Merge remote-tracking branch 'origin/main' into codex/provider-policy-hardening

This commit is contained in:
elky
2026-09-05 16:21:10 +08:00
1228 changed files with 241230 additions and 31814 deletions
@@ -505,7 +505,7 @@ fn projects_uuid_prompt_cache_identity_into_missing_session_headers() {
assert_eq!(headers.get("x-client-request-id"), None);
assert_eq!(
headers.get("user-agent"),
Some(&"codex_cli_rs/0.144.1".to_string())
Some(&"codex_cli_rs/0.153.3".to_string())
);
assert_eq!(headers.get("originator"), Some(&"codex_cli_rs".to_string()));
assert!(!headers.contains_key("version"));
@@ -615,7 +615,7 @@ fn injects_only_codex_client_headers_for_images_requests() {
);
assert_eq!(
headers.get("user-agent"),
Some(&"codex_cli_rs/0.144.1".to_string())
Some(&"codex_cli_rs/0.153.3".to_string())
);
assert_eq!(headers.get("originator"), Some(&"codex_cli_rs".to_string()));
assert!(!headers.contains_key("version"));
@@ -699,7 +699,7 @@ fn preserves_client_context_headers_and_enforces_codex_provider_identity() {
);
assert_eq!(
headers.get("user-agent"),
Some(&"codex_cli_rs/0.144.1".to_string())
Some(&"codex_cli_rs/0.153.3".to_string())
);
assert_eq!(headers.get("originator"), Some(&"codex_cli_rs".to_string()));
assert_eq!(
@@ -763,7 +763,7 @@ fn compact_projects_uuid_prompt_cache_identity_into_session_headers() {
assert_eq!(headers.get("x-client-request-id"), None);
assert_eq!(
headers.get("user-agent"),
Some(&"codex_cli_rs/0.144.1".to_string())
Some(&"codex_cli_rs/0.153.3".to_string())
);
assert_eq!(headers.get("originator"), Some(&"codex_cli_rs".to_string()));
assert!(!headers.contains_key("version"));
@@ -15,11 +15,25 @@ pub(crate) fn is_deepseek_provider(provider_type: &str, base_url: &str) -> bool
host == "deepseek.com" || host.ends_with(".deepseek.com")
}
fn is_deepseek_model(provider_model: &str) -> bool {
let provider_model = provider_model.trim().to_ascii_lowercase();
let leaf = provider_model
.rsplit(['/', ':'])
.next()
.unwrap_or(provider_model.as_str());
leaf == "deepseek" || leaf.starts_with("deepseek-") || leaf.starts_with("deepseek_")
}
fn is_deepseek_upstream(provider_type: &str, base_url: &str, provider_model: &str) -> bool {
is_deepseek_provider(provider_type, base_url) || is_deepseek_model(provider_model)
}
pub(crate) fn openai_responses_reasoning_replay_policy(
provider_type: &str,
base_url: &str,
provider_model: &str,
) -> crate::ai_serving::OpenAiResponsesReasoningReplayPolicy {
if is_deepseek_provider(provider_type, base_url) {
if is_deepseek_upstream(provider_type, base_url, provider_model) {
crate::ai_serving::OpenAiResponsesReasoningReplayPolicy::DeepSeekOpaque
} else {
crate::ai_serving::OpenAiResponsesReasoningReplayPolicy::OpenAiItemIds
@@ -33,7 +47,11 @@ pub(crate) fn apply_deepseek_tool_call_thinking_compat(
provider_api_format: &str,
original_request_body: Option<&Value>,
) {
if !is_deepseek_provider(provider_type, base_url) {
let provider_model = provider_request_body
.get("model")
.and_then(Value::as_str)
.unwrap_or_default();
if !is_deepseek_upstream(provider_type, base_url, provider_model) {
return;
}
@@ -302,11 +320,35 @@ mod tests {
));
assert!(!is_deepseek_provider("custom", "ftp://api.deepseek.com/v1"));
assert_eq!(
openai_responses_reasoning_replay_policy("custom", "https://api.deepseek.com/v1"),
openai_responses_reasoning_replay_policy(
"custom",
"https://api.deepseek.com/v1",
"deepseek-v4-flash",
),
crate::ai_serving::OpenAiResponsesReasoningReplayPolicy::DeepSeekOpaque
);
assert_eq!(
openai_responses_reasoning_replay_policy("openai", "https://api.openai.com/v1"),
openai_responses_reasoning_replay_policy(
"openai",
"https://api.openai.com/v1",
"gpt-5.6-sol",
),
crate::ai_serving::OpenAiResponsesReasoningReplayPolicy::OpenAiItemIds
);
assert_eq!(
openai_responses_reasoning_replay_policy(
"custom",
"https://api.b.ai/v1",
"deepseek-v4-flash",
),
crate::ai_serving::OpenAiResponsesReasoningReplayPolicy::DeepSeekOpaque
);
assert_eq!(
openai_responses_reasoning_replay_policy(
"custom",
"https://api.b.ai/v1",
"not-deepseek-compatible",
),
crate::ai_serving::OpenAiResponsesReasoningReplayPolicy::OpenAiItemIds
);
}
@@ -330,8 +372,11 @@ mod tests {
"input": reasoning_items.clone(),
"future_request_field": {"preserve": true}
});
let replay_policy =
openai_responses_reasoning_replay_policy("custom", "https://api.deepseek.com/v1");
let replay_policy = openai_responses_reasoning_replay_policy(
"custom",
"https://api.deepseek.com/v1",
"deepseek-v4-flash",
);
let mut provider_body = crate::ai_serving::build_standard_request_body_with_model_directives_and_request_headers_and_reasoning_replay_policy(
&request,
"openai:responses",
@@ -373,7 +418,11 @@ mod tests {
crate::ai_serving::strip_incompatible_openai_responses_reasoning_items_with_policy(
&mut deepseek,
"openai:responses",
openai_responses_reasoning_replay_policy("custom", "https://api.deepseek.com/v1"),
openai_responses_reasoning_replay_policy(
"custom",
"https://api.deepseek.com/v1",
"deepseek-v4-flash",
),
),
0
);
@@ -383,7 +432,11 @@ mod tests {
crate::ai_serving::strip_incompatible_openai_responses_reasoning_items_with_policy(
&mut openai,
"openai:responses",
openai_responses_reasoning_replay_policy("openai", "https://api.openai.com/v1"),
openai_responses_reasoning_replay_policy(
"openai",
"https://api.openai.com/v1",
"gpt-5.6-sol",
),
),
66
);
@@ -417,6 +470,52 @@ mod tests {
assert_eq!(body["messages"][1]["reasoning_content"], "");
}
#[test]
fn custom_relay_deepseek_model_adds_chat_thinking_compat() {
let mut body = json!({
"model": "deepseek-v4-flash",
"messages": [
{"role": "user", "content": "inspect the repository"},
{"role": "assistant", "content": null, "tool_calls": [{
"id": "call_1",
"type": "function",
"function": {"name": "inspect", "arguments": "{}"}
}]},
{"role": "tool", "tool_call_id": "call_1", "content": "done"}
]
});
apply_deepseek_tool_call_thinking_compat(
&mut body,
"custom",
"https://api.b.ai/v1",
"openai:chat",
None,
);
assert_eq!(body["thinking"]["type"], "enabled");
assert_eq!(body["messages"][1]["reasoning_content"], "");
}
#[test]
fn custom_relay_non_deepseek_model_is_not_rewritten() {
let original = json!({
"model": "not-deepseek-compatible",
"messages": [{"role": "assistant", "content": "done"}]
});
let mut body = original.clone();
apply_deepseek_tool_call_thinking_compat(
&mut body,
"custom",
"https://api.b.ai/v1",
"openai:chat",
None,
);
assert_eq!(body, original);
}
#[test]
fn openai_chat_deepseek_honors_disabled_thinking() {
let original = json!({"reasoning_effort": "none"});
@@ -18,6 +18,7 @@ use crate::ai_serving::planner::spec_metadata::{
};
use crate::ai_serving::GatewayControlDecision;
use crate::{AiExecutionDecision, AppState, GatewayError};
use aether_routing_core::RoutingExecutionPolicy;
use super::candidates::{
build_local_standard_candidate_attempt_source, resolve_local_standard_decision_input,
@@ -177,6 +178,13 @@ pub(crate) async fn build_local_stream_attempt_source<'a>(
#[async_trait]
impl LocalExecutionAttemptSource<AiSyncAttempt> for LocalStandardSyncAttemptSource<'_> {
fn routing_execution_policy(&self) -> Option<RoutingExecutionPolicy> {
self.input
.routing_policy
.as_ref()
.map(|policy| policy.execution_policy)
}
async fn next_execution_attempt(&mut self) -> Result<Option<AiSyncAttempt>, GatewayError> {
while let Some(attempt) = self.candidates.next_attempt().await? {
match self.build_sync_attempt(attempt).await? {
@@ -220,6 +228,13 @@ impl LocalExecutionAttemptSource<AiSyncAttempt> for LocalStandardSyncAttemptSour
#[async_trait]
impl LocalExecutionAttemptSource<AiStreamAttempt> for LocalStandardStreamAttemptSource<'_> {
fn routing_execution_policy(&self) -> Option<RoutingExecutionPolicy> {
self.input
.routing_policy
.as_ref()
.map(|policy| policy.execution_policy)
}
async fn next_execution_attempt(&mut self) -> Result<Option<AiStreamAttempt>, GatewayError> {
while let Some(attempt) = self.candidates.next_attempt().await? {
match self.build_stream_attempt(attempt).await? {
@@ -142,6 +142,7 @@ pub(super) async fn maybe_build_local_standard_decision_payload_for_candidate(
client_session_affinity: input.client_session_affinity.as_ref(),
routing_policy: input.routing_policy.as_ref(),
scheduler_affinity_epoch: eligible.orchestration.scheduler_affinity_epoch,
sticky_key_attempts: eligible.orchestration.sticky_key_attempts,
client_requested_stream: body_json
.get("stream")
.and_then(serde_json::Value::as_bool)
@@ -4,6 +4,10 @@ use std::sync::Arc;
use aether_contracts::ResolvedTransportProfile;
use serde_json::Value;
use crate::ai_serving::planner::antigravity::{
build_antigravity_v1internal_provider_request, AntigravityV1InternalRequestError,
AntigravityV1InternalRequestInput, ANTIGRAVITY_V1INTERNAL_ENVELOPE_NAME,
};
use crate::ai_serving::planner::candidate_preparation::{
prepare_header_authenticated_candidate, prepare_header_authenticated_candidate_from_auth,
OauthPreparationContext,
@@ -26,6 +30,7 @@ use crate::ai_serving::planner::standard::{
openai_provider_request_contract_failure_extra_data, openai_responses_reasoning_replay_policy,
request_body_build_failure_extra_data, request_conversion_failure_extra_data,
};
use crate::ai_serving::transport::antigravity::is_antigravity_provider_transport;
use crate::ai_serving::transport::kiro::{
build_kiro_provider_headers, build_kiro_provider_request_body,
is_kiro_claude_messages_transport, KiroProviderHeadersInput, KiroRequestAuth,
@@ -587,7 +592,7 @@ pub(crate) async fn resolve_local_standard_candidate_payload_parts(
}
};
crate::ai_serving::hydrate_openai_response_history(
state.runtime_state(),
state,
body_json,
spec_metadata.api_format,
provider_api_format,
@@ -597,6 +602,7 @@ pub(crate) async fn resolve_local_standard_candidate_payload_parts(
let reasoning_replay_policy = openai_responses_reasoning_replay_policy(
transport.provider.provider_type.as_str(),
transport.endpoint.base_url.as_str(),
prepared_candidate.mapped_model.as_str(),
);
let redaction = resolve_provider_chat_pii_redaction(
state,
@@ -836,6 +842,29 @@ pub(crate) async fn resolve_local_standard_candidate_payload_parts(
.await);
}
if normalized_provider_api_format == "gemini:generate_content"
&& is_antigravity_provider_transport(transport)
{
return Ok(build_antigravity_cross_format_payload_parts(
state,
parts,
trace_id,
body_json,
input,
attempt,
transport,
spec_metadata.api_format,
provider_api_format,
prepared_candidate.mapped_model,
prepared_candidate.auth_header,
prepared_candidate.auth_value,
provider_request_body,
upstream_is_stream,
redaction.redacted,
)
.await);
}
if normalized_provider_api_format == "gemini:generate_content"
&& is_gemini_cli_provider_transport(transport)
{
@@ -962,6 +991,145 @@ fn apply_transport_request_body_semantics(
)
}
#[allow(clippy::too_many_arguments)]
async fn build_antigravity_cross_format_payload_parts(
state: &AppState,
parts: &http::request::Parts,
trace_id: &str,
original_body_json: &serde_json::Value,
input: &LocalStandardDecisionInput,
attempt: &LocalStandardCandidateAttempt,
transport: &Arc<GatewayProviderTransportSnapshot>,
client_api_format: &str,
provider_api_format: &str,
mapped_model: String,
auth_header: String,
auth_value: String,
gemini_request_body: Value,
upstream_is_stream: bool,
request_redacted: bool,
) -> Option<LocalStandardCandidatePayloadParts> {
let candidate = &attempt.eligible.candidate;
let effective_headers = input.effective_headers(&parts.headers);
let resolved =
match build_antigravity_v1internal_provider_request(AntigravityV1InternalRequestInput {
state,
parts,
transport,
trace_id,
mapped_model: &mapped_model,
provider_api_format,
auth_header: &auth_header,
auth_value: &auth_value,
request_headers: effective_headers,
original_request_body: original_body_json,
gemini_request_body: &gemini_request_body,
upstream_is_stream,
same_format: false,
})
.await
{
Ok(resolved) => resolved,
Err(AntigravityV1InternalRequestError::TransportUnsupported) => {
mark_skipped_local_standard_candidate(
state,
input,
trace_id,
candidate,
attempt.candidate_index,
&attempt.candidate_id,
"transport_unsupported",
)
.await;
return None;
}
Err(AntigravityV1InternalRequestError::EnvelopeUnsupported) => {
mark_skipped_local_standard_candidate_with_extra_data(
state,
input,
trace_id,
candidate,
attempt.candidate_index,
&attempt.candidate_id,
"provider_request_body_build_failed",
request_body_build_failure_extra_data(
original_body_json,
client_api_format,
provider_api_format,
),
)
.await;
return None;
}
Err(AntigravityV1InternalRequestError::UpstreamUrlUnavailable) => {
mark_skipped_local_standard_candidate_with_failure_diagnostic(
state,
input,
trace_id,
candidate,
attempt.candidate_index,
&attempt.candidate_id,
"upstream_url_missing",
CandidateFailureDiagnostic::upstream_url_missing(
client_api_format,
provider_api_format,
"standard_family_antigravity_url",
),
)
.await;
return None;
}
Err(AntigravityV1InternalRequestError::HeaderRulesApplyFailed) => {
mark_skipped_local_standard_candidate_with_failure_diagnostic(
state,
input,
trace_id,
candidate,
attempt.candidate_index,
&attempt.candidate_id,
"transport_header_rules_apply_failed",
CandidateFailureDiagnostic::header_rules_apply_failed(
client_api_format,
provider_api_format,
"standard_family_antigravity_headers",
),
)
.await;
return None;
}
};
let mut provider_request_headers = resolved.headers.headers;
apply_codex_openai_special_headers(
&mut provider_request_headers,
&resolved.body,
effective_headers,
resolved.transport.provider.provider_type.as_str(),
provider_api_format,
Some(trace_id),
resolved.transport.key.decrypted_auth_config.as_deref(),
);
request_identity_response_encoding_when_redacted(
&mut provider_request_headers,
request_redacted,
);
Some(LocalStandardCandidatePayloadParts {
auth_header: resolved.headers.auth_header,
auth_value: resolved.headers.auth_value,
mapped_model,
provider_api_format: provider_api_format.to_string(),
provider_request_body: resolved.body,
provider_request_headers,
upstream_url: resolved.upstream_url,
upstream_is_stream,
envelope_name: Some(ANTIGRAVITY_V1INTERNAL_ENVELOPE_NAME),
transport: resolved.transport,
transport_profile: None,
request_redacted,
})
}
#[allow(clippy::too_many_arguments)]
async fn build_gemini_cli_cross_format_payload_parts(
state: &AppState,
@@ -195,6 +195,7 @@ pub(crate) async fn maybe_build_local_openai_chat_decision_payload_for_candidate
client_session_affinity: input.client_session_affinity.as_ref(),
routing_policy: input.routing_policy.as_ref(),
scheduler_affinity_epoch: eligible.orchestration.scheduler_affinity_epoch,
sticky_key_attempts: eligible.orchestration.sticky_key_attempts,
client_requested_stream: body_json
.get("stream")
.and_then(serde_json::Value::as_bool)
@@ -159,6 +159,7 @@ fn finalize_openai_chat_provider_request_body(
openai_responses_reasoning_replay_policy(
transport.provider.provider_type.as_str(),
transport.endpoint.base_url.as_str(),
mapped_model,
),
)
.err()
@@ -2740,7 +2741,7 @@ mod tests {
.provider_request_headers
.get("x-client-version")
.map(String::as_str),
Some("1.2.3")
Some("4.3.0")
);
assert_eq!(
payload
@@ -2760,7 +2761,7 @@ mod tests {
assert_eq!(payload.provider_request_body["model"], "gemini-2.5-pro");
assert_eq!(
payload.provider_request_body["userAgent"],
"antigravity/cli/1.0.16 (aidev_client; os_type=linux; arch=arm64; auth_method=consumer)"
"vscode/1.X.X (Antigravity/4.3.0)"
);
assert_eq!(payload.provider_request_body["requestType"], "agent");
assert!(payload.provider_request_body.get("contents").is_none());
@@ -1,3 +1,4 @@
use aether_routing_core::RoutingExecutionPolicy;
use async_trait::async_trait;
use std::collections::VecDeque;
use tracing::warn;
@@ -119,6 +120,13 @@ pub(crate) async fn build_local_openai_chat_stream_attempt_source<'a>(
#[async_trait]
impl LocalExecutionAttemptSource<AiStreamAttempt> for LocalOpenAiChatStreamAttemptSource<'_> {
fn routing_execution_policy(&self) -> Option<RoutingExecutionPolicy> {
self.input
.routing_policy
.as_ref()
.map(|policy| policy.execution_policy)
}
async fn next_execution_attempt(&mut self) -> Result<Option<AiStreamAttempt>, GatewayError> {
let select_started_at = std::time::Instant::now();
let selected = self.next_execution_attempt_with_target_select().await?;
@@ -1,3 +1,4 @@
use aether_routing_core::RoutingExecutionPolicy;
use async_trait::async_trait;
use tracing::warn;
@@ -92,6 +93,13 @@ pub(crate) async fn build_local_openai_chat_sync_attempt_source<'a>(
#[async_trait]
impl LocalExecutionAttemptSource<AiSyncAttempt> for LocalOpenAiChatSyncAttemptSource<'_> {
fn routing_execution_policy(&self) -> Option<RoutingExecutionPolicy> {
self.input
.routing_policy
.as_ref()
.map(|policy| policy.execution_policy)
}
async fn next_execution_attempt(&mut self) -> Result<Option<AiSyncAttempt>, GatewayError> {
while let Some(attempt) = self.candidates.next_attempt().await? {
match self.build_sync_attempt(attempt).await? {
@@ -1,7 +1,6 @@
use std::collections::BTreeMap;
use aether_contracts::RequestBody;
use tracing::debug;
use super::super::{
augment_sync_report_context, build_ai_execution_plan_from_decision,
@@ -10,7 +9,6 @@ use super::super::{
AiStreamAttempt,
};
use crate::ai_serving::planner::common::enforce_provider_body_stream_policy;
use crate::ai_serving::planner::redaction::sanitize_upstream_url_for_log;
use crate::ai_serving::provider_adaptation_requires_eventstream_accept;
use crate::ai_serving::transport::{
build_standard_plan_fallback_headers, build_standard_plan_fallback_openai_chat_url,
@@ -157,21 +155,16 @@ pub(crate) fn build_openai_responses_stream_plan_from_decision(
let Some(auth_pair) = take_ai_upstream_auth_pair(&mut payload) else {
return Ok(None);
};
let (url, url_source) = if let Some(upstream_url) =
take_non_empty_string(&mut payload.upstream_url)
{
(upstream_url, "upstream_url")
let url = if let Some(upstream_url) = take_non_empty_string(&mut payload.upstream_url) {
upstream_url
} else {
let Some(upstream_base_url) = take_non_empty_string(&mut payload.upstream_base_url) else {
return Ok(None);
};
(
build_standard_plan_fallback_openai_responses_url(
&upstream_base_url,
parts.uri.query(),
compact,
),
"upstream_base_url",
build_standard_plan_fallback_openai_responses_url(
&upstream_base_url,
parts.uri.query(),
compact,
)
};
let Some(provider_request_body_value) = payload.provider_request_body.take() else {
@@ -238,16 +231,7 @@ pub(crate) fn build_openai_responses_stream_plan_from_decision(
.uri
.query()
.and_then(crate::ai_serving::api::sanitize_request_query_string);
let log_decision_upstream_base_url = payload
.upstream_base_url
.as_deref()
.map(sanitize_upstream_url_for_log);
let log_decision_upstream_url = payload
.upstream_url
.as_deref()
.map(sanitize_upstream_url_for_log);
let log_plan_url = sanitize_upstream_url_for_log(plan.url.as_str());
debug!(
tracing::debug!(
event_name = "local_openai_responses_stream_plan_built",
log_type = "debug",
request_id = %plan.request_id,
@@ -255,12 +239,13 @@ pub(crate) fn build_openai_responses_stream_plan_from_decision(
provider_id = %plan.provider_id,
endpoint_id = %plan.endpoint_id,
key_id = %plan.key_id,
downstream_path_and_query = %crate::ai_serving::pure::sanitize_request_path_and_query(
parts.uri.path(),
parts.uri.query(),
).unwrap_or_else(|| "/".to_string()),
upstream_origin = %crate::handlers::shared::security_log_url_origin(&plan.url),
downstream_path = %parts.uri.path(),
downstream_query = ?log_downstream_query,
url_source,
decision_upstream_base_url = ?log_decision_upstream_base_url,
decision_upstream_url = ?log_decision_upstream_url,
plan_url = %log_plan_url,
client_api_format = %plan.client_api_format,
provider_api_format = %plan.provider_api_format,
upstream_is_stream = effective_upstream_is_stream,
@@ -1,7 +1,6 @@
use std::collections::BTreeMap;
use aether_contracts::RequestBody;
use tracing::debug;
use super::super::{
augment_sync_report_context, build_ai_execution_plan_from_decision,
@@ -10,7 +9,6 @@ use super::super::{
AiSyncAttempt,
};
use crate::ai_serving::planner::common::enforce_provider_body_stream_policy;
use crate::ai_serving::planner::redaction::sanitize_upstream_url_for_log;
use crate::ai_serving::transport::{
build_standard_plan_fallback_headers, build_standard_plan_fallback_openai_chat_url,
build_standard_plan_fallback_openai_responses_url, StandardPlanFallbackAcceptPolicy,
@@ -142,21 +140,16 @@ pub(crate) fn build_openai_responses_sync_plan_from_decision(
let Some(auth_pair) = take_ai_upstream_auth_pair(&mut payload) else {
return Ok(None);
};
let (url, url_source) = if let Some(upstream_url) =
take_non_empty_string(&mut payload.upstream_url)
{
(upstream_url, "upstream_url")
let url = if let Some(upstream_url) = take_non_empty_string(&mut payload.upstream_url) {
upstream_url
} else {
let Some(upstream_base_url) = take_non_empty_string(&mut payload.upstream_base_url) else {
return Ok(None);
};
(
build_standard_plan_fallback_openai_responses_url(
&upstream_base_url,
parts.uri.query(),
compact,
),
"upstream_base_url",
build_standard_plan_fallback_openai_responses_url(
&upstream_base_url,
parts.uri.query(),
compact,
)
};
let Some(provider_request_body_value) = payload.provider_request_body.take() else {
@@ -205,16 +198,7 @@ pub(crate) fn build_openai_responses_sync_plan_from_decision(
.uri
.query()
.and_then(crate::ai_serving::api::sanitize_request_query_string);
let log_decision_upstream_base_url = payload
.upstream_base_url
.as_deref()
.map(sanitize_upstream_url_for_log);
let log_decision_upstream_url = payload
.upstream_url
.as_deref()
.map(sanitize_upstream_url_for_log);
let log_plan_url = sanitize_upstream_url_for_log(plan.url.as_str());
debug!(
tracing::debug!(
event_name = "local_openai_responses_sync_plan_built",
log_type = "debug",
request_id = %plan.request_id,
@@ -222,12 +206,13 @@ pub(crate) fn build_openai_responses_sync_plan_from_decision(
provider_id = %plan.provider_id,
endpoint_id = %plan.endpoint_id,
key_id = %plan.key_id,
downstream_path_and_query = %crate::ai_serving::pure::sanitize_request_path_and_query(
parts.uri.path(),
parts.uri.query(),
).unwrap_or_else(|| "/".to_string()),
upstream_origin = %crate::handlers::shared::security_log_url_origin(&plan.url),
downstream_path = %parts.uri.path(),
downstream_query = ?log_downstream_query,
url_source,
decision_upstream_base_url = ?log_decision_upstream_base_url,
decision_upstream_url = ?log_decision_upstream_url,
plan_url = %log_plan_url,
client_api_format = %plan.client_api_format,
provider_api_format = %plan.provider_api_format,
upstream_is_stream = payload.upstream_is_stream,
@@ -3,7 +3,6 @@ use tracing::debug;
use crate::ai_serving::build_request_trace_proxy_value;
use crate::ai_serving::planner::decision_input::apply_provider_request_routing_policy_to_decision_with_websocket_mode;
use crate::ai_serving::planner::redaction::sanitize_upstream_url_for_log;
use crate::ai_serving::planner::report_context::{
build_local_execution_report_context, insert_native_client_envelope_name,
insert_provider_stream_event_api_format, LocalExecutionReportContextParts,
@@ -184,6 +183,7 @@ pub(crate) async fn maybe_build_local_openai_responses_decision_payload_for_cand
client_session_affinity: input.client_session_affinity.as_ref(),
routing_policy: input.routing_policy.as_ref(),
scheduler_affinity_epoch: eligible.orchestration.scheduler_affinity_epoch,
sticky_key_attempts: eligible.orchestration.sticky_key_attempts,
client_requested_stream: body_json
.get("stream")
.and_then(serde_json::Value::as_bool)
@@ -204,12 +204,10 @@ pub(crate) async fn maybe_build_local_openai_responses_decision_payload_for_cand
&resolved.transport,
);
let log_base_url = sanitize_upstream_url_for_log(resolved.transport.endpoint.base_url.as_str());
let log_request_query = parts
.uri
.query()
.and_then(crate::ai_serving::api::sanitize_request_query_string);
let log_upstream_url = sanitize_upstream_url_for_log(resolved.upstream_url.as_str());
debug!(
event_name = "local_openai_responses_decision_payload_built",
log_type = "debug",
@@ -226,9 +224,12 @@ pub(crate) async fn maybe_build_local_openai_responses_decision_payload_for_cand
client_api_format = spec_metadata.api_format,
provider_api_format = %resolved.provider_api_format,
request_path = %parts.uri.path(),
request_path_and_query = %crate::ai_serving::pure::sanitize_request_path_and_query(
parts.uri.path(),
parts.uri.query(),
).unwrap_or_else(|| "/".to_string()),
upstream_origin = %crate::handlers::shared::security_log_url_origin(&resolved.upstream_url),
request_query = ?log_request_query,
upstream_base_url = %log_base_url,
upstream_url = %log_upstream_url,
upstream_is_stream = resolved.upstream_is_stream,
has_envelope = resolved.envelope_name.is_some(),
"gateway built local openai responses decision payload"
@@ -24,7 +24,6 @@ use crate::ai_serving::planner::gemini_cli::{
};
use crate::ai_serving::planner::redaction::{
request_identity_response_encoding_when_redacted, resolve_provider_chat_pii_redaction,
sanitize_upstream_url_for_log,
};
use crate::ai_serving::planner::spec_metadata::local_openai_responses_spec_metadata;
use crate::ai_serving::planner::standard::{
@@ -428,7 +427,7 @@ pub(crate) async fn resolve_local_openai_responses_candidate_payload_parts_with_
}
};
crate::ai_serving::hydrate_openai_response_history(
state.runtime_state(),
state,
body_json,
spec_metadata.api_format,
provider_api_format,
@@ -438,6 +437,7 @@ pub(crate) async fn resolve_local_openai_responses_candidate_payload_parts_with_
let reasoning_replay_policy = openai_responses_reasoning_replay_policy(
transport.provider.provider_type.as_str(),
transport.endpoint.base_url.as_str(),
mapped_model.as_str(),
);
let redaction = resolve_provider_chat_pii_redaction(
state,
@@ -866,17 +866,10 @@ pub(crate) async fn resolve_local_openai_responses_candidate_payload_parts_with_
let (execution_strategy, conversion_mode) =
ai_local_execution_contract_for_formats(spec_metadata.api_format, provider_api_format);
let log_base_url = sanitize_upstream_url_for_log(transport.endpoint.base_url.as_str());
let log_custom_path = transport
.endpoint
.custom_path
.as_deref()
.map(sanitize_upstream_url_for_log);
let log_request_query = parts
.uri
.query()
.and_then(crate::ai_serving::api::sanitize_request_query_string);
let log_upstream_url = sanitize_upstream_url_for_log(upstream_url.as_str());
debug!(
event_name = "local_openai_responses_upstream_url_resolved",
@@ -892,12 +885,14 @@ pub(crate) async fn resolve_local_openai_responses_candidate_payload_parts_with_
provider_api_format = %provider_api_format,
execution_strategy = execution_strategy.as_str(),
conversion_mode = conversion_mode.as_str(),
base_url = %log_base_url,
custom_path = ?log_custom_path,
request_path_and_query = %crate::ai_serving::pure::sanitize_request_path_and_query(
parts.uri.path(),
parts.uri.query(),
).unwrap_or_else(|| "/".to_string()),
upstream_origin = %crate::handlers::shared::security_log_url_origin(&upstream_url),
request_path = %parts.uri.path(),
request_query = ?log_request_query,
mapped_model = %mapped_model,
upstream_url = %log_upstream_url,
upstream_is_stream,
"gateway resolved local openai responses upstream url"
);
@@ -2010,8 +2005,6 @@ async fn build_kiro_openai_responses_payload_parts(
};
let (execution_strategy, conversion_mode) =
ai_local_execution_contract_for_formats(client_api_format, provider_api_format);
let log_upstream_url = sanitize_upstream_url_for_log(upstream_url.as_str());
debug!(
event_name = "local_openai_responses_kiro_upstream_url_resolved",
log_type = "debug",
@@ -2026,7 +2019,7 @@ async fn build_kiro_openai_responses_payload_parts(
provider_api_format = %provider_api_format,
execution_strategy = execution_strategy.as_str(),
conversion_mode = conversion_mode.as_str(),
upstream_url = %log_upstream_url,
upstream_origin = %crate::handlers::shared::security_log_url_origin(&upstream_url),
upstream_is_stream,
"gateway resolved local openai responses kiro upstream url"
);
@@ -1005,6 +1005,7 @@ pub(crate) async fn maybe_build_responses_websocket_decision(
reasoning_replay_policy: openai_responses_reasoning_replay_policy(
transport.provider.provider_type.as_str(),
transport.endpoint.base_url.as_str(),
mapped_model.as_str(),
),
model_directive_patch: input
.model_directive_policy
@@ -1,3 +1,4 @@
use aether_routing_core::RoutingExecutionPolicy;
use async_trait::async_trait;
use tracing::warn;
@@ -161,6 +162,13 @@ pub(super) async fn build_local_stream_attempt_source<'a>(
#[async_trait]
impl LocalExecutionAttemptSource<AiSyncAttempt> for LocalOpenAiResponsesSyncAttemptSource<'_> {
fn routing_execution_policy(&self) -> Option<RoutingExecutionPolicy> {
self.input
.routing_policy
.as_ref()
.map(|policy| policy.execution_policy)
}
async fn next_execution_attempt(&mut self) -> Result<Option<AiSyncAttempt>, GatewayError> {
while let Some(attempt) = self.candidates.next_attempt().await? {
match self.build_sync_attempt(attempt).await? {
@@ -204,6 +212,13 @@ impl LocalExecutionAttemptSource<AiSyncAttempt> for LocalOpenAiResponsesSyncAtte
#[async_trait]
impl LocalExecutionAttemptSource<AiStreamAttempt> for LocalOpenAiResponsesStreamAttemptSource<'_> {
fn routing_execution_policy(&self) -> Option<RoutingExecutionPolicy> {
self.input
.routing_policy
.as_ref()
.map(|policy| policy.execution_policy)
}
async fn next_execution_attempt(&mut self) -> Result<Option<AiStreamAttempt>, GatewayError> {
while let Some(attempt) = self.candidates.next_attempt().await? {
match self.build_stream_attempt(attempt).await? {