {
return None;
}
Some(provider_request_body)
}
pub(crate) fn build_standard_upstream_url(
parts: &http::request::Parts,
transport: &crate::gateway::provider_transport::GatewayProviderTransportSnapshot,
mapped_model: &str,
provider_api_format: &str,
upstream_is_stream: bool,
) -> Option<String> {
let custom_path = transport
.endpoint
.custom_path
.as_deref()
.map(str::trim)
.filter(|value| !value.is_empty());
match custom_path {
Some(path) => {
build_passthrough_path_url(&transport.endpoint.base_url, path, parts.uri.query(), &[])
}
None => match provider_api_format.trim().to_ascii_lowercase().as_str() {
"openai:chat" => Some(build_openai_chat_url(
&transport.endpoint.base_url,
parts.uri.query(),
)),
"openai:cli" => Some(build_openai_cli_url(
&transport.endpoint.base_url,
parts.uri.query(),
false,
)),
"openai:compact" => Some(build_openai_cli_url(
&transport.endpoint.base_url,
parts.uri.query(),
true,
)),
"claude:chat" | "claude:cli" => Some(build_claude_messages_url(
&transport.endpoint.base_url,
parts.uri.query(),
)),
"gemini:chat" | "gemini:cli" => build_gemini_content_url(
&transport.endpoint.base_url,
mapped_model,
upstream_is_stream,
parts.uri.query(),
),
_ => None,
},
}
}
pub(crate) fn normalize_standard_request_to_openai_chat_request(
body_json: &Value,
client_api_format: &str,
request_path: &str,
) -> Option<Value> {
match client_api_format.trim().to_ascii_lowercase().as_str() {
"openai:chat" => Some(body_json.clone()),
"openai:cli" | "openai:compact" => {
normalize_openai_cli_request_to_openai_chat_request(body_json)
}
"claude:chat" | "claude:cli" => normalize_claude_request_to_openai_chat_request(body_json),
"gemini:chat" | "gemini:cli" => {
normalize_gemini_request_to_openai_chat_request(body_json, request_path)
}
_ => None,
}
}
fn build_openai_chat_request_body(
body_json: &Value,
mapped_model: &str,
upstream_is_stream: bool,
) -> Option<Value> {
let request_body_object = body_json.as_object()?;
let mut provider_request_body = serde_json::Map::from_iter(
request_body_object
.iter()
.map(|(key, value)| (key.clone(), value.clone())),
);
provider_request_body.insert("model".to_string(), Value::String(mapped_model.to_string()));
if upstream_is_stream {
provider_request_body.insert("stream".to_string(), Value::Bool(true));
}
Some(Value::Object(provider_request_body))
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn builds_openai_chat_request_from_claude_chat_source() {
let request = json!({
"model": "claude-3-7-sonnet",
"system": "You are concise.",
"messages": [
{
"role": "user",
"content": [{"type": "text", "text": "Hello from Claude"}]
}
],
"max_tokens": 128
});
let converted = build_standard_request_body(
&request,
"claude:chat",
"gpt-5",
"openai:chat",
"/v1/messages",
false,
None,
)
.expect("claude chat should convert to openai chat");
assert_eq!(converted["model"], "gpt-5");
assert_eq!(converted["messages"][0]["role"], "system");
assert_eq!(converted["messages"][0]["content"], "You are concise.");
assert_eq!(converted["messages"][1]["role"], "user");
assert_eq!(converted["messages"][1]["content"], "Hello from Claude");
}
#[test]
fn builds_claude_chat_request_from_gemini_chat_source() {
let request = json!({
"systemInstruction": {
"parts": [{"text": "Be brief."}]
},
"contents": [
{
"role": "user",
"parts": [{"text": "Hello from Gemini"}]
}
]
});
let converted = build_standard_request_body(
&request,
"gemini:chat",
"claude-sonnet-4-5",
"claude:chat",
"/v1beta/models/gemini-2.5-pro:generateContent",
false,
None,
)
.expect("gemini chat should convert to claude chat");
assert_eq!(converted["model"], "claude-sonnet-4-5");
assert_eq!(converted["messages"][0]["role"], "user");
assert!(
converted["messages"]
.to_string()
.contains("Hello from Gemini"),
"converted claude payload should retain the gemini user text: {converted}"
);
}
#[test]
fn builds_gemini_cli_request_from_claude_cli_source() {
let request = json!({
"model": "claude-sonnet-4-5",
"messages": [
{
"role": "user",
"content": [{"type": "text", "text": "Need CLI output"}]
}
],
"max_tokens": 64
});
let converted = build_standard_request_body(
&request,
"claude:cli",
"gemini-2.5-pro",
"gemini:cli",
"/v1/messages",
false,
None,
)
.expect("claude cli should convert to gemini cli");
assert_eq!(converted["contents"][0]["role"], "user");
assert_eq!(
converted["contents"][0]["parts"][0]["text"],
"Need CLI output"
);
}
}
@@ -0,0 +1,129 @@
//! Standard contract planning surface.
//!
//! This groups the public standard matrix in one place:
//! request-side conversion, matrix registry, and local standard execution entrypoints.
use axum::body::Body;
use axum::http::Response;
use crate::gateway::{
AppState, GatewayControlDecision, GatewayControlSyncDecisionResponse, GatewayError,
};
pub(crate) mod claude;
mod family;
pub(crate) mod gemini;
mod matrix;
pub(crate) mod openai;
pub(crate) use crate::gateway::ai_pipeline::conversion::{
build_core_error_body_for_client_format, request_conversion_kind,
request_conversion_transport_supported, sync_chat_response_conversion_kind,
sync_cli_response_conversion_kind, RequestConversionKind, SyncChatResponseConversionKind,
SyncCliResponseConversionKind,
};
pub(crate) use self::matrix::{
build_standard_request_body, build_standard_upstream_url,
normalize_standard_request_to_openai_chat_request,
};
pub(crate) use self::openai::{
copy_request_number_field, copy_request_number_field_as,
map_openai_reasoning_effort_to_claude_output, map_openai_reasoning_effort_to_gemini_budget,
maybe_build_stream_local_decision_payload, maybe_build_sync_local_decision_payload,
maybe_execute_stream_via_local_decision, maybe_execute_sync_via_local_decision,
parse_openai_stop_sequences, resolve_openai_chat_max_tokens, value_as_u64,
};
pub(crate) use self::openai::{
maybe_build_stream_local_openai_cli_decision_payload,
maybe_build_sync_local_openai_cli_decision_payload,
maybe_execute_stream_via_local_openai_cli_decision,
maybe_execute_sync_via_local_openai_cli_decision,
};
pub(crate) async fn maybe_execute_sync_via_local_standard_decision(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Option<Response<Body>>, GatewayError> {
if let Some(response) = self::claude::maybe_execute_sync_via_local_claude_decision(
state, parts, trace_id, decision, body_json, plan_kind,
)
.await?
{
return Ok(Some(response));
}
self::gemini::maybe_execute_sync_via_local_gemini_decision(
state, parts, trace_id, decision, body_json, plan_kind,
)
.await
}
pub(crate) async fn maybe_execute_stream_via_local_standard_decision(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Option<Response<Body>>, GatewayError> {
if let Some(response) = self::claude::maybe_execute_stream_via_local_claude_decision(
state, parts, trace_id, decision, body_json, plan_kind,
)
.await?
{
return Ok(Some(response));
}
self::gemini::maybe_execute_stream_via_local_gemini_decision(
state, parts, trace_id, decision, body_json, plan_kind,
)
.await
}
pub(crate) async fn maybe_build_sync_local_standard_decision_payload(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
if let Some(payload) = self::claude::maybe_build_sync_local_claude_decision_payload(
state, parts, trace_id, decision, body_json, plan_kind,
)
.await?
{
return Ok(Some(payload));
}
self::gemini::maybe_build_sync_local_gemini_decision_payload(
state, parts, trace_id, decision, body_json, plan_kind,
)
.await
}
pub(crate) async fn maybe_build_stream_local_standard_decision_payload(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
if let Some(payload) = self::claude::maybe_build_stream_local_claude_decision_payload(
state, parts, trace_id, decision, body_json, plan_kind,
)
.await?
{
return Ok(Some(payload));
}
self::gemini::maybe_build_stream_local_gemini_decision_payload(
state, parts, trace_id, decision, body_json, plan_kind,
)
.await
}
@@ -0,0 +1,321 @@
use axum::body::Body;
use axum::http::Response;
use serde_json::{Map, Value};
use tracing::warn;
use crate::gateway::{
execute_execution_runtime_stream, execute_execution_runtime_sync, AppState,
GatewayControlDecision, GatewayControlSyncDecisionResponse, GatewayError,
LocalExecutionRuntimeMissDiagnostic,
};
use crate::gateway::ai_pipeline::planner::{
OPENAI_CHAT_STREAM_PLAN_KIND, OPENAI_CHAT_SYNC_PLAN_KIND,
};
use crate::gateway::ai_pipeline::planner::standard::openai::{
convert_openai_chat_request_to_claude_request, convert_openai_chat_request_to_gemini_request,
convert_openai_chat_request_to_openai_cli_request, extract_openai_text_content,
parse_openai_tool_result_content,
};
mod decision;
mod plans;
use self::decision::{
mark_unused_local_openai_chat_candidates, materialize_local_openai_chat_candidate_attempts,
maybe_build_local_openai_chat_decision_payload_for_candidate, LocalOpenAiChatDecisionInput,
};
use self::plans::{
build_local_openai_chat_miss_diagnostic, build_local_openai_chat_stream_plan_and_reports,
build_local_openai_chat_sync_plan_and_reports, current_unix_secs,
list_local_openai_chat_candidates, resolve_local_openai_chat_decision_input,
set_local_openai_chat_miss_diagnostic,
};
pub(crate) async fn maybe_execute_sync_via_local_decision(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Option<Response<Body>>, GatewayError> {
let plan_and_reports = build_local_openai_chat_sync_plan_and_reports(
state, parts, trace_id, decision, body_json, plan_kind,
)
.await?;
if plan_and_reports.is_empty() {
return Ok(None);
}
let plan_count = plan_and_reports.len();
let mut remaining = plan_and_reports.into_iter();
while let Some(plan_and_report) = remaining.next() {
if let Some(response) = execute_execution_runtime_sync(
state,
parts.uri.path(),
plan_and_report.plan,
trace_id,
decision,
plan_kind,
plan_and_report.report_kind,
plan_and_report.report_context,
)
.await?
{
mark_unused_local_openai_chat_candidates(state, remaining.collect()).await;
return Ok(Some(response));
}
}
state.set_local_execution_runtime_miss_diagnostic(
trace_id,
LocalExecutionRuntimeMissDiagnostic {
candidate_count: Some(plan_count),
..build_local_openai_chat_miss_diagnostic(
decision,
plan_kind,
body_json.get("model").and_then(|value| value.as_str()),
"execution_runtime_candidates_exhausted",
)
},
);
Ok(None)
}
pub(crate) async fn maybe_execute_stream_via_local_decision(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Option<Response<Body>>, GatewayError> {
let plan_and_reports = build_local_openai_chat_stream_plan_and_reports(
state, parts, trace_id, decision, body_json, plan_kind,
)
.await?;
if plan_and_reports.is_empty() {
return Ok(None);
}
let plan_count = plan_and_reports.len();
let mut remaining = plan_and_reports.into_iter();
while let Some(plan_and_report) = remaining.next() {
if let Some(response) = execute_execution_runtime_stream(
state,
plan_and_report.plan,
trace_id,
decision,
plan_kind,
plan_and_report.report_kind,
plan_and_report.report_context,
)
.await?
{
mark_unused_local_openai_chat_candidates(state, remaining.collect()).await;
return Ok(Some(response));
}
}
state.set_local_execution_runtime_miss_diagnostic(
trace_id,
LocalExecutionRuntimeMissDiagnostic {
candidate_count: Some(plan_count),
..build_local_openai_chat_miss_diagnostic(
decision,
plan_kind,
body_json.get("model").and_then(|value| value.as_str()),
"execution_runtime_candidates_exhausted",
)
},
);
Ok(None)
}
pub(crate) async fn maybe_build_sync_local_decision_payload(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
if plan_kind != OPENAI_CHAT_SYNC_PLAN_KIND {
return Ok(None);
}
let Some(input) = resolve_local_openai_chat_decision_input(
state, trace_id, decision, body_json, plan_kind, false,
)
.await
else {
return Ok(None);
};
let candidates = match list_local_openai_chat_candidates(state, &input, false).await {
Ok(candidates) => candidates,
Err(err) => {
warn!(
trace_id = %trace_id,
error = ?err,
"gateway local openai chat sync decision scheduler selection failed"
);
return Ok(None);
}
};
let attempts =
materialize_local_openai_chat_candidate_attempts(state, trace_id, &input, candidates).await;
for attempt in attempts {
if let Some(payload) = maybe_build_local_openai_chat_decision_payload_for_candidate(
state,
parts,
trace_id,
body_json,
&input,
attempt,
OPENAI_CHAT_SYNC_PLAN_KIND,
"openai_chat_sync_success",
false,
)
.await
{
return Ok(Some(payload));
}
}
Ok(None)
}
pub(crate) async fn maybe_build_stream_local_decision_payload(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
if plan_kind != OPENAI_CHAT_STREAM_PLAN_KIND {
return Ok(None);
}
let Some(input) = resolve_local_openai_chat_decision_input(
state, trace_id, decision, body_json, plan_kind, false,
)
.await
else {
return Ok(None);
};
let candidates = match list_local_openai_chat_candidates(state, &input, true).await {
Ok(candidates) => candidates,
Err(err) => {
warn!(
trace_id = %trace_id,
error = ?err,
"gateway local openai chat stream decision scheduler selection failed"
);
return Ok(None);
}
};
let attempts =
materialize_local_openai_chat_candidate_attempts(state, trace_id, &input, candidates).await;
for attempt in attempts {
if let Some(payload) = maybe_build_local_openai_chat_decision_payload_for_candidate(
state,
parts,
trace_id,
body_json,
&input,
attempt,
OPENAI_CHAT_STREAM_PLAN_KIND,
"openai_chat_stream_success",
true,
)
.await
{
return Ok(Some(payload));
}
}
Ok(None)
}
pub(crate) fn parse_openai_stop_sequences(stop: Option<&Value>) -> Option<Vec<Value>> {
match stop {
Some(Value::String(value)) if !value.trim().is_empty() => {
Some(vec![Value::String(value.clone())])
}
Some(Value::Array(values)) => Some(
values
.iter()
.filter_map(|value| value.as_str())
.map(str::trim)
.filter(|value| !value.is_empty())
.map(|value| Value::String(value.to_string()))
.collect::<Vec<_>>(),
)
.filter(|values| !values.is_empty()),
_ => None,
}
}
pub(crate) fn resolve_openai_chat_max_tokens(request: &Map<String, Value>) -> u64 {
request
.get("max_completion_tokens")
.and_then(value_as_u64)
.or_else(|| request.get("max_tokens").and_then(value_as_u64))
.unwrap_or(4096)
}
pub(crate) fn value_as_u64(value: &Value) -> Option<u64> {
value
.as_u64()
.or_else(|| value.as_i64().and_then(|value| u64::try_from(value).ok()))
}
pub(crate) fn copy_request_number_field(
request: &Map<String, Value>,
target: &mut Map<String, Value>,
key: &str,
) {
copy_request_number_field_as(request, target, key, key);
}
pub(crate) fn copy_request_number_field_as(
request: &Map<String, Value>,
target: &mut Map<String, Value>,
source_key: &str,
target_key: &str,
) {
if let Some(value) = request.get(source_key).cloned() {
if value.is_number() {
target.insert(target_key.to_string(), value);
}
}
}
pub(crate) fn map_openai_reasoning_effort_to_claude_output(value: &str) -> Option<&'static str> {
match value.trim().to_ascii_lowercase().as_str() {
"low" => Some("low"),
"medium" => Some("medium"),
"high" | "xhigh" => Some("high"),
_ => None,
}
}
pub(crate) fn map_openai_reasoning_effort_to_gemini_budget(value: &str) -> Option<u64> {
match value.trim().to_ascii_lowercase().as_str() {
"low" => Some(1024),
"medium" => Some(4096),
"high" => Some(8192),
"xhigh" => Some(16_384),
_ => None,
}
}
@@ -0,0 +1,820 @@
use std::collections::BTreeMap;
use aether_data::repository::candidates::{RequestCandidateStatus, UpsertRequestCandidateRecord};
use serde_json::json;
use tracing::warn;
use uuid::Uuid;
use crate::gateway::ai_pipeline::conversion::{
request_conversion_direct_auth, request_conversion_kind, request_conversion_transport_supported,
};
use crate::gateway::headers::collect_control_headers;
use crate::gateway::provider_transport::{
apply_local_header_rules, build_openai_passthrough_headers, ensure_upstream_auth_header,
resolve_local_openai_chat_auth, resolve_transport_execution_timeouts,
resolve_transport_proxy_snapshot_with_tunnel_affinity, resolve_transport_tls_profile,
supports_local_openai_chat_transport, LocalResolvedOAuthRequestAuth,
};
use crate::gateway::request_candidates::record_local_request_candidate_status;
use crate::gateway::ai_pipeline::planner::plan_builders::{
LocalStreamPlanAndReport, LocalSyncPlanAndReport,
};
use crate::gateway::ai_pipeline::planner::prefer_local_tunnel_owner_candidates;
use crate::gateway::scheduler::GatewayMinimalCandidateSelectionCandidate;
use crate::gateway::{
append_execution_contract_fields_to_value, AppState, ConversionMode, ExecutionStrategy,
GatewayControlSyncDecisionResponse,
};
use crate::gateway::ai_pipeline::planner::{
EXECUTION_RUNTIME_STREAM_DECISION_ACTION, EXECUTION_RUNTIME_SYNC_DECISION_ACTION,
OPENAI_CHAT_STREAM_PLAN_KIND,
};
use super::plans::current_unix_secs;
use crate::gateway::ai_pipeline::planner::standard::openai::{
build_cross_format_openai_chat_request_body, build_cross_format_openai_chat_upstream_url,
build_local_openai_chat_request_body, build_local_openai_chat_upstream_url,
};
#[derive(Debug, Clone)]
pub(super) struct LocalOpenAiChatDecisionInput {
pub(super) auth_context: crate::gateway::GatewayControlAuthContext,
pub(super) requested_model: String,
pub(super) auth_snapshot: crate::gateway::gateway_data::StoredGatewayAuthApiKeySnapshot,
}
#[derive(Debug, Clone)]
pub(super) struct LocalOpenAiChatCandidateAttempt {
pub(super) candidate: GatewayMinimalCandidateSelectionCandidate,
pub(super) candidate_index: u32,
pub(super) candidate_id: String,
}
pub(super) async fn maybe_build_local_openai_chat_decision_payload_for_candidate(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
body_json: &serde_json::Value,
input: &LocalOpenAiChatDecisionInput,
attempt: LocalOpenAiChatCandidateAttempt,
decision_kind: &str,
report_kind: &str,
upstream_is_stream: bool,
) -> Option<GatewayControlSyncDecisionResponse> {
let LocalOpenAiChatCandidateAttempt {
candidate,
candidate_index,
candidate_id,
} = attempt;
let transport = match state
.read_provider_transport_snapshot(
&candidate.provider_id,
&candidate.endpoint_id,
&candidate.key_id,
)
.await
{
Ok(Some(snapshot)) => snapshot,
Ok(None) => {
mark_skipped_local_openai_chat_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_snapshot_missing",
)
.await;
return None;
}
Err(err) => {
warn!(
trace_id = %trace_id,
error = ?err,
"gateway local openai chat decision provider transport read failed"
);
mark_skipped_local_openai_chat_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_snapshot_read_failed",
)
.await;
return None;
}
};
let provider_api_format = transport.endpoint.api_format.trim().to_ascii_lowercase();
match provider_api_format.as_str() {
"openai:chat" => {
build_same_format_local_openai_chat_decision_payload_for_candidate(
state,
parts,
trace_id,
body_json,
input,
&candidate,
candidate_index,
&candidate_id,
decision_kind,
report_kind,
upstream_is_stream,
&transport,
)
.await
}
"claude:chat" | "gemini:chat" | "openai:cli" | "openai:compact" => {
build_cross_format_local_openai_chat_decision_payload_for_candidate(
state,
parts,
trace_id,
body_json,
input,
&candidate,
candidate_index,
&candidate_id,
decision_kind,
upstream_is_stream,
&transport,
provider_api_format.as_str(),
)
.await
}
_ => {
mark_skipped_local_openai_chat_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_unsupported",
)
.await;
None
}
}
}
#[allow(clippy::too_many_arguments)]
async fn build_same_format_local_openai_chat_decision_payload_for_candidate(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
body_json: &serde_json::Value,
input: &LocalOpenAiChatDecisionInput,
candidate: &GatewayMinimalCandidateSelectionCandidate,
candidate_index: u32,
candidate_id: &str,
decision_kind: &str,
report_kind: &str,
upstream_is_stream: bool,
transport: &crate::gateway::provider_transport::GatewayProviderTransportSnapshot,
) -> Option<GatewayControlSyncDecisionResponse> {
if !supports_local_openai_chat_transport(transport) {
mark_skipped_local_openai_chat_candidate(
state,
input,
trace_id,
candidate,
candidate_index,
candidate_id,
"transport_unsupported",
)
.await;
return None;
}
let oauth_auth = if resolve_local_openai_chat_auth(transport).is_none() {
match state.resolve_local_oauth_request_auth(transport).await {
Ok(Some(LocalResolvedOAuthRequestAuth::Header { name, value })) => Some((name, value)),
Ok(Some(LocalResolvedOAuthRequestAuth::Kiro(_))) => None,
Ok(None) => None,
Err(err) => {
warn!(
trace_id = %trace_id,
provider_type = %transport.provider.provider_type,
error = ?err,
"gateway local openai chat oauth auth resolution failed"
);
None
}
}
} else {
None
};
let Some((auth_header, auth_value)) = resolve_local_openai_chat_auth(transport).or(oauth_auth)
else {
mark_skipped_local_openai_chat_candidate(
state,
input,
trace_id,
candidate,
candidate_index,
candidate_id,
"transport_auth_unavailable",
)
.await;
return None;
};
let mapped_model = candidate.selected_provider_model_name.trim().to_string();
if mapped_model.is_empty() {
mark_skipped_local_openai_chat_candidate(
state,
input,
trace_id,
candidate,
candidate_index,
candidate_id,
"mapped_model_missing",
)
.await;
return None;
}
let Some(provider_request_body) = build_local_openai_chat_request_body(
body_json,
&mapped_model,
upstream_is_stream,
transport.endpoint.body_rules.as_ref(),
) else {
mark_skipped_local_openai_chat_candidate(
state,
input,
trace_id,
candidate,
candidate_index,
candidate_id,
"provider_request_body_missing",
)
.await;
return None;
};
let Some(upstream_url) = build_local_openai_chat_upstream_url(parts, transport) else {
mark_skipped_local_openai_chat_candidate(
state,
input,
trace_id,
candidate,
candidate_index,
candidate_id,
"upstream_url_missing",
)
.await;
return None;
};
let mut provider_request_headers = build_openai_passthrough_headers(
&parts.headers,
&auth_header,
&auth_value,
&BTreeMap::new(),
Some("application/json"),
);
if !apply_local_header_rules(
&mut provider_request_headers,
transport.endpoint.header_rules.as_ref(),
&[&auth_header, "content-type"],
&provider_request_body,
Some(body_json),
) {
mark_skipped_local_openai_chat_candidate(
state,
input,
trace_id,
candidate,
candidate_index,
candidate_id,
"transport_header_rules_apply_failed",
)
.await;
return None;
}
ensure_upstream_auth_header(&mut provider_request_headers, &auth_header, &auth_value);
if upstream_is_stream {
provider_request_headers
.entry("accept".to_string())
.or_insert_with(|| "text/event-stream".to_string());
}
let proxy = resolve_transport_proxy_snapshot_with_tunnel_affinity(state, transport).await;
let tls_profile = resolve_transport_tls_profile(transport);
let prompt_cache_key = provider_request_body
.get("prompt_cache_key")
.and_then(|value| value.as_str())
.map(str::trim)
.filter(|value| !value.is_empty())
.map(ToOwned::to_owned);
Some(GatewayControlSyncDecisionResponse {
action: if upstream_is_stream {
EXECUTION_RUNTIME_STREAM_DECISION_ACTION.to_string()
} else {
EXECUTION_RUNTIME_SYNC_DECISION_ACTION.to_string()
},
decision_kind: Some(decision_kind.to_string()),
execution_strategy: Some(ExecutionStrategy::LocalSameFormat.as_str().to_string()),
conversion_mode: Some(ConversionMode::None.as_str().to_string()),
request_id: Some(trace_id.to_string()),
candidate_id: Some(candidate_id.to_string()),
provider_name: Some(transport.provider.name.clone()),
provider_id: Some(candidate.provider_id.clone()),
endpoint_id: Some(candidate.endpoint_id.clone()),
key_id: Some(candidate.key_id.clone()),
upstream_base_url: Some(transport.endpoint.base_url.clone()),
upstream_url: Some(upstream_url.clone()),
provider_request_method: None,
auth_header: Some(auth_header),
auth_value: Some(auth_value),
provider_api_format: Some("openai:chat".to_string()),
client_api_format: Some("openai:chat".to_string()),
provider_contract: Some("openai:chat".to_string()),
client_contract: Some("openai:chat".to_string()),
model_name: Some(input.requested_model.clone()),
mapped_model: Some(mapped_model.clone()),
prompt_cache_key,
extra_headers: BTreeMap::new(),
provider_request_headers: provider_request_headers.clone(),
provider_request_body: Some(provider_request_body.clone()),
provider_request_body_base64: None,
content_type: Some("application/json".to_string()),
proxy,
tls_profile,
timeouts: resolve_transport_execution_timeouts(transport),
upstream_is_stream,
report_kind: Some(report_kind.to_string()),
report_context: Some(append_execution_contract_fields_to_value(
json!({
"user_id": input.auth_context.user_id,
"api_key_id": input.auth_context.api_key_id,
"request_id": trace_id,
"candidate_id": candidate_id,
"candidate_index": candidate_index,
"retry_index": 0,
"model": input.requested_model,
"provider_name": transport.provider.name,
"provider_id": candidate.provider_id,
"endpoint_id": candidate.endpoint_id,
"key_id": candidate.key_id,
"provider_api_format": "openai:chat",
"client_api_format": "openai:chat",
"mapped_model": mapped_model,
"upstream_url": upstream_url,
"provider_request_method": serde_json::Value::Null,
"provider_request_headers": provider_request_headers,
"provider_request_body": provider_request_body,
"original_headers": collect_control_headers(&parts.headers),
"original_request_body": body_json,
"has_envelope": false,
"needs_conversion": false,
}),
ExecutionStrategy::LocalSameFormat,
ConversionMode::None,
"openai:chat",
"openai:chat",
)),
auth_context: Some(input.auth_context.clone()),
})
}
#[allow(clippy::too_many_arguments)]
async fn build_cross_format_local_openai_chat_decision_payload_for_candidate(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
body_json: &serde_json::Value,
input: &LocalOpenAiChatDecisionInput,
candidate: &GatewayMinimalCandidateSelectionCandidate,
candidate_index: u32,
candidate_id: &str,
decision_kind: &str,
upstream_is_stream: bool,
transport: &crate::gateway::provider_transport::GatewayProviderTransportSnapshot,
provider_api_format: &str,
) -> Option<GatewayControlSyncDecisionResponse> {
let provider_api_format = provider_api_format.trim().to_ascii_lowercase();
let Some(conversion_kind) =
request_conversion_kind("openai:chat", provider_api_format.as_str())
else {
return None;
};
let transport_supported = request_conversion_transport_supported(transport, conversion_kind);
if !transport_supported {
mark_skipped_local_openai_chat_candidate(
state,
input,
trace_id,
candidate,
candidate_index,
candidate_id,
"transport_unsupported",
)
.await;
return None;
}
let resolve_auth = request_conversion_direct_auth(transport, conversion_kind);
let oauth_auth = if resolve_auth.is_none() {
match state.resolve_local_oauth_request_auth(transport).await {
Ok(Some(LocalResolvedOAuthRequestAuth::Header { name, value })) => Some((name, value)),
Ok(Some(LocalResolvedOAuthRequestAuth::Kiro(_))) => None,
Ok(None) => None,
Err(err) => {
warn!(
trace_id = %trace_id,
provider_type = %transport.provider.provider_type,
provider_api_format = %provider_api_format,
error = ?err,
"gateway local openai chat cross-format oauth auth resolution failed"
);
None
}
}
} else {
None
};
let Some((auth_header, auth_value)) = resolve_auth.or(oauth_auth) else {
mark_skipped_local_openai_chat_candidate(
state,
input,
trace_id,
candidate,
candidate_index,
candidate_id,
"transport_auth_unavailable",
)
.await;
return None;
};
let mapped_model = candidate.selected_provider_model_name.trim().to_string();
if mapped_model.is_empty() {
mark_skipped_local_openai_chat_candidate(
state,
input,
trace_id,
candidate,
candidate_index,
candidate_id,
"mapped_model_missing",
)
.await;
return None;
}
let Some(provider_request_body) = build_cross_format_openai_chat_request_body(
body_json,
&mapped_model,
provider_api_format.as_str(),
upstream_is_stream,
transport.endpoint.body_rules.as_ref(),
) else {
mark_skipped_local_openai_chat_candidate(
state,
input,
trace_id,
candidate,
candidate_index,
candidate_id,
"provider_request_body_missing",
)
.await;
return None;
};
let Some(upstream_url) = build_cross_format_openai_chat_upstream_url(
parts,
transport,
&mapped_model,
provider_api_format.as_str(),
upstream_is_stream,
) else {
mark_skipped_local_openai_chat_candidate(
state,
input,
trace_id,
candidate,
candidate_index,
candidate_id,
"upstream_url_missing",
)
.await;
return None;
};
let mut provider_request_headers = build_openai_passthrough_headers(
&parts.headers,
&auth_header,
&auth_value,
&BTreeMap::new(),
Some("application/json"),
);
if !apply_local_header_rules(
&mut provider_request_headers,
transport.endpoint.header_rules.as_ref(),
&[&auth_header, "content-type"],
&provider_request_body,
Some(body_json),
) {
mark_skipped_local_openai_chat_candidate(
state,
input,
trace_id,
candidate,
candidate_index,
candidate_id,
"transport_header_rules_apply_failed",
)
.await;
return None;
}
ensure_upstream_auth_header(&mut provider_request_headers, &auth_header, &auth_value);
if upstream_is_stream {
provider_request_headers
.entry("accept".to_string())
.or_insert_with(|| "text/event-stream".to_string());
}
let report_kind = if decision_kind == OPENAI_CHAT_STREAM_PLAN_KIND {
"openai_chat_stream_success"
} else {
"openai_chat_sync_finalize"
};
let proxy = resolve_transport_proxy_snapshot_with_tunnel_affinity(state, transport).await;
let tls_profile = resolve_transport_tls_profile(transport);
let prompt_cache_key = provider_request_body
.get("prompt_cache_key")
.and_then(|value| value.as_str())
.map(str::trim)
.filter(|value| !value.is_empty())
.map(ToOwned::to_owned);
Some(GatewayControlSyncDecisionResponse {
action: if upstream_is_stream {
EXECUTION_RUNTIME_STREAM_DECISION_ACTION.to_string()
} else {
EXECUTION_RUNTIME_SYNC_DECISION_ACTION.to_string()
},
decision_kind: Some(decision_kind.to_string()),
execution_strategy: Some(ExecutionStrategy::LocalCrossFormat.as_str().to_string()),
conversion_mode: Some(ConversionMode::Bidirectional.as_str().to_string()),
request_id: Some(trace_id.to_string()),
candidate_id: Some(candidate_id.to_string()),
provider_name: Some(transport.provider.name.clone()),
provider_id: Some(candidate.provider_id.clone()),
endpoint_id: Some(candidate.endpoint_id.clone()),
key_id: Some(candidate.key_id.clone()),
upstream_base_url: Some(transport.endpoint.base_url.clone()),
upstream_url: Some(upstream_url.clone()),
provider_request_method: None,
auth_header: Some(auth_header),
auth_value: Some(auth_value),
provider_api_format: Some(provider_api_format.clone()),
client_api_format: Some("openai:chat".to_string()),
provider_contract: Some(provider_api_format.clone()),
client_contract: Some("openai:chat".to_string()),
model_name: Some(input.requested_model.clone()),
mapped_model: Some(mapped_model.clone()),
prompt_cache_key,
extra_headers: BTreeMap::new(),
provider_request_headers: provider_request_headers.clone(),
provider_request_body: Some(provider_request_body.clone()),
provider_request_body_base64: None,
content_type: Some("application/json".to_string()),
proxy,
tls_profile,
timeouts: resolve_transport_execution_timeouts(transport),
upstream_is_stream,
report_kind: Some(report_kind.to_string()),
report_context: Some(append_execution_contract_fields_to_value(
json!({
"user_id": input.auth_context.user_id,
"api_key_id": input.auth_context.api_key_id,
"request_id": trace_id,
"candidate_id": candidate_id,
"candidate_index": candidate_index,
"retry_index": 0,
"model": input.requested_model,
"provider_name": transport.provider.name,
"provider_id": candidate.provider_id,
"endpoint_id": candidate.endpoint_id,
"key_id": candidate.key_id,
"provider_api_format": provider_api_format,
"client_api_format": "openai:chat",
"mapped_model": mapped_model,
"upstream_url": upstream_url,
"provider_request_method": serde_json::Value::Null,
"provider_request_headers": provider_request_headers,
"provider_request_body": provider_request_body,
"original_headers": collect_control_headers(&parts.headers),
"original_request_body": body_json,
"has_envelope": false,
"needs_conversion": true,
}),
ExecutionStrategy::LocalCrossFormat,
ConversionMode::Bidirectional,
"openai:chat",
provider_api_format.as_str(),
)),
auth_context: Some(input.auth_context.clone()),
})
}
pub(super) async fn mark_skipped_local_openai_chat_candidate(
state: &AppState,
input: &LocalOpenAiChatDecisionInput,
trace_id: &str,
candidate: &GatewayMinimalCandidateSelectionCandidate,
candidate_index: u32,
candidate_id: &str,
skip_reason: &'static str,
) {
state.mutate_local_execution_runtime_miss_diagnostic(trace_id, |diagnostic| {
*diagnostic
.skip_reasons
.entry(skip_reason.to_string())
.or_insert(0) += 1;
*diagnostic.skipped_candidate_count.get_or_insert(0) += 1;
});
let terminal_unix_secs = current_unix_secs();
if let Err(err) = state
.upsert_request_candidate(UpsertRequestCandidateRecord {
id: candidate_id.to_string(),
request_id: trace_id.to_string(),
user_id: Some(input.auth_context.user_id.clone()),
api_key_id: Some(input.auth_context.api_key_id.clone()),
username: None,
api_key_name: None,
candidate_index,
retry_index: 0,
provider_id: Some(candidate.provider_id.clone()),
endpoint_id: Some(candidate.endpoint_id.clone()),
key_id: Some(candidate.key_id.clone()),
status: RequestCandidateStatus::Skipped,
skip_reason: Some(skip_reason.to_string()),
is_cached: Some(false),
status_code: None,
error_type: None,
error_message: None,
latency_ms: None,
concurrent_requests: None,
extra_data: None,
required_capabilities: candidate.key_capabilities.clone(),
created_at_unix_secs: None,
started_at_unix_secs: None,
finished_at_unix_secs: Some(terminal_unix_secs),
})
.await
{
warn!(
trace_id = %trace_id,
candidate_id = %candidate_id,
skip_reason,
error = ?err,
"gateway local openai chat decision failed to persist skipped candidate"
);
}
}
pub(super) async fn materialize_local_openai_chat_candidate_attempts(
state: &AppState,
trace_id: &str,
input: &LocalOpenAiChatDecisionInput,
candidates: Vec<GatewayMinimalCandidateSelectionCandidate>,
) -> Vec<LocalOpenAiChatCandidateAttempt> {
let candidates = prefer_local_tunnel_owner_candidates(state, candidates).await;
let created_at_unix_secs = current_unix_secs();
let mut attempts = Vec::with_capacity(candidates.len());
for (candidate_index, candidate) in candidates.into_iter().enumerate() {
let generated_candidate_id = Uuid::new_v4().to_string();
let provider_api_format = candidate.endpoint_api_format.trim().to_ascii_lowercase();
let (execution_strategy, conversion_mode) = if provider_api_format == "openai:chat" {
(ExecutionStrategy::LocalSameFormat, ConversionMode::None)
} else {
(
ExecutionStrategy::LocalCrossFormat,
ConversionMode::Bidirectional,
)
};
let extra_data = append_execution_contract_fields_to_value(
json!({
"provider_api_format": provider_api_format,
"client_api_format": "openai:chat",
"global_model_id": candidate.global_model_id.clone(),
"global_model_name": candidate.global_model_name.clone(),
"model_id": candidate.model_id.clone(),
"selected_provider_model_name": candidate.selected_provider_model_name.clone(),
"mapping_matched_model": candidate.mapping_matched_model.clone(),
"provider_name": candidate.provider_name.clone(),
"key_name": candidate.key_name.clone(),
}),
execution_strategy,
conversion_mode,
"openai:chat",
candidate.endpoint_api_format.trim(),
);
let candidate_id = match state
.upsert_request_candidate(UpsertRequestCandidateRecord {
id: generated_candidate_id.clone(),
request_id: trace_id.to_string(),
user_id: Some(input.auth_context.user_id.clone()),
api_key_id: Some(input.auth_context.api_key_id.clone()),
username: None,
api_key_name: None,
candidate_index: candidate_index as u32,
retry_index: 0,
provider_id: Some(candidate.provider_id.clone()),
endpoint_id: Some(candidate.endpoint_id.clone()),
key_id: Some(candidate.key_id.clone()),
status: RequestCandidateStatus::Available,
skip_reason: None,
is_cached: Some(false),
status_code: None,
error_type: None,
error_message: None,
latency_ms: None,
concurrent_requests: None,
extra_data: Some(extra_data),
required_capabilities: candidate.key_capabilities.clone(),
created_at_unix_secs: Some(created_at_unix_secs),
started_at_unix_secs: None,
finished_at_unix_secs: None,
})
.await
{
Ok(Some(stored)) => stored.id,
Ok(None) => generated_candidate_id.clone(),
Err(err) => {
warn!(
trace_id = %trace_id,
error = ?err,
"gateway local openai chat decision request candidate upsert failed"
);
generated_candidate_id.clone()
}
};
attempts.push(LocalOpenAiChatCandidateAttempt {
candidate,
candidate_index: candidate_index as u32,
candidate_id,
});
}
attempts
}
pub(super) async fn mark_unused_local_openai_chat_candidates<T>(state: &AppState, remaining: Vec<T>)
where
T: LocalOpenAiChatPlanAndReport,
{
for plan_and_report in remaining {
record_local_request_candidate_status(
state,
plan_and_report.plan(),
plan_and_report.report_context(),
RequestCandidateStatus::Unused,
None,
None,
None,
None,
None,
None,
)
.await;
}
}
pub(super) trait LocalOpenAiChatPlanAndReport {
fn plan(&self) -> &aether_contracts::ExecutionPlan;
fn report_context(&self) -> Option<&serde_json::Value>;
}
impl LocalOpenAiChatPlanAndReport for LocalSyncPlanAndReport {
fn plan(&self) -> &aether_contracts::ExecutionPlan {
&self.plan
}
fn report_context(&self) -> Option<&serde_json::Value> {
self.report_context.as_ref()
}
}
impl LocalOpenAiChatPlanAndReport for LocalStreamPlanAndReport {
fn plan(&self) -> &aether_contracts::ExecutionPlan {
&self.plan
}
fn report_context(&self) -> Option<&serde_json::Value> {
self.report_context.as_ref()
}
}
@@ -0,0 +1,494 @@
use std::collections::{BTreeMap, BTreeSet};
use std::time::{SystemTime, UNIX_EPOCH};
use tracing::warn;
use super::{
materialize_local_openai_chat_candidate_attempts,
maybe_build_local_openai_chat_decision_payload_for_candidate, AppState, GatewayControlDecision,
GatewayError, LocalExecutionRuntimeMissDiagnostic, LocalOpenAiChatDecisionInput,
};
use crate::gateway::ai_pipeline::planner::plan_builders::{
build_openai_chat_stream_plan_from_decision, build_openai_chat_sync_plan_from_decision,
LocalStreamPlanAndReport, LocalSyncPlanAndReport,
};
use crate::gateway::ai_pipeline::planner::prefer_local_tunnel_owner_candidates;
use crate::gateway::scheduler::{
list_selectable_candidates, GatewayMinimalCandidateSelectionCandidate,
};
use crate::gateway::ai_pipeline::planner::{
OPENAI_CHAT_STREAM_PLAN_KIND, OPENAI_CHAT_SYNC_PLAN_KIND,
};
pub(super) fn build_local_openai_chat_miss_diagnostic(
decision: &GatewayControlDecision,
plan_kind: &str,
requested_model: Option<&str>,
reason: &str,
) -> LocalExecutionRuntimeMissDiagnostic {
LocalExecutionRuntimeMissDiagnostic {
reason: reason.to_string(),
route_family: decision.route_family.clone(),
route_kind: decision.route_kind.clone(),
public_path: Some(decision.public_path.clone()),
plan_kind: Some(plan_kind.to_string()),
requested_model: requested_model.map(ToOwned::to_owned),
candidate_count: None,
skipped_candidate_count: None,
skip_reasons: BTreeMap::new(),
}
}
pub(super) fn set_local_openai_chat_miss_diagnostic(
state: &AppState,
trace_id: &str,
decision: &GatewayControlDecision,
plan_kind: &str,
requested_model: Option<&str>,
reason: &str,
) {
state.set_local_execution_runtime_miss_diagnostic(
trace_id,
build_local_openai_chat_miss_diagnostic(decision, plan_kind, requested_model, reason),
);
}
pub(super) async fn build_local_openai_chat_sync_plan_and_reports(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Vec<LocalSyncPlanAndReport>, GatewayError> {
if plan_kind != OPENAI_CHAT_SYNC_PLAN_KIND {
return Ok(Vec::new());
}
let Some(input) = resolve_local_openai_chat_decision_input(
state, trace_id, decision, body_json, plan_kind, true,
)
.await
else {
return Ok(Vec::new());
};
let candidates = match list_local_openai_chat_candidates(state, &input, false).await {
Ok(candidates) => candidates,
Err(err) => {
warn!(
trace_id = %trace_id,
error = ?err,
"gateway local openai chat sync decision scheduler selection failed"
);
set_local_openai_chat_miss_diagnostic(
state,
trace_id,
decision,
plan_kind,
Some(input.requested_model.as_str()),
"scheduler_selection_failed",
);
return Ok(Vec::new());
}
};
if candidates.is_empty() {
state.set_local_execution_runtime_miss_diagnostic(
trace_id,
LocalExecutionRuntimeMissDiagnostic {
candidate_count: Some(0),
..build_local_openai_chat_miss_diagnostic(
decision,
plan_kind,
Some(input.requested_model.as_str()),
"candidate_list_empty",
)
},
);
return Ok(Vec::new());
}
state.set_local_execution_runtime_miss_diagnostic(
trace_id,
LocalExecutionRuntimeMissDiagnostic {
candidate_count: Some(candidates.len()),
..build_local_openai_chat_miss_diagnostic(
decision,
plan_kind,
Some(input.requested_model.as_str()),
"candidate_evaluation_incomplete",
)
},
);
let attempts =
materialize_local_openai_chat_candidate_attempts(state, trace_id, &input, candidates).await;
let mut plans = Vec::new();
for attempt in attempts {
let Some(payload) = maybe_build_local_openai_chat_decision_payload_for_candidate(
state,
parts,
trace_id,
body_json,
&input,
attempt,
OPENAI_CHAT_SYNC_PLAN_KIND,
"openai_chat_sync_success",
false,
)
.await
else {
continue;
};
match build_openai_chat_sync_plan_from_decision(parts, body_json, payload) {
Ok(Some(value)) => plans.push(value),
Ok(None) => {}
Err(err) => {
warn!(
trace_id = %trace_id,
error = ?err,
"gateway local openai chat sync decision plan build failed"
);
}
}
}
state.mutate_local_execution_runtime_miss_diagnostic(trace_id, |diagnostic| {
let candidate_count = diagnostic.candidate_count.unwrap_or(0);
let skipped_candidate_count = diagnostic.skipped_candidate_count.unwrap_or(0);
diagnostic.reason = if candidate_count > 0 && skipped_candidate_count >= candidate_count {
"all_candidates_skipped".to_string()
} else {
"no_local_sync_plans".to_string()
};
});
Ok(plans)
}
pub(super) async fn build_local_openai_chat_stream_plan_and_reports(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Vec<LocalStreamPlanAndReport>, GatewayError> {
if plan_kind != OPENAI_CHAT_STREAM_PLAN_KIND {
return Ok(Vec::new());
}
let Some(input) = resolve_local_openai_chat_decision_input(
state, trace_id, decision, body_json, plan_kind, true,
)
.await
else {
return Ok(Vec::new());
};
let candidates = match list_local_openai_chat_candidates(state, &input, true).await {
Ok(candidates) => candidates,
Err(err) => {
warn!(
trace_id = %trace_id,
error = ?err,
"gateway local openai chat stream decision scheduler selection failed"
);
set_local_openai_chat_miss_diagnostic(
state,
trace_id,
decision,
plan_kind,
Some(input.requested_model.as_str()),
"scheduler_selection_failed",
);
return Ok(Vec::new());
}
};
if candidates.is_empty() {
state.set_local_execution_runtime_miss_diagnostic(
trace_id,
LocalExecutionRuntimeMissDiagnostic {
candidate_count: Some(0),
..build_local_openai_chat_miss_diagnostic(
decision,
plan_kind,
Some(input.requested_model.as_str()),
"candidate_list_empty",
)
},
);
return Ok(Vec::new());
}
state.set_local_execution_runtime_miss_diagnostic(
trace_id,
LocalExecutionRuntimeMissDiagnostic {
candidate_count: Some(candidates.len()),
..build_local_openai_chat_miss_diagnostic(
decision,
plan_kind,
Some(input.requested_model.as_str()),
"candidate_evaluation_incomplete",
)
},
);
let attempts =
materialize_local_openai_chat_candidate_attempts(state, trace_id, &input, candidates).await;
let mut plans = Vec::new();
for attempt in attempts {
let Some(payload) = maybe_build_local_openai_chat_decision_payload_for_candidate(
state,
parts,
trace_id,
body_json,
&input,
attempt,
OPENAI_CHAT_STREAM_PLAN_KIND,
"openai_chat_stream_success",
true,
)
.await
else {
continue;
};
match build_openai_chat_stream_plan_from_decision(parts, body_json, payload) {
Ok(Some(value)) => plans.push(value),
Ok(None) => {}
Err(err) => {
warn!(
trace_id = %trace_id,
error = ?err,
"gateway local openai chat stream decision plan build failed"
);
}
}
}
state.mutate_local_execution_runtime_miss_diagnostic(trace_id, |diagnostic| {
let candidate_count = diagnostic.candidate_count.unwrap_or(0);
let skipped_candidate_count = diagnostic.skipped_candidate_count.unwrap_or(0);
diagnostic.reason = if candidate_count > 0 && skipped_candidate_count >= candidate_count {
"all_candidates_skipped".to_string()
} else {
"no_local_stream_plans".to_string()
};
});
Ok(plans)
}
pub(super) fn current_unix_secs() -> u64 {
SystemTime::now()
.duration_since(UNIX_EPOCH)
.unwrap_or_default()
.as_secs()
}
pub(super) async fn list_local_openai_chat_candidates(
state: &AppState,
input: &LocalOpenAiChatDecisionInput,
require_streaming: bool,
) -> Result<Vec<GatewayMinimalCandidateSelectionCandidate>, GatewayError> {
let now_unix_secs = current_unix_secs();
let mut combined = Vec::new();
let mut seen = BTreeSet::new();
let api_formats = if require_streaming {
vec!["openai:chat", "claude:chat", "gemini:chat", "openai:cli"]
} else {
vec![
"openai:chat",
"claude:chat",
"gemini:chat",
"openai:cli",
"openai:compact",
]
};
for api_format in api_formats {
let auth_snapshot = if api_format == "openai:chat" {
Some(&input.auth_snapshot)
} else {
None
};
let mut candidates = list_selectable_candidates(
state,
api_format,
&input.requested_model,
require_streaming,
auth_snapshot,
now_unix_secs,
)
.await?;
if api_format != "openai:chat" {
candidates.retain(|candidate| {
auth_snapshot_allows_cross_format_openai_chat_candidate(
&input.auth_snapshot,
&input.requested_model,
candidate,
)
});
}
for candidate in candidates {
let candidate_key = format!(
"{}:{}:{}:{}:{}",
candidate.provider_id,
candidate.endpoint_id,
candidate.key_id,
candidate.model_id,
candidate.selected_provider_model_name,
);
if seen.insert(candidate_key) {
combined.push(candidate);
}
}
}
Ok(combined)
}
fn auth_snapshot_allows_cross_format_openai_chat_candidate(
auth_snapshot: &crate::gateway::gateway_data::StoredGatewayAuthApiKeySnapshot,
requested_model: &str,
candidate: &GatewayMinimalCandidateSelectionCandidate,
) -> bool {
if let Some(allowed_providers) = auth_snapshot.effective_allowed_providers() {
let provider_allowed = allowed_providers.iter().any(|value| {
value
.trim()
.eq_ignore_ascii_case(candidate.provider_id.trim())
|| value
.trim()
.eq_ignore_ascii_case(candidate.provider_name.trim())
});
if !provider_allowed {
return false;
}
}
if let Some(allowed_models) = auth_snapshot.effective_allowed_models() {
let model_allowed = allowed_models
.iter()
.any(|value| value == requested_model || value == &candidate.global_model_name);
if !model_allowed {
return false;
}
}
true
}
pub(super) async fn resolve_local_openai_chat_decision_input(
state: &AppState,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
record_miss_diagnostic: bool,
) -> Option<LocalOpenAiChatDecisionInput> {
let Some(auth_context) = decision.auth_context.clone().filter(|auth_context| {
!auth_context.user_id.trim().is_empty() && !auth_context.api_key_id.trim().is_empty()
}) else {
warn!(
trace_id = %trace_id,
route_class = ?decision.route_class,
route_family = ?decision.route_family,
route_kind = ?decision.route_kind,
"gateway local openai chat decision skipped: missing_auth_context"
);
if record_miss_diagnostic {
set_local_openai_chat_miss_diagnostic(
state,
trace_id,
decision,
plan_kind,
body_json.get("model").and_then(|value| value.as_str()),
"missing_auth_context",
);
}
return None;
};
let Some(requested_model) = body_json
.get("model")
.and_then(|value| value.as_str())
.map(str::trim)
.filter(|value| !value.is_empty())
.map(ToOwned::to_owned)
else {
warn!(
trace_id = %trace_id,
"gateway local openai chat decision skipped: missing_requested_model"
);
if record_miss_diagnostic {
set_local_openai_chat_miss_diagnostic(
state,
trace_id,
decision,
plan_kind,
None,
"missing_requested_model",
);
}
return None;
};
let now_unix_secs = current_unix_secs();
let auth_snapshot = match state
.read_auth_api_key_snapshot(
&auth_context.user_id,
&auth_context.api_key_id,
now_unix_secs,
)
.await
{
Ok(Some(snapshot)) => snapshot,
Ok(None) => {
warn!(
trace_id = %trace_id,
user_id = %auth_context.user_id,
api_key_id = %auth_context.api_key_id,
"gateway local openai chat decision skipped: auth_snapshot_missing"
);
if record_miss_diagnostic {
set_local_openai_chat_miss_diagnostic(
state,
trace_id,
decision,
plan_kind,
Some(requested_model.as_str()),
"auth_snapshot_missing",
);
}
return None;
}
Err(err) => {
warn!(
trace_id = %trace_id,
error = ?err,
"gateway local openai chat decision auth snapshot read failed"
);
if record_miss_diagnostic {
set_local_openai_chat_miss_diagnostic(
state,
trace_id,
decision,
plan_kind,
Some(requested_model.as_str()),
"auth_snapshot_read_failed",
);
}
return None;
}
};
Some(LocalOpenAiChatDecisionInput {
auth_context,
requested_model,
auth_snapshot,
})
}
@@ -0,0 +1,168 @@
use crate::gateway::{
execute_execution_runtime_stream, execute_execution_runtime_sync, AppState,
GatewayControlDecision, GatewayControlSyncDecisionResponse, GatewayError,
};
use axum::body::Body;
use axum::http::Response;
mod decision;
mod plans;
use self::decision::{
mark_unused_local_openai_cli_candidates, materialize_local_openai_cli_candidate_attempts,
maybe_build_local_openai_cli_decision_payload_for_candidate,
resolve_local_openai_cli_decision_input,
};
use self::plans::{
build_local_stream_plan_and_reports, build_local_sync_plan_and_reports, resolve_stream_spec,
resolve_sync_spec,
};
pub(crate) async fn maybe_execute_sync_via_local_openai_cli_decision(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Option<Response<Body>>, GatewayError> {
let Some(spec) = resolve_sync_spec(plan_kind) else {
return Ok(None);
};
let plan_and_reports =
build_local_sync_plan_and_reports(state, parts, trace_id, decision, body_json, spec)
.await?;
if plan_and_reports.is_empty() {
return Ok(None);
}
let mut remaining = plan_and_reports.into_iter();
while let Some(plan_and_report) = remaining.next() {
if let Some(response) = execute_execution_runtime_sync(
state,
parts.uri.path(),
plan_and_report.plan,
trace_id,
decision,
plan_kind,
plan_and_report.report_kind,
plan_and_report.report_context,
)
.await?
{
mark_unused_local_openai_cli_candidates(state, remaining.collect()).await;
return Ok(Some(response));
}
}
Ok(None)
}
pub(crate) async fn maybe_execute_stream_via_local_openai_cli_decision(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Option<Response<Body>>, GatewayError> {
let Some(spec) = resolve_stream_spec(plan_kind) else {
return Ok(None);
};
let plan_and_reports =
build_local_stream_plan_and_reports(state, parts, trace_id, decision, body_json, spec)
.await?;
if plan_and_reports.is_empty() {
return Ok(None);
}
let mut remaining = plan_and_reports.into_iter();
while let Some(plan_and_report) = remaining.next() {
if let Some(response) = execute_execution_runtime_stream(
state,
plan_and_report.plan,
trace_id,
decision,
plan_kind,
plan_and_report.report_kind,
plan_and_report.report_context,
)
.await?
{
mark_unused_local_openai_cli_candidates(state, remaining.collect()).await;
return Ok(Some(response));
}
}
Ok(None)
}
pub(crate) async fn maybe_build_sync_local_openai_cli_decision_payload(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
let Some(spec) = resolve_sync_spec(plan_kind) else {
return Ok(None);
};
let Some(input) =
resolve_local_openai_cli_decision_input(state, trace_id, decision, body_json).await
else {
return Ok(None);
};
let attempts =
materialize_local_openai_cli_candidate_attempts(state, trace_id, &input, spec).await?;
for attempt in attempts {
if let Some(payload) = maybe_build_local_openai_cli_decision_payload_for_candidate(
state, parts, trace_id, body_json, &input, attempt, spec,
)
.await
{
return Ok(Some(payload));
}
}
Ok(None)
}
pub(crate) async fn maybe_build_stream_local_openai_cli_decision_payload(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
plan_kind: &str,
) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
let Some(spec) = resolve_stream_spec(plan_kind) else {
return Ok(None);
};
let Some(input) =
resolve_local_openai_cli_decision_input(state, trace_id, decision, body_json).await
else {
return Ok(None);
};
let attempts =
materialize_local_openai_cli_candidate_attempts(state, trace_id, &input, spec).await?;
for attempt in attempts {
if let Some(payload) = maybe_build_local_openai_cli_decision_payload_for_candidate(
state, parts, trace_id, body_json, &input, attempt, spec,
)
.await
{
return Ok(Some(payload));
}
}
Ok(None)
}
@@ -0,0 +1,771 @@
use std::collections::{BTreeMap, BTreeSet};
use aether_data::repository::candidates::{RequestCandidateStatus, UpsertRequestCandidateRecord};
use serde_json::{json, Value};
use tracing::warn;
use uuid::Uuid;
use crate::gateway::ai_pipeline::planner::standard::openai::{
build_cross_format_openai_cli_request_body, build_cross_format_openai_cli_upstream_url,
build_local_openai_cli_request_body, build_local_openai_cli_upstream_url,
};
use crate::gateway::ai_pipeline::conversion::{
request_conversion_direct_auth, request_conversion_kind, request_conversion_transport_supported,
};
use crate::gateway::headers::collect_control_headers;
use crate::gateway::provider_transport::{
apply_local_body_rules, apply_local_header_rules, build_antigravity_safe_v1internal_request,
build_antigravity_static_identity_headers, build_openai_passthrough_headers,
classify_local_antigravity_request_support, ensure_upstream_auth_header,
resolve_local_gemini_auth, resolve_local_standard_auth, resolve_transport_execution_timeouts,
resolve_transport_proxy_snapshot_with_tunnel_affinity, resolve_transport_tls_profile,
supports_local_standard_transport_with_network, AntigravityEnvelopeRequestType,
AntigravityRequestEnvelopeSupport, AntigravityRequestSideSupport,
LocalResolvedOAuthRequestAuth,
};
use crate::gateway::request_candidates::{
current_unix_secs, record_local_request_candidate_status,
};
use crate::gateway::ai_pipeline::planner::plan_builders::{
LocalStreamPlanAndReport, LocalSyncPlanAndReport,
};
use crate::gateway::ai_pipeline::planner::prefer_local_tunnel_owner_candidates;
use crate::gateway::scheduler::{
list_selectable_candidates, GatewayMinimalCandidateSelectionCandidate,
};
use crate::gateway::{
append_execution_contract_fields_to_value, AppState, ConversionMode, ExecutionStrategy,
GatewayControlDecision, GatewayControlSyncDecisionResponse, GatewayError,
};
use crate::gateway::ai_pipeline::planner::{
EXECUTION_RUNTIME_STREAM_DECISION_ACTION, EXECUTION_RUNTIME_SYNC_DECISION_ACTION,
};
const ANTIGRAVITY_ENVELOPE_NAME: &str = "antigravity:v1internal";
#[derive(Debug, Clone, Copy)]
pub(super) struct LocalOpenAiCliSpec {
pub(super) api_format: &'static str,
pub(super) decision_kind: &'static str,
pub(super) report_kind: &'static str,
pub(super) compact: bool,
pub(super) require_streaming: bool,
}
#[derive(Debug, Clone)]
pub(super) struct LocalOpenAiCliDecisionInput {
pub(super) auth_context: crate::gateway::GatewayControlAuthContext,
pub(super) requested_model: String,
pub(super) auth_snapshot: crate::gateway::gateway_data::StoredGatewayAuthApiKeySnapshot,
}
#[derive(Debug, Clone)]
pub(super) struct LocalOpenAiCliCandidateAttempt {
pub(super) candidate: GatewayMinimalCandidateSelectionCandidate,
pub(super) candidate_index: u32,
pub(super) candidate_id: String,
}
pub(super) async fn resolve_local_openai_cli_decision_input(
state: &AppState,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
) -> Option<LocalOpenAiCliDecisionInput> {
let Some(auth_context) = decision.auth_context.clone().filter(|auth_context| {
!auth_context.user_id.trim().is_empty() && !auth_context.api_key_id.trim().is_empty()
}) else {
return None;
};
let requested_model = body_json
.get("model")
.and_then(|value| value.as_str())
.map(str::trim)
.filter(|value| !value.is_empty())
.map(ToOwned::to_owned)?;
let auth_snapshot = match state
.read_auth_api_key_snapshot(
&auth_context.user_id,
&auth_context.api_key_id,
current_unix_secs(),
)
.await
{
Ok(Some(snapshot)) => snapshot,
Ok(None) => return None,
Err(err) => {
warn!(
trace_id = %trace_id,
error = ?err,
"gateway local openai cli decision auth snapshot read failed"
);
return None;
}
};
Some(LocalOpenAiCliDecisionInput {
auth_context,
requested_model,
auth_snapshot,
})
}
pub(super) async fn materialize_local_openai_cli_candidate_attempts(
state: &AppState,
trace_id: &str,
input: &LocalOpenAiCliDecisionInput,
spec: LocalOpenAiCliSpec,
) -> Result<Vec<LocalOpenAiCliCandidateAttempt>, GatewayError> {
let mut seen_candidates = BTreeSet::new();
let mut candidates = Vec::new();
for candidate_api_format in candidate_api_formats_for_spec(spec) {
let auth_snapshot = if *candidate_api_format == spec.api_format {
Some(&input.auth_snapshot)
} else {
None
};
let mut selected_candidates = list_selectable_candidates(
state,
candidate_api_format,
&input.requested_model,
spec.require_streaming,
auth_snapshot,
current_unix_secs(),
)
.await?;
if auth_snapshot.is_none() {
selected_candidates.retain(|candidate| {
auth_snapshot_allows_cross_format_openai_cli_candidate(
&input.auth_snapshot,
&input.requested_model,
candidate,
)
});
}
for candidate in selected_candidates {
let candidate_key = format!(
"{}:{}:{}:{}:{}:{}",
candidate.provider_id,
candidate.endpoint_id,
candidate.key_id,
candidate.model_id,
candidate.selected_provider_model_name,
candidate.endpoint_api_format,
);
if seen_candidates.insert(candidate_key) {
candidates.push(candidate);
}
}
}
let candidates = prefer_local_tunnel_owner_candidates(state, candidates).await;
let created_at_unix_secs = current_unix_secs();
let mut attempts = Vec::with_capacity(candidates.len());
for (candidate_index, candidate) in candidates.into_iter().enumerate() {
let generated_candidate_id = Uuid::new_v4().to_string();
let provider_api_format = candidate.endpoint_api_format.trim().to_ascii_lowercase();
let execution_strategy =
if provider_api_format == spec.api_format.trim().to_ascii_lowercase() {
ExecutionStrategy::LocalSameFormat
} else {
ExecutionStrategy::LocalCrossFormat
};
let conversion_mode =
if request_conversion_kind(spec.api_format, provider_api_format.as_str()).is_some() {
ConversionMode::Bidirectional
} else {
ConversionMode::None
};
let extra_data = append_execution_contract_fields_to_value(
json!({
"provider_api_format": provider_api_format,
"client_api_format": spec.api_format,
"global_model_id": candidate.global_model_id.clone(),
"global_model_name": candidate.global_model_name.clone(),
"model_id": candidate.model_id.clone(),
"selected_provider_model_name": candidate.selected_provider_model_name.clone(),
"mapping_matched_model": candidate.mapping_matched_model.clone(),
"provider_name": candidate.provider_name.clone(),
"key_name": candidate.key_name.clone(),
}),
execution_strategy,
conversion_mode,
spec.api_format,
candidate.endpoint_api_format.as_str(),
);
let candidate_id = match state
.upsert_request_candidate(UpsertRequestCandidateRecord {
id: generated_candidate_id.clone(),
request_id: trace_id.to_string(),
user_id: Some(input.auth_context.user_id.clone()),
api_key_id: Some(input.auth_context.api_key_id.clone()),
username: None,
api_key_name: None,
candidate_index: candidate_index as u32,
retry_index: 0,
provider_id: Some(candidate.provider_id.clone()),
endpoint_id: Some(candidate.endpoint_id.clone()),
key_id: Some(candidate.key_id.clone()),
status: RequestCandidateStatus::Available,
skip_reason: None,
is_cached: Some(false),
status_code: None,
error_type: None,
error_message: None,
latency_ms: None,
concurrent_requests: None,
extra_data: Some(extra_data),
required_capabilities: candidate.key_capabilities.clone(),
created_at_unix_secs: Some(created_at_unix_secs),
started_at_unix_secs: None,
finished_at_unix_secs: None,
})
.await
{
Ok(Some(stored)) => stored.id,
Ok(None) => generated_candidate_id.clone(),
Err(err) => {
warn!(
trace_id = %trace_id,
api_format = spec.api_format,
error = ?err,
"gateway local openai cli decision request candidate upsert failed"
);
generated_candidate_id.clone()
}
};
attempts.push(LocalOpenAiCliCandidateAttempt {
candidate,
candidate_index: candidate_index as u32,
candidate_id,
});
}
Ok(attempts)
}
fn auth_snapshot_allows_cross_format_openai_cli_candidate(
auth_snapshot: &crate::gateway::gateway_data::StoredGatewayAuthApiKeySnapshot,
requested_model: &str,
candidate: &GatewayMinimalCandidateSelectionCandidate,
) -> bool {
if let Some(allowed_providers) = auth_snapshot.effective_allowed_providers() {
let provider_allowed = allowed_providers.iter().any(|value| {
value
.trim()
.eq_ignore_ascii_case(candidate.provider_id.trim())
|| value
.trim()
.eq_ignore_ascii_case(candidate.provider_name.trim())
});
if !provider_allowed {
return false;
}
}
if let Some(allowed_models) = auth_snapshot.effective_allowed_models() {
let model_allowed = allowed_models
.iter()
.any(|value| value == requested_model || value == &candidate.global_model_name);
if !model_allowed {
return false;
}
}
true
}
pub(super) async fn maybe_build_local_openai_cli_decision_payload_for_candidate(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
body_json: &serde_json::Value,
input: &LocalOpenAiCliDecisionInput,
attempt: LocalOpenAiCliCandidateAttempt,
spec: LocalOpenAiCliSpec,
) -> Option<GatewayControlSyncDecisionResponse> {
let LocalOpenAiCliCandidateAttempt {
candidate,
candidate_index,
candidate_id,
} = attempt;
let provider_api_format = candidate.endpoint_api_format.trim().to_ascii_lowercase();
let transport = match state
.read_provider_transport_snapshot(
&candidate.provider_id,
&candidate.endpoint_id,
&candidate.key_id,
)
.await
{
Ok(Some(snapshot)) => snapshot,
Ok(None) => {
mark_skipped_local_openai_cli_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_snapshot_missing",
)
.await;
return None;
}
Err(err) => {
warn!(
trace_id = %trace_id,
api_format = spec.api_format,
error = ?err,
"gateway local openai cli decision provider transport read failed"
);
mark_skipped_local_openai_cli_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_snapshot_read_failed",
)
.await;
return None;
}
};
let is_antigravity = transport
.provider
.provider_type
.trim()
.eq_ignore_ascii_case("antigravity");
let same_format = provider_api_format == spec.api_format.trim().to_ascii_lowercase();
let conversion_kind = request_conversion_kind(spec.api_format, provider_api_format.as_str());
let transport_supported = if same_format {
supports_local_standard_transport_with_network(&transport, provider_api_format.as_str())
} else {
match conversion_kind {
Some(_) if is_antigravity && provider_api_format == "gemini:cli" => true,
Some(kind) => request_conversion_transport_supported(&transport, kind),
None => false,
}
};
if !transport_supported {
mark_skipped_local_openai_cli_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_unsupported",
)
.await;
return None;
}
let resolved_auth = if same_format {
match provider_api_format.as_str() {
"gemini:cli" => resolve_local_gemini_auth(&transport),
"claude:cli" | "openai:cli" | "openai:compact" => {
resolve_local_standard_auth(&transport)
}
_ => None,
}
} else {
conversion_kind.and_then(|kind| request_conversion_direct_auth(&transport, kind))
};
let oauth_auth = if resolved_auth.is_none() {
match state.resolve_local_oauth_request_auth(&transport).await {
Ok(Some(LocalResolvedOAuthRequestAuth::Header { name, value })) => Some((name, value)),
Ok(Some(LocalResolvedOAuthRequestAuth::Kiro(_))) => None,
Ok(None) => None,
Err(err) => {
warn!(
trace_id = %trace_id,
api_format = spec.api_format,
provider_type = %transport.provider.provider_type,
error = ?err,
"gateway local openai cli oauth auth resolution failed"
);
None
}
}
} else {
None
};
let Some((auth_header, auth_value)) = resolved_auth.or(oauth_auth) else {
mark_skipped_local_openai_cli_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_auth_unavailable",
)
.await;
return None;
};
let mapped_model = candidate.selected_provider_model_name.trim().to_string();
if mapped_model.is_empty() {
mark_skipped_local_openai_cli_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"mapped_model_missing",
)
.await;
return None;
}
let needs_bidirectional_conversion = !same_format && conversion_kind.is_some();
let Some(base_provider_request_body) = (if needs_bidirectional_conversion {
build_cross_format_openai_cli_request_body(
body_json,
&mapped_model,
spec.api_format,
provider_api_format.as_str(),
spec.require_streaming,
transport.endpoint.body_rules.as_ref(),
)
} else {
build_local_openai_cli_request_body(
body_json,
&mapped_model,
spec.require_streaming,
transport.endpoint.body_rules.as_ref(),
)
}) else {
mark_skipped_local_openai_cli_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"provider_request_body_missing",
)
.await;
return None;
};
let antigravity_auth = if is_antigravity {
match classify_local_antigravity_request_support(
&transport,
&base_provider_request_body,
AntigravityEnvelopeRequestType::Agent,
) {
AntigravityRequestSideSupport::Supported(spec) => Some(spec.auth),
AntigravityRequestSideSupport::Unsupported(_) => {
mark_skipped_local_openai_cli_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_unsupported",
)
.await;
return None;
}
}
} else {
None
};
let provider_request_body = if let Some(antigravity_auth) = antigravity_auth.as_ref() {
match build_antigravity_safe_v1internal_request(
antigravity_auth,
trace_id,
&mapped_model,
&base_provider_request_body,
AntigravityEnvelopeRequestType::Agent,
) {
AntigravityRequestEnvelopeSupport::Supported(envelope) => envelope,
AntigravityRequestEnvelopeSupport::Unsupported(_) => {
mark_skipped_local_openai_cli_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"provider_request_body_missing",
)
.await;
return None;
}
}
} else {
base_provider_request_body
};
let upstream_is_stream = spec.require_streaming || is_antigravity;
let Some(upstream_url) = (if needs_bidirectional_conversion {
build_cross_format_openai_cli_upstream_url(
parts,
&transport,
&mapped_model,
spec.api_format,
provider_api_format.as_str(),
upstream_is_stream,
)
} else {
build_local_openai_cli_upstream_url(
parts,
&transport,
provider_api_format.as_str() == "openai:compact",
)
}) else {
mark_skipped_local_openai_cli_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"upstream_url_missing",
)
.await;
return None;
};
let mut provider_request_headers = build_openai_passthrough_headers(
&parts.headers,
&auth_header,
&auth_value,
&antigravity_auth
.as_ref()
.map(build_antigravity_static_identity_headers)
.unwrap_or_default(),
Some("application/json"),
);
if !apply_local_header_rules(
&mut provider_request_headers,
transport.endpoint.header_rules.as_ref(),
&[&auth_header, "content-type"],
&provider_request_body,
Some(body_json),
) {
mark_skipped_local_openai_cli_candidate(
state,
input,
trace_id,
&candidate,
candidate_index,
&candidate_id,
"transport_header_rules_apply_failed",
)
.await;
return None;
}
ensure_upstream_auth_header(&mut provider_request_headers, &auth_header, &auth_value);
if upstream_is_stream {
provider_request_headers
.entry("accept".to_string())
.or_insert_with(|| "text/event-stream".to_string());
}
let prompt_cache_key = provider_request_body
.get("prompt_cache_key")
.and_then(|value| value.as_str())
.map(str::trim)
.filter(|value| !value.is_empty())
.map(ToOwned::to_owned);
let proxy = resolve_transport_proxy_snapshot_with_tunnel_affinity(state, &transport).await;
let tls_profile = resolve_transport_tls_profile(&transport);
let execution_strategy = if provider_api_format == spec.api_format {
ExecutionStrategy::LocalSameFormat
} else {
ExecutionStrategy::LocalCrossFormat
};
let conversion_mode = if needs_bidirectional_conversion {
ConversionMode::Bidirectional
} else {
ConversionMode::None
};
Some(GatewayControlSyncDecisionResponse {
action: if spec.require_streaming {
EXECUTION_RUNTIME_STREAM_DECISION_ACTION.to_string()
} else {
EXECUTION_RUNTIME_SYNC_DECISION_ACTION.to_string()
},
decision_kind: Some(spec.decision_kind.to_string()),
execution_strategy: Some(execution_strategy.as_str().to_string()),
conversion_mode: Some(conversion_mode.as_str().to_string()),
request_id: Some(trace_id.to_string()),
candidate_id: Some(candidate_id.clone()),
provider_name: Some(transport.provider.name.clone()),
provider_id: Some(candidate.provider_id.clone()),
endpoint_id: Some(candidate.endpoint_id.clone()),
key_id: Some(candidate.key_id.clone()),
upstream_base_url: Some(transport.endpoint.base_url.clone()),
upstream_url: Some(upstream_url.clone()),
provider_request_method: None,
auth_header: Some(auth_header),
auth_value: Some(auth_value),
provider_api_format: Some(provider_api_format.clone()),
client_api_format: Some(spec.api_format.to_string()),
provider_contract: Some(provider_api_format.clone()),
client_contract: Some(spec.api_format.to_string()),
model_name: Some(input.requested_model.clone()),
mapped_model: Some(mapped_model.clone()),
prompt_cache_key,
extra_headers: BTreeMap::new(),
provider_request_headers: provider_request_headers.clone(),
provider_request_body: Some(provider_request_body.clone()),
provider_request_body_base64: None,
content_type: Some("application/json".to_string()),
proxy,
tls_profile,
timeouts: resolve_transport_execution_timeouts(&transport),
upstream_is_stream,
report_kind: Some(spec.report_kind.to_string()),
report_context: Some(append_execution_contract_fields_to_value(
json!({
"user_id": input.auth_context.user_id,
"api_key_id": input.auth_context.api_key_id,
"request_id": trace_id,
"candidate_id": candidate_id,
"candidate_index": candidate_index,
"retry_index": 0,
"model": input.requested_model,
"provider_name": transport.provider.name,
"provider_id": candidate.provider_id,
"endpoint_id": candidate.endpoint_id,
"key_id": candidate.key_id,
"provider_api_format": provider_api_format,
"client_api_format": spec.api_format,
"mapped_model": mapped_model,
"upstream_url": upstream_url,
"provider_request_method": serde_json::Value::Null,
"provider_request_headers": provider_request_headers,
"provider_request_body": provider_request_body,
"original_headers": collect_control_headers(&parts.headers),
"original_request_body": body_json,
"has_envelope": is_antigravity,
"envelope_name": if is_antigravity {
Some(ANTIGRAVITY_ENVELOPE_NAME)
} else {
None
},
"needs_conversion": needs_bidirectional_conversion,
}),
execution_strategy,
conversion_mode,
spec.api_format,
candidate.endpoint_api_format.as_str(),
)),
auth_context: Some(input.auth_context.clone()),
})
}
fn candidate_api_formats_for_spec(spec: LocalOpenAiCliSpec) -> &'static [&'static str] {
match spec.api_format {
"openai:compact" => &["openai:compact", "openai:cli", "claude:cli", "gemini:cli"],
"openai:cli" => &["openai:cli", "claude:cli", "gemini:cli"],
_ => &[],
}
}
async fn mark_skipped_local_openai_cli_candidate(
state: &AppState,
input: &LocalOpenAiCliDecisionInput,
trace_id: &str,
candidate: &GatewayMinimalCandidateSelectionCandidate,
candidate_index: u32,
candidate_id: &str,
skip_reason: &'static str,
) {
if let Err(err) = state
.upsert_request_candidate(UpsertRequestCandidateRecord {
id: candidate_id.to_string(),
request_id: trace_id.to_string(),
user_id: Some(input.auth_context.user_id.clone()),
api_key_id: Some(input.auth_context.api_key_id.clone()),
username: None,
api_key_name: None,
candidate_index,
retry_index: 0,
provider_id: Some(candidate.provider_id.clone()),
endpoint_id: Some(candidate.endpoint_id.clone()),
key_id: Some(candidate.key_id.clone()),
status: RequestCandidateStatus::Skipped,
skip_reason: Some(skip_reason.to_string()),
is_cached: Some(false),
status_code: None,
error_type: None,
error_message: None,
latency_ms: None,
concurrent_requests: None,
extra_data: None,
required_capabilities: candidate.key_capabilities.clone(),
created_at_unix_secs: None,
started_at_unix_secs: None,
finished_at_unix_secs: Some(current_unix_secs()),
})
.await
{
warn!(
trace_id = %trace_id,
candidate_id = %candidate_id,
skip_reason,
error = ?err,
"gateway local openai cli decision failed to persist skipped candidate"
);
}
}
pub(super) async fn mark_unused_local_openai_cli_candidates<T>(state: &AppState, remaining: Vec<T>)
where
T: LocalOpenAiCliPlanAndReport,
{
for plan_and_report in remaining {
record_local_request_candidate_status(
state,
plan_and_report.plan(),
plan_and_report.report_context(),
RequestCandidateStatus::Unused,
None,
None,
None,
None,
None,
None,
)
.await;
}
}
pub(super) trait LocalOpenAiCliPlanAndReport {
fn plan(&self) -> &aether_contracts::ExecutionPlan;
fn report_context(&self) -> Option<&serde_json::Value>;
}
impl LocalOpenAiCliPlanAndReport for LocalSyncPlanAndReport {
fn plan(&self) -> &aether_contracts::ExecutionPlan {
&self.plan
}
fn report_context(&self) -> Option<&serde_json::Value> {
self.report_context.as_ref()
}
}
impl LocalOpenAiCliPlanAndReport for LocalStreamPlanAndReport {
fn plan(&self) -> &aether_contracts::ExecutionPlan {
&self.plan
}
fn report_context(&self) -> Option<&serde_json::Value> {
self.report_context.as_ref()
}
}
@@ -0,0 +1,144 @@
use tracing::warn;
use super::decision::{
materialize_local_openai_cli_candidate_attempts,
maybe_build_local_openai_cli_decision_payload_for_candidate,
resolve_local_openai_cli_decision_input, LocalOpenAiCliSpec,
};
use crate::gateway::ai_pipeline::planner::plan_builders::{
build_openai_cli_stream_plan_from_decision, build_openai_cli_sync_plan_from_decision,
LocalStreamPlanAndReport, LocalSyncPlanAndReport,
};
use crate::gateway::{AppState, GatewayControlDecision, GatewayError};
use crate::gateway::ai_pipeline::planner::{
OPENAI_CLI_STREAM_PLAN_KIND, OPENAI_CLI_SYNC_PLAN_KIND, OPENAI_COMPACT_STREAM_PLAN_KIND,
OPENAI_COMPACT_SYNC_PLAN_KIND,
};
pub(super) fn resolve_sync_spec(plan_kind: &str) -> Option<LocalOpenAiCliSpec> {
match plan_kind {
OPENAI_CLI_SYNC_PLAN_KIND => Some(LocalOpenAiCliSpec {
api_format: "openai:cli",
decision_kind: OPENAI_CLI_SYNC_PLAN_KIND,
report_kind: "openai_cli_sync_success",
compact: false,
require_streaming: false,
}),
OPENAI_COMPACT_SYNC_PLAN_KIND => Some(LocalOpenAiCliSpec {
api_format: "openai:compact",
decision_kind: OPENAI_COMPACT_SYNC_PLAN_KIND,
report_kind: "openai_cli_sync_success",
compact: true,
require_streaming: false,
}),
_ => None,
}
}
pub(super) fn resolve_stream_spec(plan_kind: &str) -> Option<LocalOpenAiCliSpec> {
match plan_kind {
OPENAI_CLI_STREAM_PLAN_KIND => Some(LocalOpenAiCliSpec {
api_format: "openai:cli",
decision_kind: OPENAI_CLI_STREAM_PLAN_KIND,
report_kind: "openai_cli_stream_success",
compact: false,
require_streaming: true,
}),
OPENAI_COMPACT_STREAM_PLAN_KIND => Some(LocalOpenAiCliSpec {
api_format: "openai:compact",
decision_kind: OPENAI_COMPACT_STREAM_PLAN_KIND,
report_kind: "openai_cli_stream_success",
compact: true,
require_streaming: true,
}),
_ => None,
}
}
pub(super) async fn build_local_sync_plan_and_reports(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
spec: LocalOpenAiCliSpec,
) -> Result<Vec<LocalSyncPlanAndReport>, GatewayError> {
let Some(input) =
resolve_local_openai_cli_decision_input(state, trace_id, decision, body_json).await
else {
return Ok(Vec::new());
};
let attempts =
materialize_local_openai_cli_candidate_attempts(state, trace_id, &input, spec).await?;
let mut plans = Vec::new();
for attempt in attempts {
let Some(payload) = maybe_build_local_openai_cli_decision_payload_for_candidate(
state, parts, trace_id, body_json, &input, attempt, spec,
)
.await
else {
continue;
};
match build_openai_cli_sync_plan_from_decision(parts, body_json, payload, spec.compact) {
Ok(Some(value)) => plans.push(value),
Ok(None) => {}
Err(err) => {
warn!(
trace_id = %trace_id,
api_format = spec.api_format,
error = ?err,
"gateway local openai cli sync decision plan build failed"
);
}
}
}
Ok(plans)
}
pub(super) async fn build_local_stream_plan_and_reports(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
decision: &GatewayControlDecision,
body_json: &serde_json::Value,
spec: LocalOpenAiCliSpec,
) -> Result<Vec<LocalStreamPlanAndReport>, GatewayError> {
let Some(input) =
resolve_local_openai_cli_decision_input(state, trace_id, decision, body_json).await
else {
return Ok(Vec::new());
};
let attempts =
materialize_local_openai_cli_candidate_attempts(state, trace_id, &input, spec).await?;
let mut plans = Vec::new();
for attempt in attempts {
let Some(payload) = maybe_build_local_openai_cli_decision_payload_for_candidate(
state, parts, trace_id, body_json, &input, attempt, spec,
)
.await
else {
continue;
};
match build_openai_cli_stream_plan_from_decision(parts, body_json, payload, spec.compact) {
Ok(Some(value)) => plans.push(value),
Ok(None) => {}
Err(err) => {
warn!(
trace_id = %trace_id,
api_format = spec.api_format,
error = ?err,
"gateway local openai cli stream decision plan build failed"
);
}
}
}
Ok(plans)
}
@@ -0,0 +1,31 @@
mod chat;
mod cli;
mod normalize_chat;
mod normalize_cli;
pub(crate) use crate::gateway::ai_pipeline::conversion::request::{
convert_openai_chat_request_to_claude_request, convert_openai_chat_request_to_gemini_request,
convert_openai_chat_request_to_openai_cli_request, extract_openai_text_content,
normalize_openai_cli_request_to_openai_chat_request, parse_openai_tool_result_content,
};
pub(crate) use self::chat::{
copy_request_number_field, copy_request_number_field_as,
map_openai_reasoning_effort_to_claude_output, map_openai_reasoning_effort_to_gemini_budget,
maybe_build_stream_local_decision_payload, maybe_build_sync_local_decision_payload,
maybe_execute_stream_via_local_decision, maybe_execute_sync_via_local_decision,
parse_openai_stop_sequences, resolve_openai_chat_max_tokens, value_as_u64,
};
pub(crate) use self::cli::{
maybe_build_stream_local_openai_cli_decision_payload,
maybe_build_sync_local_openai_cli_decision_payload,
maybe_execute_stream_via_local_openai_cli_decision,
maybe_execute_sync_via_local_openai_cli_decision,
};
pub(crate) use self::normalize_chat::{
build_cross_format_openai_chat_request_body, build_cross_format_openai_chat_upstream_url,
build_local_openai_chat_request_body, build_local_openai_chat_upstream_url,
};
pub(crate) use self::normalize_cli::{
build_cross_format_openai_cli_request_body, build_cross_format_openai_cli_upstream_url,
build_local_openai_cli_request_body, build_local_openai_cli_upstream_url,
};
@@ -0,0 +1,5 @@
mod request;
pub(crate) use self::request::{
build_cross_format_openai_chat_request_body, build_cross_format_openai_chat_upstream_url,
build_local_openai_chat_request_body, build_local_openai_chat_upstream_url,
};
@@ -0,0 +1,143 @@
use serde_json::Value;
use crate::gateway::ai_pipeline::conversion::request::{
convert_openai_chat_request_to_claude_request, convert_openai_chat_request_to_gemini_request,
convert_openai_chat_request_to_openai_cli_request,
};
use crate::gateway::ai_pipeline::conversion::{request_conversion_kind, RequestConversionKind};
use crate::gateway::provider_transport::{
apply_local_body_rules, build_claude_messages_url, build_gemini_content_url,
build_openai_chat_url, build_openai_cli_url, build_passthrough_path_url,
};
pub(crate) fn build_local_openai_chat_request_body(
body_json: &serde_json::Value,
mapped_model: &str,
upstream_is_stream: bool,
body_rules: Option<&serde_json::Value>,
) -> Option<serde_json::Value> {
let request_body_object = body_json.as_object()?;
let mut provider_request_body = serde_json::Map::from_iter(
request_body_object
.iter()
.map(|(key, value)| (key.clone(), value.clone())),
);
provider_request_body.insert(
"model".to_string(),
serde_json::Value::String(mapped_model.to_string()),
);
if upstream_is_stream {
provider_request_body.insert("stream".to_string(), serde_json::Value::Bool(true));
}
let mut provider_request_body = serde_json::Value::Object(provider_request_body);
if !apply_local_body_rules(&mut provider_request_body, body_rules, Some(body_json)) {
return None;
}
Some(provider_request_body)
}
pub(crate) fn build_local_openai_chat_upstream_url(
parts: &http::request::Parts,
transport: &crate::gateway::provider_transport::GatewayProviderTransportSnapshot,
) -> Option<String> {
let custom_path = transport
.endpoint
.custom_path
.as_deref()
.map(str::trim)
.filter(|value| !value.is_empty());
match custom_path {
Some(path) => {
build_passthrough_path_url(&transport.endpoint.base_url, path, parts.uri.query(), &[])
}
None => Some(build_openai_chat_url(
&transport.endpoint.base_url,
parts.uri.query(),
)),
}
}
pub(crate) fn build_cross_format_openai_chat_request_body(
body_json: &Value,
mapped_model: &str,
provider_api_format: &str,
upstream_is_stream: bool,
body_rules: Option<&Value>,
) -> Option<Value> {
let conversion_kind = request_conversion_kind("openai:chat", provider_api_format)?;
let mut provider_request_body = match conversion_kind {
RequestConversionKind::ToClaudeStandard => convert_openai_chat_request_to_claude_request(
body_json,
mapped_model,
upstream_is_stream,
)?,
RequestConversionKind::ToGeminiStandard => convert_openai_chat_request_to_gemini_request(
body_json,
mapped_model,
upstream_is_stream,
)?,
RequestConversionKind::ToOpenAIFamilyCli => {
convert_openai_chat_request_to_openai_cli_request(
body_json,
mapped_model,
upstream_is_stream,
false,
)?
}
RequestConversionKind::ToOpenAICompact => {
convert_openai_chat_request_to_openai_cli_request(body_json, mapped_model, false, true)?
}
_ => return None,
};
if !apply_local_body_rules(&mut provider_request_body, body_rules, Some(body_json)) {
return None;
}
Some(provider_request_body)
}
pub(crate) fn build_cross_format_openai_chat_upstream_url(
parts: &http::request::Parts,
transport: &crate::gateway::provider_transport::GatewayProviderTransportSnapshot,
mapped_model: &str,
provider_api_format: &str,
upstream_is_stream: bool,
) -> Option<String> {
let conversion_kind = request_conversion_kind("openai:chat", provider_api_format)?;
let custom_path = transport
.endpoint
.custom_path
.as_deref()
.map(str::trim)
.filter(|value| !value.is_empty());
match custom_path {
Some(path) => {
build_passthrough_path_url(&transport.endpoint.base_url, path, parts.uri.query(), &[])
}
None => match conversion_kind {
RequestConversionKind::ToClaudeStandard => Some(build_claude_messages_url(
&transport.endpoint.base_url,
parts.uri.query(),
)),
RequestConversionKind::ToGeminiStandard => build_gemini_content_url(
&transport.endpoint.base_url,
mapped_model,
upstream_is_stream,
parts.uri.query(),
),
RequestConversionKind::ToOpenAIFamilyCli => Some(build_openai_cli_url(
&transport.endpoint.base_url,
parts.uri.query(),
false,
)),
RequestConversionKind::ToOpenAICompact => Some(build_openai_cli_url(
&transport.endpoint.base_url,
parts.uri.query(),
true,
)),
_ => None,
},
}
}
@@ -0,0 +1,203 @@
use std::collections::BTreeMap;
use serde_json::{json, Map, Value};
use url::form_urlencoded;
use crate::gateway::ai_pipeline::conversion::request::{
convert_openai_chat_request_to_claude_request, convert_openai_chat_request_to_gemini_request,
convert_openai_chat_request_to_openai_cli_request, extract_openai_text_content,
normalize_openai_cli_request_to_openai_chat_request, parse_openai_tool_result_content,
};
use crate::gateway::ai_pipeline::conversion::{request_conversion_kind, RequestConversionKind};
use crate::gateway::provider_transport::{
apply_local_body_rules, build_antigravity_v1internal_url, build_claude_messages_url,
build_gemini_content_url, build_openai_cli_url, build_passthrough_path_url,
AntigravityRequestUrlAction,
};
pub(crate) fn build_local_openai_cli_request_body(
body_json: &Value,
mapped_model: &str,
require_streaming: bool,
body_rules: Option<&Value>,
) -> Option<Value> {
let request_body_object = body_json.as_object()?;
let mut provider_request_body = serde_json::Map::from_iter(
request_body_object
.iter()
.map(|(key, value)| (key.clone(), value.clone())),
);
provider_request_body.insert("model".to_string(), Value::String(mapped_model.to_string()));
if require_streaming {
provider_request_body.insert("stream".to_string(), Value::Bool(true));
}
let mut provider_request_body = Value::Object(provider_request_body);
if !apply_local_body_rules(&mut provider_request_body, body_rules, Some(body_json)) {
return None;
}
Some(provider_request_body)
}
pub(crate) fn build_cross_format_openai_cli_request_body(
body_json: &Value,
mapped_model: &str,
client_api_format: &str,
provider_api_format: &str,
upstream_is_stream: bool,
body_rules: Option<&Value>,
) -> Option<Value> {
let chat_like_request = normalize_openai_cli_request_to_openai_chat_request(body_json)?;
let conversion_kind = request_conversion_kind(client_api_format, provider_api_format)?;
let mut provider_request_body = match conversion_kind {
RequestConversionKind::ToOpenAIFamilyCli => {
convert_openai_chat_request_to_openai_cli_request(
&chat_like_request,
mapped_model,
upstream_is_stream,
false,
)?
}
RequestConversionKind::ToOpenAICompact => {
convert_openai_chat_request_to_openai_cli_request(
&chat_like_request,
mapped_model,
false,
true,
)?
}
RequestConversionKind::ToClaudeStandard => convert_openai_chat_request_to_claude_request(
&chat_like_request,
mapped_model,
upstream_is_stream,
)?,
RequestConversionKind::ToGeminiStandard => convert_openai_chat_request_to_gemini_request(
&chat_like_request,
mapped_model,
upstream_is_stream,
)?,
_ => return None,
};
if !apply_local_body_rules(&mut provider_request_body, body_rules, Some(body_json)) {
return None;
}
Some(provider_request_body)
}
pub(crate) fn build_local_openai_cli_upstream_url(
parts: &http::request::Parts,
transport: &crate::gateway::provider_transport::GatewayProviderTransportSnapshot,
compact: bool,
) -> Option<String> {
let custom_path = transport
.endpoint
.custom_path
.as_deref()
.map(str::trim)
.filter(|value| !value.is_empty());
match custom_path {
Some(path) => {
build_passthrough_path_url(&transport.endpoint.base_url, path, parts.uri.query(), &[])
}
None => Some(build_openai_cli_url(
&transport.endpoint.base_url,
parts.uri.query(),
compact,
)),
}
}
pub(crate) fn build_cross_format_openai_cli_upstream_url(
parts: &http::request::Parts,
transport: &crate::gateway::provider_transport::GatewayProviderTransportSnapshot,
mapped_model: &str,
client_api_format: &str,
provider_api_format: &str,
upstream_is_stream: bool,
) -> Option<String> {
let conversion_kind = request_conversion_kind(client_api_format, provider_api_format)?;
if transport
.provider
.provider_type
.trim()
.eq_ignore_ascii_case("antigravity")
{
let query = parts.uri.query().map(|query| {
form_urlencoded::parse(query.as_bytes())
.into_owned()
.collect::<BTreeMap<String, String>>()
});
return build_antigravity_v1internal_url(
&transport.endpoint.base_url,
if upstream_is_stream {
AntigravityRequestUrlAction::StreamGenerateContent
} else {
AntigravityRequestUrlAction::GenerateContent
},
query.as_ref(),
);
}
let custom_path = transport
.endpoint
.custom_path
.as_deref()
.map(str::trim)
.filter(|value| !value.is_empty());
match custom_path {
Some(path) => {
build_passthrough_path_url(&transport.endpoint.base_url, path, parts.uri.query(), &[])
}
None => match conversion_kind {
RequestConversionKind::ToOpenAIFamilyCli => Some(build_openai_cli_url(
&transport.endpoint.base_url,
parts.uri.query(),
false,
)),
RequestConversionKind::ToOpenAICompact => Some(build_openai_cli_url(
&transport.endpoint.base_url,
parts.uri.query(),
true,
)),
RequestConversionKind::ToClaudeStandard => Some(build_claude_messages_url(
&transport.endpoint.base_url,
parts.uri.query(),
)),
RequestConversionKind::ToGeminiStandard => build_gemini_content_url(
&transport.endpoint.base_url,
mapped_model,
upstream_is_stream,
parts.uri.query(),
),
_ => None,
},
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn builds_openai_family_cross_format_request_body_from_compact_source() {
let body_json = json!({
"model": "gpt-5",
"input": "hello",
});
let provider_request_body = build_cross_format_openai_cli_request_body(
&body_json,
"gpt-5-upstream",
"openai:compact",
"openai:cli",
false,
None,
)
.expect("compact to openai cli body should build");
assert_eq!(provider_request_body["model"], "gpt-5-upstream");
assert_eq!(provider_request_body["input"][0]["type"], "message");
assert_eq!(provider_request_body["input"][0]["role"], "user");
}
}
@@ -0,0 +1,623 @@
use aether_contracts::{ExecutionPlan, RequestBody};
use super::{
augment_sync_report_context, generic_decision_missing_exact_provider_request,
GatewayControlSyncDecisionResponse, GatewayError, LocalStreamPlanAndReport,
LocalSyncPlanAndReport,
};
use crate::gateway::ai_pipeline::private_surfaces::provider_adaptation_requires_eventstream_accept;
use crate::gateway::provider_transport::{
build_openai_chat_url, build_openai_cli_url, build_openai_passthrough_headers,
ensure_upstream_auth_header,
};
pub(crate) fn build_openai_chat_sync_plan_from_decision(
parts: &http::request::Parts,
body_json: &serde_json::Value,
payload: GatewayControlSyncDecisionResponse,
) -> Result<Option<LocalSyncPlanAndReport>, GatewayError> {
let Some(request_id) = payload
.request_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(provider_id) = payload
.provider_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(endpoint_id) = payload
.endpoint_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(key_id) = payload
.key_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(auth_header) = payload
.auth_header
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(auth_value) = payload
.auth_value
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(provider_api_format) = payload
.provider_api_format
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(client_api_format) = payload
.client_api_format
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let url = if let Some(upstream_url) = payload
.upstream_url
.clone()
.filter(|value| !value.trim().is_empty())
{
upstream_url
} else {
let Some(upstream_base_url) = payload
.upstream_base_url
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
build_openai_chat_url(&upstream_base_url, parts.uri.query())
};
let provider_request_body_value = if let Some(body) = payload.provider_request_body.clone() {
body
} else {
let Some(request_body_object) = body_json.as_object() else {
return Ok(None);
};
let mut provider_request_body = serde_json::Map::from_iter(
request_body_object
.iter()
.map(|(key, value)| (key.clone(), value.clone())),
);
if let Some(mapped_model) = payload
.mapped_model
.as_ref()
.filter(|value| !value.trim().is_empty())
{
provider_request_body.insert(
"model".to_string(),
serde_json::Value::String(mapped_model.clone()),
);
}
if payload.upstream_is_stream {
provider_request_body.insert("stream".to_string(), serde_json::Value::Bool(true));
}
if let Some(prompt_cache_key) = payload
.prompt_cache_key
.as_ref()
.filter(|value| !value.trim().is_empty())
{
let existing = provider_request_body
.get("prompt_cache_key")
.and_then(|value| value.as_str())
.map(str::trim)
.unwrap_or_default();
if existing.is_empty() {
provider_request_body.insert(
"prompt_cache_key".to_string(),
serde_json::Value::String(prompt_cache_key.clone()),
);
}
}
serde_json::Value::Object(provider_request_body)
};
let mut provider_request_headers = if payload.provider_request_headers.is_empty() {
build_openai_passthrough_headers(
&parts.headers,
&auth_header,
&auth_value,
&payload.extra_headers,
payload.content_type.as_deref(),
)
} else {
payload.provider_request_headers.clone()
};
ensure_upstream_auth_header(&mut provider_request_headers, &auth_header, &auth_value);
if payload.upstream_is_stream {
provider_request_headers
.entry("accept".to_string())
.or_insert_with(|| "text/event-stream".to_string());
}
let plan = ExecutionPlan {
request_id,
candidate_id: payload.candidate_id.clone(),
provider_name: payload.provider_name.clone(),
provider_id,
endpoint_id,
key_id,
method: "POST".to_string(),
url,
headers: std::mem::take(&mut provider_request_headers),
content_type: payload
.content_type
.clone()
.or_else(|| Some("application/json".to_string())),
content_encoding: None,
body: RequestBody::from_json(provider_request_body_value.clone()),
stream: payload.upstream_is_stream,
client_api_format,
provider_api_format,
model_name: payload.model_name.clone(),
proxy: payload.proxy.clone(),
tls_profile: payload.tls_profile.clone(),
timeouts: payload.timeouts.clone(),
};
let report_context = augment_sync_report_context(
payload.report_context,
&plan.headers,
&provider_request_body_value,
)?;
Ok(Some(LocalSyncPlanAndReport {
plan,
report_kind: payload.report_kind,
report_context,
}))
}
pub(crate) fn build_openai_cli_sync_plan_from_decision(
parts: &http::request::Parts,
_body_json: &serde_json::Value,
payload: GatewayControlSyncDecisionResponse,
compact: bool,
) -> Result<Option<LocalSyncPlanAndReport>, GatewayError> {
if generic_decision_missing_exact_provider_request(&payload) {
return Ok(None);
}
let Some(request_id) = payload
.request_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(provider_id) = payload
.provider_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(endpoint_id) = payload
.endpoint_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(key_id) = payload
.key_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let auth_header = payload
.auth_header
.clone()
.filter(|value| !value.trim().is_empty());
let auth_value = payload
.auth_value
.clone()
.filter(|value| !value.trim().is_empty());
if auth_header.is_some() != auth_value.is_some() {
return Ok(None);
}
let Some(provider_api_format) = payload
.provider_api_format
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(client_api_format) = payload
.client_api_format
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let url = if let Some(upstream_url) = payload
.upstream_url
.clone()
.filter(|value| !value.trim().is_empty())
{
upstream_url
} else {
let Some(upstream_base_url) = payload
.upstream_base_url
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
build_openai_cli_url(&upstream_base_url, parts.uri.query(), compact)
};
let Some(provider_request_body_value) = payload.provider_request_body.clone() else {
return Ok(None);
};
let mut provider_request_headers = payload.provider_request_headers.clone();
if let (Some(auth_header), Some(auth_value)) = (auth_header.as_deref(), auth_value.as_deref()) {
ensure_upstream_auth_header(&mut provider_request_headers, auth_header, auth_value);
}
if payload.upstream_is_stream && !provider_request_headers.contains_key("accept") {
provider_request_headers.insert("accept".to_string(), "text/event-stream".to_string());
}
let plan = ExecutionPlan {
request_id,
candidate_id: payload.candidate_id.clone(),
provider_name: payload.provider_name.clone(),
provider_id,
endpoint_id,
key_id,
method: "POST".to_string(),
url,
headers: std::mem::take(&mut provider_request_headers),
content_type: payload
.content_type
.clone()
.or_else(|| Some("application/json".to_string())),
content_encoding: None,
body: RequestBody::from_json(provider_request_body_value.clone()),
stream: payload.upstream_is_stream,
client_api_format,
provider_api_format,
model_name: payload.model_name.clone(),
proxy: payload.proxy.clone(),
tls_profile: payload.tls_profile.clone(),
timeouts: payload.timeouts.clone(),
};
let report_context = augment_sync_report_context(
payload.report_context,
&plan.headers,
&provider_request_body_value,
)?;
Ok(Some(LocalSyncPlanAndReport {
plan,
report_kind: payload.report_kind,
report_context,
}))
}
pub(crate) fn build_openai_chat_stream_plan_from_decision(
parts: &http::request::Parts,
body_json: &serde_json::Value,
payload: GatewayControlSyncDecisionResponse,
) -> Result<Option<LocalStreamPlanAndReport>, GatewayError> {
let Some(request_id) = payload
.request_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(provider_id) = payload
.provider_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(endpoint_id) = payload
.endpoint_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(key_id) = payload
.key_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(auth_header) = payload
.auth_header
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(auth_value) = payload
.auth_value
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(provider_api_format) = payload
.provider_api_format
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(client_api_format) = payload
.client_api_format
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let url = if let Some(upstream_url) = payload
.upstream_url
.clone()
.filter(|value| !value.trim().is_empty())
{
upstream_url
} else {
let Some(upstream_base_url) = payload
.upstream_base_url
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
build_openai_chat_url(&upstream_base_url, parts.uri.query())
};
let provider_request_body_value = if let Some(body) = payload.provider_request_body.clone() {
body
} else {
let Some(request_body_object) = body_json.as_object() else {
return Ok(None);
};
let mut provider_request_body = serde_json::Map::from_iter(
request_body_object
.iter()
.map(|(key, value)| (key.clone(), value.clone())),
);
if let Some(mapped_model) = payload
.mapped_model
.as_ref()
.filter(|value| !value.trim().is_empty())
{
provider_request_body.insert(
"model".to_string(),
serde_json::Value::String(mapped_model.clone()),
);
}
provider_request_body.insert("stream".to_string(), serde_json::Value::Bool(true));
if let Some(prompt_cache_key) = payload
.prompt_cache_key
.as_ref()
.filter(|value| !value.trim().is_empty())
{
let existing = provider_request_body
.get("prompt_cache_key")
.and_then(|value| value.as_str())
.map(str::trim)
.unwrap_or_default();
if existing.is_empty() {
provider_request_body.insert(
"prompt_cache_key".to_string(),
serde_json::Value::String(prompt_cache_key.clone()),
);
}
}
serde_json::Value::Object(provider_request_body)
};
let mut provider_request_headers = if payload.provider_request_headers.is_empty() {
build_openai_passthrough_headers(
&parts.headers,
&auth_header,
&auth_value,
&payload.extra_headers,
payload.content_type.as_deref(),
)
} else {
payload.provider_request_headers.clone()
};
ensure_upstream_auth_header(&mut provider_request_headers, &auth_header, &auth_value);
provider_request_headers.insert("accept".to_string(), "text/event-stream".to_string());
let plan = ExecutionPlan {
request_id,
candidate_id: payload.candidate_id.clone(),
provider_name: payload.provider_name.clone(),
provider_id,
endpoint_id,
key_id,
method: "POST".to_string(),
url,
headers: std::mem::take(&mut provider_request_headers),
content_type: payload
.content_type
.clone()
.or_else(|| Some("application/json".to_string())),
content_encoding: None,
body: RequestBody::from_json(provider_request_body_value.clone()),
stream: true,
client_api_format,
provider_api_format,
model_name: payload.model_name.clone(),
proxy: payload.proxy.clone(),
tls_profile: payload.tls_profile.clone(),
timeouts: payload.timeouts.clone(),
};
let report_context = augment_sync_report_context(
payload.report_context,
&plan.headers,
&provider_request_body_value,
)?;
Ok(Some(LocalStreamPlanAndReport {
plan,
report_kind: payload.report_kind,
report_context,
}))
}
pub(crate) fn build_openai_cli_stream_plan_from_decision(
parts: &http::request::Parts,
_body_json: &serde_json::Value,
payload: GatewayControlSyncDecisionResponse,
compact: bool,
) -> Result<Option<LocalStreamPlanAndReport>, GatewayError> {
if generic_decision_missing_exact_provider_request(&payload) {
return Ok(None);
}
let Some(request_id) = payload
.request_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(provider_id) = payload
.provider_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(endpoint_id) = payload
.endpoint_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(key_id) = payload
.key_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let auth_header = payload
.auth_header
.clone()
.filter(|value| !value.trim().is_empty());
let auth_value = payload
.auth_value
.clone()
.filter(|value| !value.trim().is_empty());
if auth_header.is_some() != auth_value.is_some() {
return Ok(None);
}
let Some(provider_api_format) = payload
.provider_api_format
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(client_api_format) = payload
.client_api_format
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let url = if let Some(upstream_url) = payload
.upstream_url
.clone()
.filter(|value| !value.trim().is_empty())
{
upstream_url
} else {
let Some(upstream_base_url) = payload
.upstream_base_url
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
build_openai_cli_url(&upstream_base_url, parts.uri.query(), compact)
};
let Some(provider_request_body_value) = payload.provider_request_body.clone() else {
return Ok(None);
};
let envelope_name = payload
.report_context
.as_ref()
.and_then(|context| context.get("envelope_name"))
.and_then(serde_json::Value::as_str);
let mut provider_request_headers = payload.provider_request_headers.clone();
if let (Some(auth_header), Some(auth_value)) = (auth_header.as_deref(), auth_value.as_deref()) {
ensure_upstream_auth_header(&mut provider_request_headers, auth_header, auth_value);
}
if provider_adaptation_requires_eventstream_accept(envelope_name, provider_api_format.as_str())
{
provider_request_headers
.entry("accept".to_string())
.or_insert_with(|| "application/vnd.amazon.eventstream".to_string());
} else {
provider_request_headers.insert("accept".to_string(), "text/event-stream".to_string());
}
let plan = ExecutionPlan {
request_id,
candidate_id: payload.candidate_id.clone(),
provider_name: payload.provider_name.clone(),
provider_id,
endpoint_id,
key_id,
method: "POST".to_string(),
url,
headers: std::mem::take(&mut provider_request_headers),
content_type: payload
.content_type
.clone()
.or_else(|| Some("application/json".to_string())),
content_encoding: None,
body: RequestBody::from_json(provider_request_body_value.clone()),
stream: true,
client_api_format,
provider_api_format,
model_name: payload.model_name.clone(),
proxy: payload.proxy.clone(),
tls_profile: payload.tls_profile.clone(),
timeouts: payload.timeouts.clone(),
};
let report_context = augment_sync_report_context(
payload.report_context,
&plan.headers,
&provider_request_body_value,
)?;
Ok(Some(LocalStreamPlanAndReport {
plan,
report_kind: payload.report_kind,
report_context,
}))
}
@@ -0,0 +1,253 @@
use aether_contracts::RequestBody;
use super::augment_sync_report_context;
use super::*;
use crate::gateway::ai_pipeline::private_surfaces::provider_adaptation_requires_eventstream_accept;
use crate::gateway::provider_transport::ensure_upstream_auth_header;
pub(crate) fn build_standard_sync_plan_from_decision(
_parts: &http::request::Parts,
_body_json: &serde_json::Value,
payload: GatewayControlSyncDecisionResponse,
) -> Result<Option<LocalSyncPlanAndReport>, GatewayError> {
if generic_decision_missing_exact_provider_request(&payload) {
return Ok(None);
}
let Some(request_id) = payload
.request_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(provider_id) = payload
.provider_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(endpoint_id) = payload
.endpoint_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(key_id) = payload
.key_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(url) = payload
.upstream_url
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let auth_header = payload
.auth_header
.clone()
.filter(|value| !value.trim().is_empty());
let auth_value = payload
.auth_value
.clone()
.filter(|value| !value.trim().is_empty());
if auth_header.is_some() != auth_value.is_some() {
return Ok(None);
}
let Some(provider_api_format) = payload
.provider_api_format
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(client_api_format) = payload
.client_api_format
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(provider_request_body_value) = payload.provider_request_body.clone() else {
return Ok(None);
};
let mut provider_request_headers = payload.provider_request_headers.clone();
if let (Some(auth_header), Some(auth_value)) = (auth_header.as_deref(), auth_value.as_deref()) {
ensure_upstream_auth_header(&mut provider_request_headers, auth_header, auth_value);
}
if payload.upstream_is_stream {
provider_request_headers
.entry("accept".to_string())
.or_insert_with(|| "text/event-stream".to_string());
}
let plan = ExecutionPlan {
request_id,
candidate_id: payload.candidate_id.clone(),
provider_name: payload.provider_name.clone(),
provider_id,
endpoint_id,
key_id,
method: "POST".to_string(),
url,
headers: std::mem::take(&mut provider_request_headers),
content_type: payload
.content_type
.clone()
.or_else(|| Some("application/json".to_string())),
content_encoding: None,
body: RequestBody::from_json(provider_request_body_value.clone()),
stream: payload.upstream_is_stream,
client_api_format,
provider_api_format,
model_name: payload.model_name.clone(),
proxy: payload.proxy.clone(),
tls_profile: payload.tls_profile.clone(),
timeouts: payload.timeouts.clone(),
};
let report_context = augment_sync_report_context(
payload.report_context,
&plan.headers,
&provider_request_body_value,
)?;
Ok(Some(LocalSyncPlanAndReport {
plan,
report_kind: payload.report_kind,
report_context,
}))
}
pub(crate) fn build_standard_stream_plan_from_decision(
_parts: &http::request::Parts,
_body_json: &serde_json::Value,
payload: GatewayControlSyncDecisionResponse,
_inject_stream_flag: bool,
) -> Result<Option<LocalStreamPlanAndReport>, GatewayError> {
if generic_decision_missing_exact_provider_request(&payload) {
return Ok(None);
}
let Some(request_id) = payload
.request_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(provider_id) = payload
.provider_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(endpoint_id) = payload
.endpoint_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(key_id) = payload
.key_id
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(url) = payload
.upstream_url
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let auth_header = payload
.auth_header
.clone()
.filter(|value| !value.trim().is_empty());
let auth_value = payload
.auth_value
.clone()
.filter(|value| !value.trim().is_empty());
if auth_header.is_some() != auth_value.is_some() {
return Ok(None);
}
let Some(provider_api_format) = payload
.provider_api_format
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(client_api_format) = payload
.client_api_format
.clone()
.filter(|value| !value.trim().is_empty())
else {
return Ok(None);
};
let Some(provider_request_body_value) = payload.provider_request_body.clone() else {
return Ok(None);
};
let envelope_name = payload
.report_context
.as_ref()
.and_then(|context| context.get("envelope_name"))
.and_then(serde_json::Value::as_str);
let mut provider_request_headers = payload.provider_request_headers.clone();
if let (Some(auth_header), Some(auth_value)) = (auth_header.as_deref(), auth_value.as_deref()) {
ensure_upstream_auth_header(&mut provider_request_headers, auth_header, auth_value);
}
if provider_adaptation_requires_eventstream_accept(envelope_name, provider_api_format.as_str())
{
provider_request_headers
.entry("accept".to_string())
.or_insert_with(|| "application/vnd.amazon.eventstream".to_string());
} else {
provider_request_headers.insert("accept".to_string(), "text/event-stream".to_string());
}
let plan = ExecutionPlan {
request_id,
candidate_id: payload.candidate_id.clone(),
provider_name: payload.provider_name.clone(),
provider_id,
endpoint_id,
key_id,
method: "POST".to_string(),
url,
headers: std::mem::take(&mut provider_request_headers),
content_type: payload
.content_type
.clone()
.or_else(|| Some("application/json".to_string())),
content_encoding: None,
body: RequestBody::from_json(provider_request_body_value.clone()),
stream: true,
client_api_format,
provider_api_format,
model_name: payload.model_name.clone(),
proxy: payload.proxy.clone(),
tls_profile: payload.tls_profile.clone(),
timeouts: payload.timeouts.clone(),
};
let report_context = augment_sync_report_context(
payload.report_context,
&plan.headers,
&provider_request_body_value,
)?;
Ok(Some(LocalStreamPlanAndReport {
plan,
report_kind: payload.report_kind,
report_context,
}))
}