mirror of
https://github.com/fawney19/Aether.git
synced 2026-10-04 08:27:46 +08:00
feat(codex): align Search and execution protocol
This commit is contained in:
@@ -104,6 +104,7 @@ fn is_known_admin_monitoring_api_format(value: &str) -> bool {
|
||||
"openai:chat"
|
||||
| "openai:responses"
|
||||
| "openai:responses:compact"
|
||||
| "openai:search"
|
||||
| "openai:image"
|
||||
| "openai:video"
|
||||
| "openai:embedding"
|
||||
@@ -492,6 +493,20 @@ mod tests {
|
||||
);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn parses_v2_scheduler_affinity_key_for_openai_search() {
|
||||
let parsed = parse_admin_monitoring_scheduler_affinity_key(
|
||||
"scheduler_affinity:v2:user-key-1:openai:search:gpt-5.6-sol:codex:sessionhash",
|
||||
)
|
||||
.expect("Search scheduler key should parse");
|
||||
|
||||
assert_eq!(parsed.affinity_key, "user-key-1");
|
||||
assert_eq!(parsed.api_format, "openai:search");
|
||||
assert_eq!(parsed.model_name, "gpt-5.6-sol");
|
||||
assert_eq!(parsed.client_family.as_deref(), Some("codex"));
|
||||
assert_eq!(parsed.session_hash.as_deref(), Some("sessionhash"));
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn parses_v2_scheduler_affinity_key_with_three_segment_api_format() {
|
||||
let parsed = parse_admin_monitoring_scheduler_affinity_key(
|
||||
|
||||
@@ -6,7 +6,8 @@ use crate::handlers::admin::provider::shared::payloads::{
|
||||
AdminProviderCreateRequest, AdminProviderUpdatePatch,
|
||||
};
|
||||
use crate::handlers::admin::provider::write::provider::{
|
||||
reconcile_admin_fixed_provider_template_endpoints, reconcile_admin_fixed_provider_template_keys,
|
||||
reconcile_admin_fixed_provider_template_endpoints,
|
||||
reconcile_admin_fixed_provider_template_endpoints_after_update,
|
||||
};
|
||||
use crate::handlers::admin::request::{AdminAppState, AdminRequestContext};
|
||||
use crate::handlers::admin::shared::attach_admin_audit_response;
|
||||
@@ -80,7 +81,6 @@ pub(crate) async fn maybe_build_local_admin_provider_writes_response(
|
||||
.is_some()
|
||||
{
|
||||
reconcile_admin_fixed_provider_template_endpoints(state, &created_provider).await?;
|
||||
reconcile_admin_fixed_provider_template_keys(state, &created_provider).await?;
|
||||
}
|
||||
return Ok(Some(attach_admin_audit_response(
|
||||
Json(json!({
|
||||
@@ -161,8 +161,12 @@ pub(crate) async fn maybe_build_local_admin_provider_writes_response(
|
||||
.fixed_provider_template(&updated_record.provider_type)
|
||||
.is_some()
|
||||
{
|
||||
reconcile_admin_fixed_provider_template_endpoints(state, &updated_record).await?;
|
||||
reconcile_admin_fixed_provider_template_keys(state, &updated_record).await?;
|
||||
reconcile_admin_fixed_provider_template_endpoints_after_update(
|
||||
state,
|
||||
&existing_provider,
|
||||
&updated_record,
|
||||
)
|
||||
.await?;
|
||||
}
|
||||
return Ok(Some(
|
||||
match state
|
||||
|
||||
@@ -29,10 +29,7 @@ use crate::handlers::shared::{
|
||||
provider_key_status_snapshot_payload,
|
||||
};
|
||||
use crate::model_fetch::ModelFetchRuntimeState;
|
||||
use crate::provider_key_auth::{
|
||||
provider_key_auth_semantics, provider_key_configured_api_formats,
|
||||
provider_key_inherits_provider_api_formats,
|
||||
};
|
||||
use crate::provider_key_auth::provider_key_auth_semantics;
|
||||
use crate::provider_transport::antigravity::{
|
||||
build_antigravity_safe_v1internal_request, build_antigravity_static_identity_headers,
|
||||
classify_local_antigravity_request_support, AntigravityEnvelopeRequestType,
|
||||
@@ -543,6 +540,22 @@ fn provider_query_build_test_request_body_for_api_format(
|
||||
model: &str,
|
||||
route_path: &str,
|
||||
client_api_format: &str,
|
||||
) -> Value {
|
||||
provider_query_build_test_request_body_for_api_format_with_search_session(
|
||||
payload,
|
||||
model,
|
||||
route_path,
|
||||
client_api_format,
|
||||
None,
|
||||
)
|
||||
}
|
||||
|
||||
fn provider_query_build_test_request_body_for_api_format_with_search_session(
|
||||
payload: &Value,
|
||||
model: &str,
|
||||
route_path: &str,
|
||||
client_api_format: &str,
|
||||
search_session_id: Option<&str>,
|
||||
) -> Value {
|
||||
let client_api_format = provider_query_normalize_api_format_alias(client_api_format);
|
||||
let override_custom_model = route_path.ends_with("/test-model-failover")
|
||||
@@ -568,7 +581,7 @@ fn provider_query_build_test_request_body_for_api_format(
|
||||
);
|
||||
} else if matches!(
|
||||
client_api_format.as_str(),
|
||||
"openai:responses" | "openai:responses:compact"
|
||||
"openai:responses" | "openai:responses:compact" | "openai:search"
|
||||
) && !value_has_non_empty_text(object.get("input"))
|
||||
{
|
||||
if let Some(prompt) = object
|
||||
@@ -580,7 +593,7 @@ fn provider_query_build_test_request_body_for_api_format(
|
||||
}
|
||||
if matches!(
|
||||
client_api_format.as_str(),
|
||||
"openai:responses" | "openai:responses:compact"
|
||||
"openai:responses" | "openai:responses:compact" | "openai:search"
|
||||
) && value_has_non_empty_text(object.get("input"))
|
||||
{
|
||||
object.remove("prompt");
|
||||
@@ -590,6 +603,9 @@ fn provider_query_build_test_request_body_for_api_format(
|
||||
{
|
||||
object.remove("messages");
|
||||
}
|
||||
if client_api_format == "openai:search" {
|
||||
provider_query_ensure_search_test_fields(object, payload, search_session_id);
|
||||
}
|
||||
}
|
||||
return body;
|
||||
}
|
||||
@@ -620,6 +636,15 @@ fn provider_query_build_test_request_body_for_api_format(
|
||||
"temperature": 0.7,
|
||||
"stream": true,
|
||||
}),
|
||||
"openai:search" => json!({
|
||||
"id": provider_query_search_test_session_id(search_session_id),
|
||||
"model": model,
|
||||
"input": message,
|
||||
"commands": {
|
||||
"search_query": [{"q": message}]
|
||||
},
|
||||
"max_output_tokens": 256,
|
||||
}),
|
||||
"claude:messages" => json!({
|
||||
"model": model,
|
||||
"messages": [{
|
||||
@@ -680,7 +705,7 @@ fn provider_query_insert_default_test_conversation(
|
||||
.entry("top_n".to_string())
|
||||
.or_insert_with(|| Value::from(4_u64));
|
||||
}
|
||||
"openai:responses" | "openai:responses:compact" => {
|
||||
"openai:responses" | "openai:responses:compact" | "openai:search" => {
|
||||
object.insert("input".to_string(), Value::String(message));
|
||||
}
|
||||
"claude:messages" => {
|
||||
@@ -698,6 +723,48 @@ fn provider_query_insert_default_test_conversation(
|
||||
}
|
||||
}
|
||||
|
||||
fn provider_query_search_test_session_id(search_session_id: Option<&str>) -> String {
|
||||
search_session_id
|
||||
.map(str::trim)
|
||||
.filter(|value| !value.is_empty())
|
||||
.map(|value| format!("aether-model-test-{value}"))
|
||||
.unwrap_or_else(|| format!("aether-model-test-{}", Uuid::new_v4().simple()))
|
||||
}
|
||||
|
||||
fn provider_query_ensure_search_test_fields(
|
||||
object: &mut Map<String, Value>,
|
||||
payload: &Value,
|
||||
search_session_id: Option<&str>,
|
||||
) {
|
||||
let query = object
|
||||
.get("input")
|
||||
.and_then(Value::as_str)
|
||||
.map(str::trim)
|
||||
.filter(|value| !value.is_empty())
|
||||
.map(ToOwned::to_owned)
|
||||
.or_else(|| provider_query_extract_message(payload))
|
||||
.unwrap_or_else(|| DEFAULT_PROVIDER_QUERY_TEST_MESSAGE.to_string());
|
||||
let has_session_id = object
|
||||
.get("id")
|
||||
.and_then(Value::as_str)
|
||||
.is_some_and(|value| !value.trim().is_empty());
|
||||
if !has_session_id {
|
||||
object.insert(
|
||||
"id".to_string(),
|
||||
provider_query_search_test_session_id(search_session_id).into(),
|
||||
);
|
||||
}
|
||||
object
|
||||
.entry("input".to_string())
|
||||
.or_insert_with(|| Value::String(query.clone()));
|
||||
object
|
||||
.entry("commands".to_string())
|
||||
.or_insert_with(|| json!({"search_query": [{"q": query}]}));
|
||||
object
|
||||
.entry("max_output_tokens".to_string())
|
||||
.or_insert_with(|| Value::from(256_u64));
|
||||
}
|
||||
|
||||
fn provider_query_grok_test_client_api_format(provider_api_format: &str) -> &'static str {
|
||||
match provider_query_normalize_api_format_alias(provider_api_format).as_str() {
|
||||
"openai:responses" | "openai:responses:compact" => "openai:responses",
|
||||
@@ -767,7 +834,7 @@ fn provider_query_request_body_has_conversation_for_api_format(
|
||||
client_api_format: &str,
|
||||
) -> bool {
|
||||
match provider_query_normalize_api_format_alias(client_api_format).as_str() {
|
||||
"openai:responses" | "openai:responses:compact" => {
|
||||
"openai:responses" | "openai:responses:compact" | "openai:search" => {
|
||||
value_has_non_empty_text(body.get("input"))
|
||||
|| value_has_non_empty_text(body.get("prompt"))
|
||||
}
|
||||
@@ -874,15 +941,11 @@ fn provider_query_key_supports_endpoint(
|
||||
provider_type: &str,
|
||||
endpoint_api_format: &str,
|
||||
) -> bool {
|
||||
if provider_key_inherits_provider_api_formats(key, provider_type) {
|
||||
return true;
|
||||
}
|
||||
let formats = provider_key_configured_api_formats(key);
|
||||
let endpoint_api_format = provider_query_normalize_api_format_alias(endpoint_api_format);
|
||||
formats.is_empty()
|
||||
|| formats
|
||||
.iter()
|
||||
.any(|value| provider_query_normalize_api_format_alias(value) == endpoint_api_format)
|
||||
crate::handlers::shared::provider_catalog_key_supports_format(
|
||||
key,
|
||||
provider_type,
|
||||
endpoint_api_format,
|
||||
)
|
||||
}
|
||||
|
||||
async fn provider_query_select_preferred_non_kiro_endpoint(
|
||||
@@ -1624,6 +1687,15 @@ fn provider_query_standard_execution_response_body(
|
||||
{
|
||||
return None;
|
||||
}
|
||||
if result.status_code < 400
|
||||
&& provider_query_normalize_api_format_alias(provider_api_format) == "openai:search"
|
||||
&& !body
|
||||
.get("output")
|
||||
.and_then(Value::as_str)
|
||||
.is_some_and(|value| !value.trim().is_empty())
|
||||
{
|
||||
return None;
|
||||
}
|
||||
Some(body)
|
||||
}
|
||||
|
||||
@@ -2864,12 +2936,14 @@ async fn provider_query_execute_standard_test_candidate(
|
||||
crate::ai_serving::normalize_api_format_alias(provider_api_format);
|
||||
let client_api_format =
|
||||
provider_query_standard_test_client_api_format(normalized_provider_api_format.as_str());
|
||||
let original_request_body = provider_query_build_test_request_body_for_api_format(
|
||||
payload,
|
||||
&candidate.effective_model,
|
||||
route_path,
|
||||
client_api_format,
|
||||
);
|
||||
let original_request_body =
|
||||
provider_query_build_test_request_body_for_api_format_with_search_session(
|
||||
payload,
|
||||
&candidate.effective_model,
|
||||
route_path,
|
||||
client_api_format,
|
||||
Some(trace_id),
|
||||
);
|
||||
if crate::provider_transport::is_windsurf_provider_transport(&transport)
|
||||
&& provider_query_normalize_api_format_alias(candidate.endpoint.api_format.as_str())
|
||||
== "openai:chat"
|
||||
@@ -3014,6 +3088,45 @@ async fn provider_query_execute_standard_test_candidate(
|
||||
);
|
||||
provider_request_body
|
||||
}
|
||||
"openai:search" => {
|
||||
let Some(mut provider_request_body) =
|
||||
crate::provider_transport::build_same_format_provider_request_body(
|
||||
crate::provider_transport::SameFormatProviderRequestBodyInput {
|
||||
body_json: &request_body,
|
||||
mapped_model: request_model,
|
||||
client_api_format,
|
||||
provider_api_format,
|
||||
source_model: request_body.get("model").and_then(Value::as_str),
|
||||
family: crate::provider_transport::SameFormatProviderFamily::Standard,
|
||||
body_rules: transport.endpoint.body_rules.as_ref(),
|
||||
request_headers: Some(&incoming_request_headers),
|
||||
upstream_is_stream,
|
||||
force_body_stream_field: require_body_stream_field,
|
||||
kiro_auth_config: None,
|
||||
is_claude_code: false,
|
||||
enable_model_directives: false,
|
||||
},
|
||||
)
|
||||
else {
|
||||
return Ok(provider_query_skipped_execution_outcome(
|
||||
request_body.clone(),
|
||||
format!("Provider request body could not be built for {provider_api_format}"),
|
||||
));
|
||||
};
|
||||
if let Err(err) = crate::provider_transport::apply_transport_request_body_semantics(
|
||||
&mut provider_request_body,
|
||||
&transport,
|
||||
normalized_provider_api_format.as_str(),
|
||||
) {
|
||||
return Ok(provider_query_skipped_execution_outcome(
|
||||
provider_request_body,
|
||||
format!(
|
||||
"Provider request body is not compatible with transport semantics: {err}"
|
||||
),
|
||||
));
|
||||
}
|
||||
provider_request_body
|
||||
}
|
||||
"openai:embedding"
|
||||
| "gemini:embedding"
|
||||
| "jina:embedding"
|
||||
@@ -3077,7 +3190,7 @@ async fn provider_query_execute_standard_test_candidate(
|
||||
);
|
||||
if matches!(
|
||||
normalized_provider_api_format.as_str(),
|
||||
"openai:chat" | "openai:responses" | "openai:responses:compact"
|
||||
"openai:chat" | "openai:responses" | "openai:responses:compact" | "openai:search"
|
||||
) && crate::ai_serving::finalize_openai_provider_request_with_codex_model_capabilities(
|
||||
&mut provider_request_body,
|
||||
crate::ai_serving::OpenAiProviderRequestFinalization {
|
||||
@@ -3175,6 +3288,7 @@ async fn provider_query_execute_standard_test_candidate(
|
||||
"openai:chat"
|
||||
| "openai:responses"
|
||||
| "openai:responses:compact"
|
||||
| "openai:search"
|
||||
| "claude:messages"
|
||||
| "gemini:generate_content"
|
||||
| "openai:embedding"
|
||||
@@ -3190,6 +3304,7 @@ async fn provider_query_execute_standard_test_candidate(
|
||||
"openai:chat"
|
||||
| "openai:responses"
|
||||
| "openai:responses:compact"
|
||||
| "openai:search"
|
||||
| "openai:embedding"
|
||||
| "jina:embedding"
|
||||
| "doubao:embedding"
|
||||
@@ -3255,7 +3370,7 @@ async fn provider_query_execute_standard_test_candidate(
|
||||
&BTreeMap::new(),
|
||||
Some("application/json"),
|
||||
),
|
||||
"openai:responses" | "openai:responses:compact" => {
|
||||
"openai:responses" | "openai:responses:compact" | "openai:search" => {
|
||||
crate::provider_transport::auth::build_complete_passthrough_headers_with_auth(
|
||||
&parts.headers,
|
||||
auth_header.as_deref().unwrap_or_default(),
|
||||
@@ -3321,7 +3436,9 @@ async fn provider_query_execute_standard_test_candidate(
|
||||
response_body: None,
|
||||
});
|
||||
}
|
||||
if crate::ai_serving::is_openai_responses_family_format(provider_api_format) {
|
||||
if crate::ai_serving::is_openai_responses_family_format(provider_api_format)
|
||||
|| crate::ai_serving::api_format_alias_matches(provider_api_format, "openai:search")
|
||||
{
|
||||
crate::ai_serving::apply_codex_openai_special_headers(
|
||||
&mut request_headers,
|
||||
&provider_request_body,
|
||||
|
||||
@@ -33,6 +33,8 @@ pub(super) fn provider_query_standard_test_client_api_format(
|
||||
let normalized_api_format = crate::ai_serving::normalize_api_format_alias(provider_api_format);
|
||||
if normalized_api_format == "openai:responses:compact" {
|
||||
"openai:responses:compact"
|
||||
} else if normalized_api_format == "openai:search" {
|
||||
"openai:search"
|
||||
} else if crate::ai_serving::is_embedding_api_format(&normalized_api_format) {
|
||||
"openai:embedding"
|
||||
} else if crate::ai_serving::is_rerank_api_format(&normalized_api_format) {
|
||||
@@ -71,6 +73,7 @@ pub(super) fn provider_query_standard_test_unsupported_reason(
|
||||
}
|
||||
"openai:responses"
|
||||
| "openai:responses:compact"
|
||||
| "openai:search"
|
||||
| "claude:messages"
|
||||
| "openai:embedding"
|
||||
| "jina:embedding"
|
||||
@@ -261,6 +264,7 @@ pub(super) fn provider_query_test_adapter_for_provider_api_format(
|
||||
"openai:chat"
|
||||
| "openai:responses"
|
||||
| "openai:responses:compact"
|
||||
| "openai:search"
|
||||
| "claude:messages"
|
||||
| "gemini:generate_content"
|
||||
| "gemini:interactions"
|
||||
@@ -365,6 +369,7 @@ pub(super) fn provider_query_transport_supports_model_test_execution(
|
||||
}
|
||||
"openai:responses"
|
||||
| "openai:responses:compact"
|
||||
| "openai:search"
|
||||
| "openai:embedding"
|
||||
| "jina:embedding"
|
||||
| "doubao:embedding"
|
||||
|
||||
+39
-2
@@ -101,8 +101,7 @@ fn provider_query_model_mapping_matches_endpoint(
|
||||
) -> bool {
|
||||
let api_format_matches = mapping.api_formats.as_ref().is_none_or(|api_formats| {
|
||||
api_formats.iter().any(|value| {
|
||||
aether_scheduler_core::normalize_api_format(value)
|
||||
== aether_scheduler_core::normalize_api_format(&endpoint.api_format)
|
||||
aether_ai_formats::api_format_permission_covers(value, &endpoint.api_format)
|
||||
})
|
||||
});
|
||||
if !api_format_matches {
|
||||
@@ -367,3 +366,41 @@ fn provider_query_parse_mapping_string_list_array(
|
||||
}
|
||||
Ok(parsed)
|
||||
}
|
||||
|
||||
#[cfg(test)]
|
||||
mod tests {
|
||||
use super::*;
|
||||
|
||||
fn sample_endpoint(api_format: &str) -> StoredProviderCatalogEndpoint {
|
||||
StoredProviderCatalogEndpoint::new(
|
||||
"endpoint-1".to_string(),
|
||||
"provider-1".to_string(),
|
||||
api_format.to_string(),
|
||||
None,
|
||||
None,
|
||||
true,
|
||||
)
|
||||
.expect("endpoint should build")
|
||||
}
|
||||
|
||||
fn sample_mapping(api_format: &str) -> StoredProviderModelMapping {
|
||||
StoredProviderModelMapping {
|
||||
name: "gpt-5.6-luna".to_string(),
|
||||
priority: 1,
|
||||
api_formats: Some(vec![api_format.to_string()]),
|
||||
endpoint_ids: None,
|
||||
}
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn responses_mapping_scope_covers_search_in_one_direction() {
|
||||
assert!(provider_query_model_mapping_matches_endpoint(
|
||||
&sample_mapping("openai:responses"),
|
||||
&sample_endpoint("openai:search"),
|
||||
));
|
||||
assert!(!provider_query_model_mapping_matches_endpoint(
|
||||
&sample_mapping("openai:search"),
|
||||
&sample_endpoint("openai:responses"),
|
||||
));
|
||||
}
|
||||
}
|
||||
|
||||
@@ -475,6 +475,118 @@ fn provider_query_compact_test_request_body_defaults_to_responses_input() {
|
||||
assert!(body.get("messages").is_none());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn provider_query_search_test_request_body_defaults_to_typed_search_input() {
|
||||
let payload = json!({"message": "find current documentation"});
|
||||
|
||||
let client_api_format = provider_query_standard_test_client_api_format("openai:search");
|
||||
let body = provider_query_build_test_request_body_for_api_format(
|
||||
&payload,
|
||||
"gpt-5.6-sol",
|
||||
"/api/admin/provider-query/test-model",
|
||||
client_api_format,
|
||||
);
|
||||
|
||||
assert_eq!(client_api_format, "openai:search");
|
||||
assert!(body["id"]
|
||||
.as_str()
|
||||
.is_some_and(|value| value.starts_with("aether-model-test-")));
|
||||
assert_eq!(body["model"], json!("gpt-5.6-sol"));
|
||||
assert_eq!(body["input"], json!("find current documentation"));
|
||||
assert_eq!(
|
||||
body["commands"]["search_query"][0]["q"],
|
||||
json!("find current documentation")
|
||||
);
|
||||
assert_eq!(body["max_output_tokens"], json!(256));
|
||||
assert!(body.get("messages").is_none());
|
||||
assert!(body.get("stream").is_none());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn provider_query_search_test_completes_missing_protocol_fields() {
|
||||
let payload = json!({
|
||||
"request_body": {
|
||||
"model": "gpt-5.6-luna",
|
||||
"input": "find release notes"
|
||||
}
|
||||
});
|
||||
|
||||
let body = provider_query_build_test_request_body_for_api_format_with_search_session(
|
||||
&payload,
|
||||
"gpt-5.6-luna",
|
||||
"/api/admin/provider-query/test-model",
|
||||
"openai:search",
|
||||
Some("trace-model-test-1"),
|
||||
);
|
||||
|
||||
assert_eq!(body["id"], json!("aether-model-test-trace-model-test-1"));
|
||||
assert_eq!(
|
||||
body["commands"]["search_query"][0]["q"],
|
||||
json!("find release notes")
|
||||
);
|
||||
assert_eq!(body["max_output_tokens"], json!(256));
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn provider_query_search_test_preserves_a_non_empty_client_session_id() {
|
||||
let payload = json!({
|
||||
"request_body": {
|
||||
"id": "client-search-session",
|
||||
"model": "gpt-5.6-luna",
|
||||
"input": "find release notes"
|
||||
}
|
||||
});
|
||||
|
||||
let body = provider_query_build_test_request_body_for_api_format_with_search_session(
|
||||
&payload,
|
||||
"gpt-5.6-luna",
|
||||
"/api/admin/provider-query/test-model",
|
||||
"openai:search",
|
||||
Some("trace-model-test-1"),
|
||||
);
|
||||
|
||||
assert_eq!(body["id"], json!("client-search-session"));
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn provider_query_search_success_requires_non_empty_output() {
|
||||
fn result(body: Value) -> aether_contracts::ExecutionResult {
|
||||
aether_contracts::ExecutionResult {
|
||||
request_id: "provider-search-test".to_string(),
|
||||
candidate_id: Some("candidate-0".to_string()),
|
||||
status_code: 200,
|
||||
headers: BTreeMap::new(),
|
||||
body: Some(aether_contracts::ResponseBody {
|
||||
json_body: Some(body),
|
||||
body_bytes_b64: None,
|
||||
}),
|
||||
telemetry: None,
|
||||
error: None,
|
||||
}
|
||||
}
|
||||
|
||||
assert!(provider_query_standard_execution_response_body(
|
||||
"openai:search",
|
||||
&result(json!({})),
|
||||
None,
|
||||
)
|
||||
.is_none());
|
||||
assert!(provider_query_standard_execution_response_body(
|
||||
"openai:search",
|
||||
&result(json!({"output": " "})),
|
||||
None,
|
||||
)
|
||||
.is_none());
|
||||
assert_eq!(
|
||||
provider_query_standard_execution_response_body(
|
||||
"openai:search",
|
||||
&result(json!({"output": "search result", "encrypted_output": "ciphertext"})),
|
||||
None,
|
||||
),
|
||||
Some(json!({"output": "search result", "encrypted_output": "ciphertext"}))
|
||||
);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn provider_query_embedding_test_request_body_defaults_to_embedding_input() {
|
||||
let payload = json!({"message": "hello from embedding"});
|
||||
@@ -637,6 +749,10 @@ fn provider_query_test_adapter_routes_fixed_provider_endpoint_types() {
|
||||
provider_query_test_adapter_for_provider_api_format("codex", "openai:responses:compact"),
|
||||
Some(ProviderQueryTestAdapter::Standard)
|
||||
);
|
||||
assert_eq!(
|
||||
provider_query_test_adapter_for_provider_api_format("codex", "openai:search"),
|
||||
Some(ProviderQueryTestAdapter::Standard)
|
||||
);
|
||||
assert_eq!(
|
||||
provider_query_test_adapter_for_provider_api_format("chatgpt_web", "openai:image"),
|
||||
Some(ProviderQueryTestAdapter::OpenAiImage)
|
||||
@@ -722,6 +838,10 @@ fn provider_query_endpoint_priority_prefers_text_before_cli_and_image() {
|
||||
provider_query_model_test_endpoint_priority("codex", "openai:responses:compact"),
|
||||
Some(1)
|
||||
);
|
||||
assert_eq!(
|
||||
provider_query_model_test_endpoint_priority("codex", "openai:search"),
|
||||
Some(1)
|
||||
);
|
||||
assert_eq!(
|
||||
provider_query_model_test_endpoint_priority("chatgpt_web", "openai:image"),
|
||||
Some(2)
|
||||
|
||||
@@ -3,11 +3,79 @@ mod endpoint;
|
||||
mod template;
|
||||
mod update;
|
||||
|
||||
fn normalize_provider_request_timeout(value: Option<f64>) -> Result<Option<f64>, String> {
|
||||
let max = aether_contracts::MAX_EXECUTION_REQUEST_TIMEOUT_SECS as f64;
|
||||
match value {
|
||||
Some(value) if (1.0..=max).contains(&value) => Ok(Some(value)),
|
||||
Some(_) => Err(format!(
|
||||
"request_timeout 必须是 1 到 {} 之间的数字",
|
||||
aether_contracts::MAX_EXECUTION_REQUEST_TIMEOUT_SECS
|
||||
)),
|
||||
None => Ok(None),
|
||||
}
|
||||
}
|
||||
|
||||
fn normalize_provider_stream_first_byte_timeout(value: Option<f64>) -> Result<Option<f64>, String> {
|
||||
let max = aether_contracts::MAX_EXECUTION_STREAM_FIRST_BYTE_TIMEOUT_SECS as f64;
|
||||
match value {
|
||||
Some(value) if (1.0..=max).contains(&value) => Ok(Some(value)),
|
||||
Some(_) => Err(format!(
|
||||
"stream_first_byte_timeout 必须是 1 到 {} 之间的数字",
|
||||
aether_contracts::MAX_EXECUTION_STREAM_FIRST_BYTE_TIMEOUT_SECS
|
||||
)),
|
||||
None => Ok(None),
|
||||
}
|
||||
}
|
||||
|
||||
pub(crate) use self::create::build_admin_create_provider_record;
|
||||
pub(crate) use self::endpoint::build_admin_fixed_provider_endpoint_record;
|
||||
pub(crate) use self::template::{
|
||||
apply_admin_fixed_provider_endpoint_template_overrides,
|
||||
reconcile_admin_fixed_provider_template_endpoints,
|
||||
reconcile_admin_fixed_provider_template_keys,
|
||||
reconcile_admin_fixed_provider_template_endpoints_after_update,
|
||||
};
|
||||
pub(crate) use self::update::build_admin_update_provider_record;
|
||||
|
||||
#[cfg(test)]
|
||||
mod tests {
|
||||
use super::{normalize_provider_request_timeout, normalize_provider_stream_first_byte_timeout};
|
||||
|
||||
#[test]
|
||||
fn provider_request_timeout_accepts_the_execution_protocol_range() {
|
||||
assert_eq!(normalize_provider_request_timeout(Some(1.0)), Ok(Some(1.0)));
|
||||
assert_eq!(
|
||||
normalize_provider_request_timeout(Some(
|
||||
aether_contracts::MAX_EXECUTION_REQUEST_TIMEOUT_SECS as f64,
|
||||
)),
|
||||
Ok(Some(
|
||||
aether_contracts::MAX_EXECUTION_REQUEST_TIMEOUT_SECS as f64,
|
||||
))
|
||||
);
|
||||
assert!(normalize_provider_request_timeout(Some(0.0)).is_err());
|
||||
assert!(normalize_provider_request_timeout(Some(
|
||||
aether_contracts::MAX_EXECUTION_REQUEST_TIMEOUT_SECS as f64 + 1.0,
|
||||
))
|
||||
.is_err());
|
||||
assert!(normalize_provider_request_timeout(Some(f64::NAN)).is_err());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn provider_stream_first_byte_timeout_keeps_its_protocol_range() {
|
||||
assert_eq!(
|
||||
normalize_provider_stream_first_byte_timeout(Some(1.0)),
|
||||
Ok(Some(1.0))
|
||||
);
|
||||
assert_eq!(
|
||||
normalize_provider_stream_first_byte_timeout(Some(
|
||||
aether_contracts::MAX_EXECUTION_STREAM_FIRST_BYTE_TIMEOUT_SECS as f64,
|
||||
)),
|
||||
Ok(Some(
|
||||
aether_contracts::MAX_EXECUTION_STREAM_FIRST_BYTE_TIMEOUT_SECS as f64,
|
||||
))
|
||||
);
|
||||
assert!(normalize_provider_stream_first_byte_timeout(Some(
|
||||
aether_contracts::MAX_EXECUTION_STREAM_FIRST_BYTE_TIMEOUT_SECS as f64 + 1.0,
|
||||
))
|
||||
.is_err());
|
||||
}
|
||||
}
|
||||
|
||||
@@ -107,16 +107,9 @@ pub(crate) async fn build_admin_create_provider_record(
|
||||
None => Some(2),
|
||||
};
|
||||
let proxy = normalize_json_object(payload.proxy, "proxy")?;
|
||||
let stream_first_byte_timeout_secs = match payload.stream_first_byte_timeout {
|
||||
Some(value) if (1.0..=300.0).contains(&value) => Some(value),
|
||||
Some(_) => return Err("stream_first_byte_timeout 必须是 1 到 300 之间的数字".to_string()),
|
||||
None => None,
|
||||
};
|
||||
let request_timeout_secs = match payload.request_timeout {
|
||||
Some(value) if (1.0..=600.0).contains(&value) => Some(value),
|
||||
Some(_) => return Err("request_timeout 必须是 1 到 600 之间的数字".to_string()),
|
||||
None => None,
|
||||
};
|
||||
let stream_first_byte_timeout_secs =
|
||||
super::normalize_provider_stream_first_byte_timeout(payload.stream_first_byte_timeout)?;
|
||||
let request_timeout_secs = super::normalize_provider_request_timeout(payload.request_timeout)?;
|
||||
|
||||
let mut config_map = normalize_json_object(payload.config, "config")?
|
||||
.and_then(|value| value.as_object().cloned())
|
||||
|
||||
@@ -1,11 +1,11 @@
|
||||
use super::endpoint::{
|
||||
build_admin_fixed_provider_endpoint_defaults, build_admin_fixed_provider_endpoint_record,
|
||||
AdminFixedProviderEndpointDefaults,
|
||||
};
|
||||
use crate::handlers::admin::request::AdminAppState;
|
||||
use crate::provider_key_auth::provider_key_is_oauth_managed;
|
||||
use crate::GatewayError;
|
||||
use aether_data_contracts::repository::provider_catalog::{
|
||||
StoredProviderCatalogEndpoint, StoredProviderCatalogKey, StoredProviderCatalogProvider,
|
||||
StoredProviderCatalogEndpoint, StoredProviderCatalogProvider,
|
||||
};
|
||||
use aether_provider_transport::provider_types::{
|
||||
fixed_provider_template, FixedProviderEndpointTemplate, FixedProviderTemplate,
|
||||
@@ -36,6 +36,30 @@ struct FixedProviderEndpointMetadata {
|
||||
pub(crate) async fn reconcile_admin_fixed_provider_template_endpoints(
|
||||
state: &AdminAppState<'_>,
|
||||
provider: &StoredProviderCatalogProvider,
|
||||
) -> Result<(), GatewayError> {
|
||||
reconcile_admin_fixed_provider_template_endpoints_with_adoption_provider(
|
||||
state, provider, provider,
|
||||
)
|
||||
.await
|
||||
}
|
||||
|
||||
pub(crate) async fn reconcile_admin_fixed_provider_template_endpoints_after_update(
|
||||
state: &AdminAppState<'_>,
|
||||
previous_provider: &StoredProviderCatalogProvider,
|
||||
updated_provider: &StoredProviderCatalogProvider,
|
||||
) -> Result<(), GatewayError> {
|
||||
reconcile_admin_fixed_provider_template_endpoints_with_adoption_provider(
|
||||
state,
|
||||
updated_provider,
|
||||
previous_provider,
|
||||
)
|
||||
.await
|
||||
}
|
||||
|
||||
async fn reconcile_admin_fixed_provider_template_endpoints_with_adoption_provider(
|
||||
state: &AdminAppState<'_>,
|
||||
provider: &StoredProviderCatalogProvider,
|
||||
adoption_provider: &StoredProviderCatalogProvider,
|
||||
) -> Result<(), GatewayError> {
|
||||
let Some(template) = state.fixed_provider_template(&provider.provider_type) else {
|
||||
return Ok(());
|
||||
@@ -53,8 +77,9 @@ pub(crate) async fn reconcile_admin_fixed_provider_template_endpoints(
|
||||
match existing_endpoint {
|
||||
Some(existing_endpoint) => {
|
||||
matched_endpoint_ids.insert(existing_endpoint.id.clone());
|
||||
let updated = reconcile_fixed_provider_endpoint(
|
||||
let updated = reconcile_fixed_provider_endpoint_with_adoption_provider(
|
||||
provider,
|
||||
adoption_provider,
|
||||
existing_endpoint,
|
||||
template,
|
||||
endpoint_template,
|
||||
@@ -115,31 +140,6 @@ pub(crate) async fn reconcile_admin_fixed_provider_template_endpoints(
|
||||
Ok(())
|
||||
}
|
||||
|
||||
pub(crate) async fn reconcile_admin_fixed_provider_template_keys(
|
||||
state: &AdminAppState<'_>,
|
||||
provider: &StoredProviderCatalogProvider,
|
||||
) -> Result<(), GatewayError> {
|
||||
let Some(_) = state.fixed_provider_template(&provider.provider_type) else {
|
||||
return Ok(());
|
||||
};
|
||||
|
||||
let existing_keys = state
|
||||
.list_provider_catalog_keys_by_provider_ids(std::slice::from_ref(&provider.id))
|
||||
.await?;
|
||||
for existing_key in existing_keys {
|
||||
let Some(updated_key) = reconcile_fixed_provider_key(provider, &existing_key) else {
|
||||
continue;
|
||||
};
|
||||
let Some(_) = state.update_provider_catalog_key(&updated_key).await? else {
|
||||
return Err(GatewayError::Internal(
|
||||
"provider catalog key writer unavailable".to_string(),
|
||||
));
|
||||
};
|
||||
}
|
||||
|
||||
Ok(())
|
||||
}
|
||||
|
||||
pub(crate) fn apply_admin_fixed_provider_endpoint_template_overrides(
|
||||
provider: &StoredProviderCatalogProvider,
|
||||
existing_endpoint: &StoredProviderCatalogEndpoint,
|
||||
@@ -156,8 +156,14 @@ pub(crate) fn apply_admin_fixed_provider_endpoint_template_overrides(
|
||||
|
||||
let defaults =
|
||||
build_admin_fixed_provider_endpoint_defaults(provider, template, endpoint_template)?;
|
||||
let mut metadata = fixed_provider_endpoint_metadata(existing_endpoint)
|
||||
.unwrap_or_else(|| managed_fixed_provider_endpoint_metadata(template, endpoint_template));
|
||||
let mut metadata = fixed_provider_endpoint_metadata(existing_endpoint).unwrap_or_else(|| {
|
||||
adopt_fixed_provider_endpoint_metadata(
|
||||
existing_endpoint,
|
||||
&defaults,
|
||||
template,
|
||||
endpoint_template,
|
||||
)
|
||||
});
|
||||
let mut overrides = metadata.overrides.clone();
|
||||
|
||||
sync_override_if_changed(
|
||||
@@ -250,17 +256,48 @@ pub(crate) fn apply_admin_fixed_provider_endpoint_template_overrides(
|
||||
Ok(())
|
||||
}
|
||||
|
||||
#[cfg(test)]
|
||||
fn reconcile_fixed_provider_endpoint(
|
||||
provider: &StoredProviderCatalogProvider,
|
||||
existing_endpoint: &StoredProviderCatalogEndpoint,
|
||||
template: &FixedProviderTemplate,
|
||||
endpoint_template: &FixedProviderEndpointTemplate,
|
||||
) -> Result<StoredProviderCatalogEndpoint, String> {
|
||||
reconcile_fixed_provider_endpoint_with_adoption_provider(
|
||||
provider,
|
||||
provider,
|
||||
existing_endpoint,
|
||||
template,
|
||||
endpoint_template,
|
||||
)
|
||||
}
|
||||
|
||||
fn reconcile_fixed_provider_endpoint_with_adoption_provider(
|
||||
provider: &StoredProviderCatalogProvider,
|
||||
adoption_provider: &StoredProviderCatalogProvider,
|
||||
existing_endpoint: &StoredProviderCatalogEndpoint,
|
||||
template: &FixedProviderTemplate,
|
||||
endpoint_template: &FixedProviderEndpointTemplate,
|
||||
) -> Result<StoredProviderCatalogEndpoint, String> {
|
||||
let defaults =
|
||||
build_admin_fixed_provider_endpoint_defaults(provider, template, endpoint_template)?;
|
||||
let mut updated = existing_endpoint.clone();
|
||||
let metadata = fixed_provider_endpoint_metadata(existing_endpoint)
|
||||
.unwrap_or_else(|| managed_fixed_provider_endpoint_metadata(template, endpoint_template));
|
||||
let metadata = match fixed_provider_endpoint_metadata(existing_endpoint) {
|
||||
Some(metadata) => metadata,
|
||||
None => {
|
||||
let adoption_defaults = build_admin_fixed_provider_endpoint_defaults(
|
||||
adoption_provider,
|
||||
template,
|
||||
endpoint_template,
|
||||
)?;
|
||||
adopt_fixed_provider_endpoint_metadata(
|
||||
existing_endpoint,
|
||||
&adoption_defaults,
|
||||
template,
|
||||
endpoint_template,
|
||||
)
|
||||
}
|
||||
};
|
||||
|
||||
updated.api_format = defaults.api_format.clone();
|
||||
updated.api_family = Some(defaults.api_family.clone());
|
||||
@@ -455,6 +492,77 @@ fn managed_fixed_provider_endpoint_metadata(
|
||||
}
|
||||
}
|
||||
|
||||
fn adopt_fixed_provider_endpoint_metadata(
|
||||
existing_endpoint: &StoredProviderCatalogEndpoint,
|
||||
defaults: &AdminFixedProviderEndpointDefaults,
|
||||
template: &FixedProviderTemplate,
|
||||
endpoint_template: &FixedProviderEndpointTemplate,
|
||||
) -> FixedProviderEndpointMetadata {
|
||||
let mut metadata = managed_fixed_provider_endpoint_metadata(template, endpoint_template);
|
||||
sync_override(
|
||||
&mut metadata.overrides,
|
||||
OVERRIDE_BASE_URL,
|
||||
&existing_endpoint.base_url,
|
||||
&defaults.base_url,
|
||||
);
|
||||
sync_override(
|
||||
&mut metadata.overrides,
|
||||
OVERRIDE_CUSTOM_PATH,
|
||||
&existing_endpoint.custom_path,
|
||||
&defaults.custom_path,
|
||||
);
|
||||
sync_override(
|
||||
&mut metadata.overrides,
|
||||
OVERRIDE_HEADER_RULES,
|
||||
&existing_endpoint.header_rules,
|
||||
&defaults.header_rules,
|
||||
);
|
||||
sync_override(
|
||||
&mut metadata.overrides,
|
||||
OVERRIDE_BODY_RULES,
|
||||
&existing_endpoint.body_rules,
|
||||
&defaults.body_rules,
|
||||
);
|
||||
sync_override(
|
||||
&mut metadata.overrides,
|
||||
OVERRIDE_MAX_RETRIES,
|
||||
&existing_endpoint.max_retries,
|
||||
&defaults.max_retries,
|
||||
);
|
||||
sync_override(
|
||||
&mut metadata.overrides,
|
||||
OVERRIDE_IS_ACTIVE,
|
||||
&existing_endpoint.is_active,
|
||||
&defaults.is_active,
|
||||
);
|
||||
sync_override(
|
||||
&mut metadata.overrides,
|
||||
OVERRIDE_PROXY,
|
||||
&existing_endpoint.proxy,
|
||||
&defaults.proxy,
|
||||
);
|
||||
sync_override(
|
||||
&mut metadata.overrides,
|
||||
OVERRIDE_FORMAT_ACCEPTANCE_CONFIG,
|
||||
&existing_endpoint.format_acceptance_config,
|
||||
&defaults.format_acceptance_config,
|
||||
);
|
||||
|
||||
let existing_config = endpoint_config_without_metadata(existing_endpoint.config.as_ref());
|
||||
for (key, desired) in fixed_provider_endpoint_config_defaults(endpoint_template) {
|
||||
let Some(actual) = existing_config.get(&key) else {
|
||||
continue;
|
||||
};
|
||||
sync_override(
|
||||
&mut metadata.overrides,
|
||||
&config_override_key(&key),
|
||||
actual,
|
||||
&desired,
|
||||
);
|
||||
}
|
||||
metadata
|
||||
}
|
||||
|
||||
fn upsert_fixed_provider_endpoint_metadata(
|
||||
endpoint: &mut StoredProviderCatalogEndpoint,
|
||||
metadata: &FixedProviderEndpointMetadata,
|
||||
@@ -513,22 +621,6 @@ fn current_unix_secs() -> u64 {
|
||||
.unwrap_or(0)
|
||||
}
|
||||
|
||||
fn reconcile_fixed_provider_key(
|
||||
provider: &StoredProviderCatalogProvider,
|
||||
existing_key: &StoredProviderCatalogKey,
|
||||
) -> Option<StoredProviderCatalogKey> {
|
||||
if !provider_key_is_oauth_managed(existing_key, &provider.provider_type)
|
||||
|| existing_key.api_formats.is_none()
|
||||
{
|
||||
return None;
|
||||
}
|
||||
|
||||
let mut updated = existing_key.clone();
|
||||
updated.api_formats = None;
|
||||
updated.updated_at_unix_secs = Some(current_unix_secs());
|
||||
Some(updated)
|
||||
}
|
||||
|
||||
fn sync_override<T>(overrides: &mut BTreeSet<String>, key: &str, actual: &T, desired: &T)
|
||||
where
|
||||
T: PartialEq,
|
||||
@@ -624,4 +716,68 @@ mod tests {
|
||||
.expect("endpoint should reconcile");
|
||||
assert_eq!(reconciled.base_url, "http://127.0.0.1:18181/v1");
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn fixed_provider_endpoint_reconcile_adopts_existing_customization() {
|
||||
let provider = sample_codex_provider();
|
||||
let template = fixed_provider_template("codex").expect("codex template should exist");
|
||||
let endpoint_template = template
|
||||
.endpoints
|
||||
.iter()
|
||||
.find(|endpoint| endpoint.api_format == "openai:responses")
|
||||
.expect("responses endpoint template should exist");
|
||||
|
||||
let mut existing = sample_codex_endpoint("http://127.0.0.1:18181/backend-api/codex");
|
||||
existing.is_active = false;
|
||||
existing.max_retries = Some(9);
|
||||
existing.proxy = Some(serde_json::json!({"url": "http://proxy.internal:8080"}));
|
||||
existing.config = Some(serde_json::json!({
|
||||
"upstream_stream_policy": "force_non_stream",
|
||||
"custom_transport_option": true
|
||||
}));
|
||||
|
||||
let reconciled =
|
||||
reconcile_fixed_provider_endpoint(&provider, &existing, template, endpoint_template)
|
||||
.expect("endpoint should reconcile");
|
||||
assert_eq!(
|
||||
reconciled.base_url,
|
||||
"http://127.0.0.1:18181/backend-api/codex"
|
||||
);
|
||||
assert!(!reconciled.is_active);
|
||||
assert_eq!(reconciled.max_retries, Some(9));
|
||||
assert_eq!(
|
||||
reconciled.proxy,
|
||||
Some(serde_json::json!({"url": "http://proxy.internal:8080"}))
|
||||
);
|
||||
assert_eq!(
|
||||
reconciled
|
||||
.config
|
||||
.as_ref()
|
||||
.and_then(|value| value.get("upstream_stream_policy")),
|
||||
Some(&serde_json::json!("force_non_stream"))
|
||||
);
|
||||
assert_eq!(
|
||||
reconciled
|
||||
.config
|
||||
.as_ref()
|
||||
.and_then(|value| value.get("custom_transport_option")),
|
||||
Some(&serde_json::json!(true))
|
||||
);
|
||||
let metadata = fixed_provider_endpoint_metadata(&reconciled)
|
||||
.expect("fixed provider metadata should exist");
|
||||
for key in [
|
||||
"base_url",
|
||||
"is_active",
|
||||
"max_retries",
|
||||
"proxy",
|
||||
"config.upstream_stream_policy",
|
||||
] {
|
||||
assert!(metadata.overrides.contains(key), "missing override {key}");
|
||||
}
|
||||
|
||||
let second =
|
||||
reconcile_fixed_provider_endpoint(&provider, &reconciled, template, endpoint_template)
|
||||
.expect("endpoint should reconcile idempotently");
|
||||
assert_eq!(second, reconciled);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -202,21 +202,13 @@ pub(crate) async fn build_admin_update_provider_record(
|
||||
}
|
||||
|
||||
if fields.contains("stream_first_byte_timeout") {
|
||||
updated.stream_first_byte_timeout_secs = match payload.stream_first_byte_timeout {
|
||||
Some(value) if (1.0..=300.0).contains(&value) => Some(value),
|
||||
Some(_) => {
|
||||
return Err("stream_first_byte_timeout 必须是 1 到 300 之间的数字".to_string());
|
||||
}
|
||||
None => None,
|
||||
};
|
||||
updated.stream_first_byte_timeout_secs =
|
||||
super::normalize_provider_stream_first_byte_timeout(payload.stream_first_byte_timeout)?;
|
||||
}
|
||||
|
||||
if fields.contains("request_timeout") {
|
||||
updated.request_timeout_secs = match payload.request_timeout {
|
||||
Some(value) if (1.0..=600.0).contains(&value) => Some(value),
|
||||
Some(_) => return Err("request_timeout 必须是 1 到 600 之间的数字".to_string()),
|
||||
None => None,
|
||||
};
|
||||
updated.request_timeout_secs =
|
||||
super::normalize_provider_request_timeout(payload.request_timeout)?;
|
||||
}
|
||||
|
||||
if fields.contains("enable_format_conversion") {
|
||||
|
||||
@@ -1,5 +1,26 @@
|
||||
use super::*;
|
||||
|
||||
fn validate_admin_endpoint_stream_policy(
|
||||
api_format: &str,
|
||||
config: Option<&serde_json::Value>,
|
||||
) -> Result<(), String> {
|
||||
if !crate::ai_serving::api_format_alias_matches(api_format, "openai:search") {
|
||||
return Ok(());
|
||||
}
|
||||
let requested = config
|
||||
.and_then(serde_json::Value::as_object)
|
||||
.and_then(|config| {
|
||||
config
|
||||
.get("upstream_stream_policy")
|
||||
.or_else(|| config.get("upstreamStreamPolicy"))
|
||||
.or_else(|| config.get("upstream_stream"))
|
||||
});
|
||||
if requested.is_some_and(crate::handlers::public::admin_requested_force_stream) {
|
||||
return Err("OpenAI Search 端点仅支持非流式上游请求".to_string());
|
||||
}
|
||||
Ok(())
|
||||
}
|
||||
|
||||
impl<'a> AdminAppState<'a> {
|
||||
pub(crate) async fn build_admin_keys_grouped_by_format_payload(
|
||||
&self,
|
||||
@@ -232,6 +253,7 @@ impl<'a> AdminAppState<'a> {
|
||||
let (normalized_api_format, api_family, endpoint_kind) =
|
||||
admin_endpoint_signature_parts(&payload.api_format)
|
||||
.ok_or_else(|| format!("无效的 api_format: {}", payload.api_format))?;
|
||||
validate_admin_endpoint_stream_policy(normalized_api_format, payload.config.as_ref())?;
|
||||
let base_url = normalize_admin_base_url(&payload.base_url)?;
|
||||
|
||||
let existing_endpoints = self
|
||||
@@ -337,6 +359,13 @@ impl<'a> AdminAppState<'a> {
|
||||
&update_fields,
|
||||
)?;
|
||||
|
||||
if fields.contains("config") {
|
||||
validate_admin_endpoint_stream_policy(
|
||||
existing_endpoint.api_format.as_str(),
|
||||
updated.config.as_ref(),
|
||||
)?;
|
||||
}
|
||||
|
||||
if provider_type == "codex"
|
||||
&& crate::ai_serving::is_openai_responses_format(&existing_endpoint.api_format)
|
||||
{
|
||||
@@ -389,3 +418,40 @@ impl<'a> AdminAppState<'a> {
|
||||
Ok(updated)
|
||||
}
|
||||
}
|
||||
|
||||
#[cfg(test)]
|
||||
mod tests {
|
||||
use serde_json::json;
|
||||
|
||||
use super::validate_admin_endpoint_stream_policy;
|
||||
|
||||
#[test]
|
||||
fn search_endpoint_rejects_explicit_streaming_policy_for_all_config_keys() {
|
||||
for (api_format, key, value) in [
|
||||
(
|
||||
"openai:search",
|
||||
"upstream_stream_policy",
|
||||
json!("force_stream"),
|
||||
),
|
||||
("openai:search", "upstreamStreamPolicy", json!(true)),
|
||||
("/v1/alpha/search", "upstream_stream", json!("sse")),
|
||||
] {
|
||||
let config = json!({(key): value});
|
||||
assert!(validate_admin_endpoint_stream_policy(api_format, Some(&config),).is_err());
|
||||
}
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn search_endpoint_accepts_non_streaming_and_unrelated_config() {
|
||||
assert!(validate_admin_endpoint_stream_policy(
|
||||
"openai:search",
|
||||
Some(&json!({"upstream_stream_policy": "force_non_stream"})),
|
||||
)
|
||||
.is_ok());
|
||||
assert!(validate_admin_endpoint_stream_policy(
|
||||
"openai:responses",
|
||||
Some(&json!({"upstream_stream_policy": "force_stream"})),
|
||||
)
|
||||
.is_ok());
|
||||
}
|
||||
}
|
||||
|
||||
@@ -83,6 +83,8 @@ const OPENAI_RESPONSES_LOCAL_EXECUTION_RUNTIME_MISS_DETAIL: &str =
|
||||
"当前 OpenAI Responses 请求无法在本地执行:没有匹配到可用的执行路径";
|
||||
const OPENAI_RESPONSES_COMPACT_LOCAL_EXECUTION_RUNTIME_MISS_DETAIL: &str =
|
||||
"当前 OpenAI Responses Compact 请求无法在本地执行:没有匹配到可用的执行路径";
|
||||
const OPENAI_SEARCH_LOCAL_EXECUTION_RUNTIME_MISS_DETAIL: &str =
|
||||
"当前 OpenAI Search 请求无法在本地执行:没有匹配到可用的执行路径";
|
||||
const OPENAI_VIDEO_LOCAL_EXECUTION_RUNTIME_MISS_DETAIL: &str =
|
||||
"当前 OpenAI Video 请求无法在本地执行:没有匹配到可用的执行路径";
|
||||
const CLAUDE_MESSAGES_LOCAL_EXECUTION_RUNTIME_MISS_DETAIL: &str =
|
||||
@@ -565,8 +567,9 @@ async fn maybe_forward_public_request_to_tunnel_owner(
|
||||
serde_json::from_slice::<serde_json::Value>(body.as_ref()).ok()
|
||||
});
|
||||
let client_session_affinity =
|
||||
crate::client_session_affinity::client_session_affinity_from_parts(
|
||||
parts,
|
||||
crate::client_session_affinity::client_session_affinity_from_api_request(
|
||||
api_format,
|
||||
&parts.headers,
|
||||
body_json.as_ref(),
|
||||
);
|
||||
let Some(target) = crate::scheduler::affinity::read_cached_scheduler_affinity_target(
|
||||
@@ -636,7 +639,27 @@ async fn maybe_forward_public_request_to_tunnel_owner(
|
||||
owner.relay_base_url.trim_end_matches('/'),
|
||||
request_context.request_path_and_query()
|
||||
);
|
||||
let mut upstream_request = state.client.request(parts.method.clone(), owner_url);
|
||||
let is_stream =
|
||||
owner_forward_request_is_stream(parts, decision, buffered_body.unwrap_or(&empty_body));
|
||||
let transport_timeouts =
|
||||
crate::provider_transport::resolve_transport_execution_timeouts(&transport);
|
||||
let non_stream_timeout =
|
||||
crate::execution_runtime::transport::resolve_non_stream_total_timeout_for_request(
|
||||
is_stream,
|
||||
&transport.endpoint.api_format,
|
||||
transport_timeouts.as_ref(),
|
||||
);
|
||||
let stream_first_byte_timeout =
|
||||
crate::execution_runtime::transport::resolve_stream_first_byte_timeout_for_request(
|
||||
is_stream,
|
||||
transport_timeouts.as_ref(),
|
||||
);
|
||||
let mut upstream_request = state
|
||||
.owner_forward_client
|
||||
.request(parts.method.clone(), owner_url);
|
||||
if let Some(timeout) = non_stream_timeout {
|
||||
upstream_request = upstream_request.timeout(timeout);
|
||||
}
|
||||
for (name, value) in &parts.headers {
|
||||
if should_skip_request_header(name.as_str()) || name == http::header::HOST {
|
||||
continue;
|
||||
@@ -682,14 +705,15 @@ async fn maybe_forward_public_request_to_tunnel_owner(
|
||||
upstream_request.header(TRUSTED_AUTH_BALANCE_HEADER, balance_remaining.to_string());
|
||||
}
|
||||
|
||||
let upstream_response = upstream_request
|
||||
.body(buffered_body.cloned().unwrap_or_default())
|
||||
.send()
|
||||
.await
|
||||
.map_err(|err| GatewayError::UpstreamUnavailable {
|
||||
trace_id: request_context.trace_id.clone(),
|
||||
message: format!("owner gateway affinity forward failed: {err}"),
|
||||
})?;
|
||||
let upstream_response = crate::tunnel::send_owner_forward_request(
|
||||
upstream_request.body(buffered_body.cloned().unwrap_or_default()),
|
||||
stream_first_byte_timeout,
|
||||
)
|
||||
.await
|
||||
.map_err(|message| GatewayError::UpstreamUnavailable {
|
||||
trace_id: request_context.trace_id.clone(),
|
||||
message: format!("owner gateway affinity forward failed: {message}"),
|
||||
})?;
|
||||
|
||||
let mut response = build_sync_aware_affinity_forward_response(
|
||||
request_context,
|
||||
@@ -707,6 +731,29 @@ async fn maybe_forward_public_request_to_tunnel_owner(
|
||||
Ok(Some(response))
|
||||
}
|
||||
|
||||
fn owner_forward_request_is_stream(
|
||||
parts: &http::request::Parts,
|
||||
decision: &GatewayControlDecision,
|
||||
body_bytes: &Bytes,
|
||||
) -> bool {
|
||||
let Some(plan_kind) =
|
||||
crate::ai_serving::api::resolve_execution_runtime_stream_plan_kind(parts, decision)
|
||||
else {
|
||||
return false;
|
||||
};
|
||||
let Some((body_json, body_base64)) =
|
||||
crate::ai_serving::api::parse_direct_request_body(parts, body_bytes)
|
||||
else {
|
||||
return false;
|
||||
};
|
||||
crate::ai_serving::api::is_matching_stream_request(
|
||||
plan_kind,
|
||||
parts,
|
||||
&body_json,
|
||||
body_base64.as_deref(),
|
||||
)
|
||||
}
|
||||
|
||||
fn upstream_response_is_sse(headers: &reqwest::header::HeaderMap) -> bool {
|
||||
headers
|
||||
.get(http::header::CONTENT_TYPE)
|
||||
@@ -2372,6 +2419,7 @@ fn local_execution_runtime_miss_route_label(
|
||||
"/v1/chat/completions" => "OpenAI Chat Completions",
|
||||
"/v1/responses" => "OpenAI Responses",
|
||||
"/v1/responses/compact" => "OpenAI Responses Compact",
|
||||
"/v1/alpha/search" => "OpenAI Search",
|
||||
"/v1/messages" => "Claude Messages",
|
||||
path if path.starts_with("/v1/videos") => "OpenAI Video",
|
||||
path if path.starts_with("/upload/v1beta/files") || path.starts_with("/v1beta/files") => {
|
||||
@@ -2414,6 +2462,7 @@ fn local_execution_runtime_miss_route_detail(
|
||||
"/v1/responses/compact" => {
|
||||
Some(OPENAI_RESPONSES_COMPACT_LOCAL_EXECUTION_RUNTIME_MISS_DETAIL)
|
||||
}
|
||||
"/v1/alpha/search" => Some(OPENAI_SEARCH_LOCAL_EXECUTION_RUNTIME_MISS_DETAIL),
|
||||
"/v1/messages" => Some(CLAUDE_MESSAGES_LOCAL_EXECUTION_RUNTIME_MISS_DETAIL),
|
||||
path if path.starts_with("/v1/videos") => {
|
||||
Some(OPENAI_VIDEO_LOCAL_EXECUTION_RUNTIME_MISS_DETAIL)
|
||||
@@ -2437,14 +2486,128 @@ mod tests {
|
||||
use super::{
|
||||
api_key_remote_ip_allowed, buffer_and_normalize_request_body,
|
||||
diagnostic_is_auth_api_key_concurrency_limited, local_execution_runtime_miss_detail,
|
||||
restore_redacted_stream_execution_response, restore_redacted_sync_execution_response,
|
||||
GatewayControlDecision, LocalExecutionRuntimeMissDiagnostic, RequestBodyBufferError,
|
||||
RequestBodyBufferPolicy,
|
||||
owner_forward_request_is_stream, restore_redacted_stream_execution_response,
|
||||
restore_redacted_sync_execution_response, GatewayControlDecision,
|
||||
LocalExecutionRuntimeMissDiagnostic, RequestBodyBufferError, RequestBodyBufferPolicy,
|
||||
};
|
||||
use axum::body::{to_bytes, Body, Bytes};
|
||||
use axum::http::{header, HeaderMap, Method, Response};
|
||||
use serde_json::json;
|
||||
|
||||
#[test]
|
||||
fn owner_forward_uses_search_protocol_timeout_semantics() {
|
||||
let request = http::Request::builder()
|
||||
.method(Method::POST)
|
||||
.uri("/v1/alpha/search")
|
||||
.header(header::CONTENT_TYPE, "application/json")
|
||||
.body(())
|
||||
.expect("request should build");
|
||||
let (parts, _) = request.into_parts();
|
||||
let decision = GatewayControlDecision::synthetic(
|
||||
"/v1/alpha/search",
|
||||
Some("ai_public".to_string()),
|
||||
Some("openai".to_string()),
|
||||
Some("search".to_string()),
|
||||
Some("openai:search".to_string()),
|
||||
);
|
||||
let body =
|
||||
Bytes::from_static(br#"{"model":"gpt-5.6-sol","input":"find docs","stream":true}"#);
|
||||
let is_stream = owner_forward_request_is_stream(&parts, &decision, &body);
|
||||
let timeouts = aether_contracts::ExecutionTimeouts {
|
||||
total_ms: Some(aether_contracts::MAX_EXECUTION_REQUEST_TIMEOUT_MS),
|
||||
first_byte_ms: Some(10),
|
||||
..aether_contracts::ExecutionTimeouts::default()
|
||||
};
|
||||
|
||||
assert!(!is_stream);
|
||||
assert_eq!(
|
||||
crate::execution_runtime::transport::resolve_non_stream_total_timeout_for_request(
|
||||
is_stream,
|
||||
"openai:search",
|
||||
Some(&timeouts),
|
||||
),
|
||||
Some(Duration::from_millis(
|
||||
aether_contracts::MAX_EXECUTION_REQUEST_TIMEOUT_MS
|
||||
))
|
||||
);
|
||||
assert_eq!(
|
||||
crate::execution_runtime::transport::resolve_stream_first_byte_timeout_for_request(
|
||||
is_stream,
|
||||
Some(&timeouts),
|
||||
),
|
||||
None
|
||||
);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn owner_forward_keeps_streaming_for_stream_capable_protocols() {
|
||||
let chat_request = http::Request::builder()
|
||||
.method(Method::POST)
|
||||
.uri("/v1/chat/completions")
|
||||
.header(header::CONTENT_TYPE, "application/json")
|
||||
.body(())
|
||||
.expect("request should build");
|
||||
let (chat_parts, _) = chat_request.into_parts();
|
||||
let chat_decision = GatewayControlDecision::synthetic(
|
||||
"/v1/chat/completions",
|
||||
Some("ai_public".to_string()),
|
||||
Some("openai".to_string()),
|
||||
Some("chat".to_string()),
|
||||
Some("openai:chat".to_string()),
|
||||
);
|
||||
|
||||
assert!(owner_forward_request_is_stream(
|
||||
&chat_parts,
|
||||
&chat_decision,
|
||||
&Bytes::from_static(br#"{"model":"gpt-5.6-sol","stream":true}"#),
|
||||
));
|
||||
assert!(!owner_forward_request_is_stream(
|
||||
&chat_parts,
|
||||
&chat_decision,
|
||||
&Bytes::from_static(br#"{"model":"gpt-5.6-sol","stream":false}"#),
|
||||
));
|
||||
|
||||
let image_request = http::Request::builder()
|
||||
.method(Method::POST)
|
||||
.uri("/v1/images/generations")
|
||||
.header(header::CONTENT_TYPE, "application/json")
|
||||
.body(())
|
||||
.expect("request should build");
|
||||
let (image_parts, _) = image_request.into_parts();
|
||||
let image_decision = GatewayControlDecision::synthetic(
|
||||
"/v1/images/generations",
|
||||
Some("ai_public".to_string()),
|
||||
Some("openai".to_string()),
|
||||
Some("image".to_string()),
|
||||
Some("openai:image".to_string()),
|
||||
);
|
||||
assert!(owner_forward_request_is_stream(
|
||||
&image_parts,
|
||||
&image_decision,
|
||||
&Bytes::from_static(br#"{"model":"gpt-image-1","stream":true}"#),
|
||||
));
|
||||
|
||||
let compact_request = http::Request::builder()
|
||||
.method(Method::POST)
|
||||
.uri("/v1/responses/compact")
|
||||
.header(header::CONTENT_TYPE, "application/json")
|
||||
.body(())
|
||||
.expect("request should build");
|
||||
let (compact_parts, _) = compact_request.into_parts();
|
||||
let compact_decision = GatewayControlDecision::synthetic(
|
||||
"/v1/responses/compact",
|
||||
Some("ai_public".to_string()),
|
||||
Some("openai".to_string()),
|
||||
Some("responses:compact".to_string()),
|
||||
Some("openai:responses:compact".to_string()),
|
||||
);
|
||||
assert!(!owner_forward_request_is_stream(
|
||||
&compact_parts,
|
||||
&compact_decision,
|
||||
&Bytes::from_static(br#"{"model":"gpt-5.6-sol","stream":true}"#),
|
||||
));
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn api_key_remote_ip_allows_unrestricted_keys() {
|
||||
let remote_ip = "203.0.113.10".parse().expect("valid ip");
|
||||
|
||||
@@ -220,14 +220,7 @@ fn users_me_usage_api_format_defaults_to_non_stream(item: &StoredRequestUsageAud
|
||||
let Some(value) = api_format else {
|
||||
return false;
|
||||
};
|
||||
matches!(
|
||||
crate::ai_serving::normalize_api_format_alias(value).as_str(),
|
||||
"openai:chat"
|
||||
| "openai:responses"
|
||||
| "openai:responses:compact"
|
||||
| "openai:image"
|
||||
| "claude:messages"
|
||||
)
|
||||
aether_ai_formats::api_format_defaults_to_non_stream(value)
|
||||
}
|
||||
|
||||
fn users_me_usage_request_body_implies_default_non_stream(item: &StoredRequestUsageAudit) -> bool {
|
||||
@@ -1846,6 +1839,27 @@ mod tests {
|
||||
assert_eq!(active_payload["client_is_stream"], false);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn user_usage_stream_defaults_to_non_stream_for_openai_search() {
|
||||
let item = StoredRequestUsageAudit {
|
||||
is_stream: false,
|
||||
api_format: Some("openai:search".to_string()),
|
||||
request_body: Some(json!({
|
||||
"id": "session-search-1",
|
||||
"model": "gpt-5.6-sol",
|
||||
"input": "current documentation"
|
||||
})),
|
||||
..sample_usage("completed")
|
||||
};
|
||||
|
||||
assert!(!users_me_usage_client_is_stream(&item));
|
||||
|
||||
let record_payload =
|
||||
build_users_me_usage_record_payload(&item, false, &BTreeMap::new(), false);
|
||||
assert_eq!(record_payload["client_requested_stream"], false);
|
||||
assert_eq!(record_payload["client_is_stream"], false);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn user_usage_upstream_stream_prefers_request_metadata_flag() {
|
||||
let item = StoredRequestUsageAudit {
|
||||
|
||||
@@ -38,7 +38,7 @@ pub(crate) fn provider_catalog_key_supports_format(
|
||||
}
|
||||
formats
|
||||
.iter()
|
||||
.any(|candidate| crate::ai_serving::api_format_alias_matches(candidate, api_format))
|
||||
.any(|candidate| aether_ai_formats::api_format_permission_covers(candidate, api_format))
|
||||
}
|
||||
|
||||
pub(crate) fn decrypt_catalog_secret_with_fallbacks(
|
||||
@@ -2732,6 +2732,25 @@ mod tests {
|
||||
.expect("key transport should build")
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn responses_key_scope_covers_search_in_one_direction() {
|
||||
let mut responses_key = sample_catalog_key();
|
||||
responses_key.api_formats = Some(json!(["openai:responses"]));
|
||||
assert!(provider_catalog_key_supports_format(
|
||||
&responses_key,
|
||||
"codex",
|
||||
"openai:search",
|
||||
));
|
||||
|
||||
let mut search_key = sample_catalog_key();
|
||||
search_key.api_formats = Some(json!(["openai:search"]));
|
||||
assert!(!provider_catalog_key_supports_format(
|
||||
&search_key,
|
||||
"codex",
|
||||
"openai:responses",
|
||||
));
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn masked_catalog_api_key_handles_unicode_plaintext_without_panicking() {
|
||||
let state = AppState::new().expect("gateway should build");
|
||||
|
||||
Reference in New Issue
Block a user