mirror of
https://github.com/fawney19/Aether.git
synced 2026-09-02 01:10:23 +08:00
feat: wire OpenAI responses session affinity
This commit is contained in:
@@ -32,6 +32,7 @@ use crate::ai_serving::{
|
|||||||
resolve_local_decision_execution_runtime_auth_context, ExecutionRuntimeAuthContext,
|
resolve_local_decision_execution_runtime_auth_context, ExecutionRuntimeAuthContext,
|
||||||
GatewayControlDecision, PlannerAppState,
|
GatewayControlDecision, PlannerAppState,
|
||||||
};
|
};
|
||||||
|
use crate::client_session_affinity::client_session_affinity_from_parts;
|
||||||
use crate::{AppState, GatewayError};
|
use crate::{AppState, GatewayError};
|
||||||
|
|
||||||
use super::LocalOpenAiResponsesSpec;
|
use super::LocalOpenAiResponsesSpec;
|
||||||
@@ -42,6 +43,7 @@ pub(crate) use crate::ai_serving::planner::decision_input::LocalRequestedModelDe
|
|||||||
|
|
||||||
pub(crate) async fn resolve_local_openai_responses_decision_input(
|
pub(crate) async fn resolve_local_openai_responses_decision_input(
|
||||||
state: &AppState,
|
state: &AppState,
|
||||||
|
parts: &http::request::Parts,
|
||||||
trace_id: &str,
|
trace_id: &str,
|
||||||
decision: &GatewayControlDecision,
|
decision: &GatewayControlDecision,
|
||||||
body_json: &serde_json::Value,
|
body_json: &serde_json::Value,
|
||||||
@@ -128,6 +130,7 @@ pub(crate) async fn resolve_local_openai_responses_decision_input(
|
|||||||
|
|
||||||
let mut input = build_local_requested_model_decision_input(resolved_input, requested_model);
|
let mut input = build_local_requested_model_decision_input(resolved_input, requested_model);
|
||||||
input.request_auth_channel = decision.request_auth_channel.clone();
|
input.request_auth_channel = decision.request_auth_channel.clone();
|
||||||
|
input.client_session_affinity = client_session_affinity_from_parts(parts, Some(body_json));
|
||||||
Some(input)
|
Some(input)
|
||||||
}
|
}
|
||||||
|
|
||||||
@@ -154,6 +157,7 @@ pub(crate) async fn materialize_local_openai_responses_candidate_attempts(
|
|||||||
spec_metadata.require_streaming,
|
spec_metadata.require_streaming,
|
||||||
input.required_capabilities.as_ref(),
|
input.required_capabilities.as_ref(),
|
||||||
&input.auth_snapshot,
|
&input.auth_snapshot,
|
||||||
|
input.client_session_affinity.as_ref(),
|
||||||
true,
|
true,
|
||||||
LocalCandidatePreselectionKeyMode::ProviderEndpointKeyModelAndApiFormat,
|
LocalCandidatePreselectionKeyMode::ProviderEndpointKeyModelAndApiFormat,
|
||||||
)
|
)
|
||||||
@@ -164,6 +168,7 @@ pub(crate) async fn materialize_local_openai_responses_candidate_attempts(
|
|||||||
spec_metadata.api_format,
|
spec_metadata.api_format,
|
||||||
Some(&input.requested_model),
|
Some(&input.requested_model),
|
||||||
Some(&input.auth_snapshot),
|
Some(&input.auth_snapshot),
|
||||||
|
input.client_session_affinity.as_ref(),
|
||||||
input.required_capabilities.as_ref(),
|
input.required_capabilities.as_ref(),
|
||||||
sticky_session_token.as_deref(),
|
sticky_session_token.as_deref(),
|
||||||
input.request_auth_channel.as_deref(),
|
input.request_auth_channel.as_deref(),
|
||||||
@@ -249,6 +254,7 @@ pub(crate) async fn build_local_openai_responses_candidate_attempt_source<'a>(
|
|||||||
&input.requested_model,
|
&input.requested_model,
|
||||||
spec_metadata.require_streaming,
|
spec_metadata.require_streaming,
|
||||||
&input.auth_snapshot,
|
&input.auth_snapshot,
|
||||||
|
input.client_session_affinity.as_ref(),
|
||||||
input.required_capabilities.as_ref(),
|
input.required_capabilities.as_ref(),
|
||||||
sticky_session_token.as_deref(),
|
sticky_session_token.as_deref(),
|
||||||
input.request_auth_channel.as_deref(),
|
input.request_auth_channel.as_deref(),
|
||||||
|
|||||||
@@ -101,7 +101,7 @@ pub(crate) async fn maybe_build_sync_local_openai_responses_decision_payload(
|
|||||||
};
|
};
|
||||||
|
|
||||||
let Some(input) = resolve_local_openai_responses_decision_input(
|
let Some(input) = resolve_local_openai_responses_decision_input(
|
||||||
state, trace_id, decision, body_json, plan_kind,
|
state, parts, trace_id, decision, body_json, plan_kind,
|
||||||
)
|
)
|
||||||
.await
|
.await
|
||||||
else {
|
else {
|
||||||
@@ -139,7 +139,7 @@ pub(crate) async fn maybe_build_stream_local_openai_responses_decision_payload(
|
|||||||
};
|
};
|
||||||
|
|
||||||
let Some(input) = resolve_local_openai_responses_decision_input(
|
let Some(input) = resolve_local_openai_responses_decision_input(
|
||||||
state, trace_id, decision, body_json, plan_kind,
|
state, parts, trace_id, decision, body_json, plan_kind,
|
||||||
)
|
)
|
||||||
.await
|
.await
|
||||||
else {
|
else {
|
||||||
|
|||||||
@@ -57,6 +57,7 @@ pub(super) async fn build_local_sync_attempt_source<'a>(
|
|||||||
let spec_metadata = local_openai_responses_spec_metadata(spec);
|
let spec_metadata = local_openai_responses_spec_metadata(spec);
|
||||||
let Some(input) = resolve_local_openai_responses_decision_input(
|
let Some(input) = resolve_local_openai_responses_decision_input(
|
||||||
state,
|
state,
|
||||||
|
parts,
|
||||||
trace_id,
|
trace_id,
|
||||||
decision,
|
decision,
|
||||||
body_json,
|
body_json,
|
||||||
@@ -108,6 +109,7 @@ pub(super) async fn build_local_stream_attempt_source<'a>(
|
|||||||
let spec_metadata = local_openai_responses_spec_metadata(spec);
|
let spec_metadata = local_openai_responses_spec_metadata(spec);
|
||||||
let Some(input) = resolve_local_openai_responses_decision_input(
|
let Some(input) = resolve_local_openai_responses_decision_input(
|
||||||
state,
|
state,
|
||||||
|
parts,
|
||||||
trace_id,
|
trace_id,
|
||||||
decision,
|
decision,
|
||||||
body_json,
|
body_json,
|
||||||
@@ -291,6 +293,7 @@ pub(super) async fn build_local_sync_plan_and_reports(
|
|||||||
let spec_metadata = local_openai_responses_spec_metadata(spec);
|
let spec_metadata = local_openai_responses_spec_metadata(spec);
|
||||||
let Some(input) = resolve_local_openai_responses_decision_input(
|
let Some(input) = resolve_local_openai_responses_decision_input(
|
||||||
state,
|
state,
|
||||||
|
parts,
|
||||||
trace_id,
|
trace_id,
|
||||||
decision,
|
decision,
|
||||||
body_json,
|
body_json,
|
||||||
@@ -362,6 +365,7 @@ pub(super) async fn build_local_stream_plan_and_reports(
|
|||||||
let spec_metadata = local_openai_responses_spec_metadata(spec);
|
let spec_metadata = local_openai_responses_spec_metadata(spec);
|
||||||
let Some(input) = resolve_local_openai_responses_decision_input(
|
let Some(input) = resolve_local_openai_responses_decision_input(
|
||||||
state,
|
state,
|
||||||
|
parts,
|
||||||
trace_id,
|
trace_id,
|
||||||
decision,
|
decision,
|
||||||
body_json,
|
body_json,
|
||||||
|
|||||||
Reference in New Issue
Block a user