feat(codex): 按操作语义路由 Responses V2 压缩

(cherry picked from commit 2fc604e047)

apps/aether-gateway/src/scheduler/candidate/enumeration.rs

@@ -2,7 +2,7 @@ use aether_scheduler_core::SchedulerMinimalCandidateSelectionCandidate;
 
 use crate::data::auth::GatewayAuthApiKeySnapshot;
 use crate::data::candidate_selection::{
-    enumerate_minimal_candidate_selection_with_required_capabilities,
+    enumerate_minimal_candidate_selection_with_required_capabilities_for_request_operation,
     MinimalCandidateSelectionRowSource,
 };
 use crate::GatewayError;
@@ -15,8 +15,9 @@ pub(super) async fn enumerate_scheduler_candidates(
     required_capabilities: Option<&serde_json::Value>,
     auth_snapshot: Option<&GatewayAuthApiKeySnapshot>,
     enable_model_directives: bool,
+    request_operation: Option<&str>,
 ) -> Result<Vec<SchedulerMinimalCandidateSelectionCandidate>, GatewayError> {
-    enumerate_minimal_candidate_selection_with_required_capabilities(
+    enumerate_minimal_candidate_selection_with_required_capabilities_for_request_operation(
         selection_row_source,
         api_format,
         global_model_name,
@@ -24,6 +25,7 @@ pub(super) async fn enumerate_scheduler_candidates(
         auth_snapshot,
         required_capabilities,
         enable_model_directives,
+        request_operation,
     )
     .await
     .map_err(|err| GatewayError::Internal(err.to_string()))

apps/aether-gateway/src/scheduler/candidate/mod.rs

@@ -116,6 +116,42 @@ pub(crate) async fn list_selectable_candidates_with_skip_reasons(
         client_session_affinity,
         now_unix_secs,
         enable_model_directives,
+        None,
+    )
+    .await
+}
+
+pub(crate) async fn list_selectable_candidates_with_skip_reasons_for_request_operation(
+    selection_row_source: &(impl MinimalCandidateSelectionRowSource + Sync),
+    runtime_state: &impl SchedulerRuntimeState,
+    api_format: &str,
+    global_model_name: &str,
+    require_streaming: bool,
+    required_capabilities: Option<&serde_json::Value>,
+    auth_snapshot: Option<&GatewayAuthApiKeySnapshot>,
+    client_session_affinity: Option<&ClientSessionAffinity>,
+    now_unix_secs: u64,
+    enable_model_directives: bool,
+    request_operation: Option<&str>,
+) -> Result<
+    (
+        Vec<SchedulerMinimalCandidateSelectionCandidate>,
+        Vec<SchedulerSkippedCandidate>,
+    ),
+    GatewayError,
+> {
+    collect_selectable_candidates_with_skip_reasons(
+        selection_row_source,
+        runtime_state,
+        api_format,
+        global_model_name,
+        require_streaming,
+        required_capabilities,
+        auth_snapshot,
+        client_session_affinity,
+        now_unix_secs,
+        enable_model_directives,
+        request_operation,
     )
     .await
 }
@@ -237,6 +273,7 @@ pub(crate) async fn list_selectable_candidates_for_required_capability_without_r
             client_session_affinity,
             now_unix_secs,
             false,
+            None,
         )
         .await?;
         all_attempts_blocked_by_auth_limit &=

apps/aether-gateway/src/scheduler/candidate/selection.rs

@@ -81,6 +81,7 @@ pub(super) async fn select_minimal_candidate(
         required_capabilities,
         auth_snapshot,
         enable_model_directives,
+        None,
     )
     .await?;
     let selected = collect_selectable_enumerated_candidates_with_skip_reasons(
@@ -137,6 +138,7 @@ pub(super) async fn collect_selectable_candidates(
         client_session_affinity,
         now_unix_secs,
         enable_model_directives,
+        None,
     )
     .await?
     .0)
@@ -153,6 +155,7 @@ pub(super) async fn collect_selectable_candidates_with_skip_reasons(
     client_session_affinity: Option<&ClientSessionAffinity>,
     now_unix_secs: u64,
     enable_model_directives: bool,
+    request_operation: Option<&str>,
 ) -> Result<
     (
         Vec<SchedulerMinimalCandidateSelectionCandidate>,
@@ -174,6 +177,7 @@ pub(super) async fn collect_selectable_candidates_with_skip_reasons(
         required_capabilities,
         auth_snapshot,
         enable_model_directives,
+        request_operation,
     )
     .await?;
     collect_selectable_enumerated_candidates_with_skip_reasons(

apps/aether-gateway/src/scheduler/candidate/tests/affinity.rs

@@ -107,6 +107,7 @@ async fn same_priority_candidates_are_distributed_by_affinity_key() {
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: None,
+        operations: None,
     }]);
 
     let mut second = sample_row();
@@ -124,6 +125,7 @@ async fn same_priority_candidates_are_distributed_by_affinity_key() {
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: None,
+        operations: None,
     }]);
 
     let candidates = Arc::new(InMemoryMinimalCandidateSelectionReadRepository::seed(vec![
@@ -142,6 +144,7 @@ async fn same_priority_candidates_are_distributed_by_affinity_key() {
         enumerate_minimal_candidate_selection(EnumerateMinimalCandidateSelectionInput {
             rows,
             normalized_api_format: "openai:chat",
+            request_operation: None,
             requested_model_name: "gpt-4.1",
             resolved_global_model_name: "gpt-4.1",
             require_streaming: false,

apps/aether-gateway/src/scheduler/candidate/tests/model.rs

@@ -45,6 +45,7 @@ fn provider_model_mapping_respects_endpoint_scope() {
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: Some(vec!["endpoint-selected".to_string()]),
+        operations: None,
     }]);
 
     assert_eq!(
@@ -101,6 +102,7 @@ fn resolves_requested_global_model_from_provider_model_alias() {
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: None,
+        operations: None,
     }]);
 
     let resolved = resolve_requested_global_model_name(&[row], "gpt-5.2", "openai:chat");
@@ -156,6 +158,7 @@ async fn enumerate_minimal_candidate_selection_resolves_provider_model_alias() {
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: None,
+        operations: None,
     }]);
 
     let candidates = Arc::new(InMemoryMinimalCandidateSelectionReadRepository::seed(vec![
@@ -193,6 +196,7 @@ async fn enumerate_minimal_candidate_selection_filters_endpoint_scoped_alias_row
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: Some(vec!["endpoint-selected".to_string()]),
+        operations: None,
     }]);
 
     let mut other = selected.clone();
@@ -248,6 +252,7 @@ async fn enumerate_minimal_candidate_selection_keeps_only_resolved_global_model_
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: None,
+        operations: None,
     }]);
 
     let candidates = Arc::new(InMemoryMinimalCandidateSelectionReadRepository::seed(vec![
@@ -287,6 +292,7 @@ async fn enumerate_minimal_candidate_selection_allows_resolved_global_model_in_a
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: None,
+        operations: None,
     }]);
 
     let candidates = Arc::new(InMemoryMinimalCandidateSelectionReadRepository::seed(vec![

apps/aether-gateway/src/scheduler/candidate/tests/selection.rs

@@ -105,6 +105,7 @@ async fn collect_selectable_candidates_with_skip_reasons(
         None,
         now_unix_secs,
         false,
+        None,
     )
     .await
 }
@@ -843,6 +844,7 @@ async fn selects_next_candidate_when_first_provider_quota_is_exhausted() {
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: None,
+        operations: None,
     }]);
     first.key_global_priority_by_format = Some(serde_json::json!({"openai:chat": 1}));
 
@@ -858,6 +860,7 @@ async fn selects_next_candidate_when_first_provider_quota_is_exhausted() {
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: None,
+        operations: None,
     }]);
     second.key_global_priority_by_format = Some(serde_json::json!({"openai:chat": 2}));
 
@@ -914,6 +917,7 @@ async fn cooled_down_when_recent_failures_are_recorded_for_same_key() {
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: None,
+        operations: None,
     }]);
     first.key_global_priority_by_format = Some(serde_json::json!({"openai:chat": 1}));
 
@@ -929,6 +933,7 @@ async fn cooled_down_when_recent_failures_are_recorded_for_same_key() {
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: None,
+        operations: None,
     }]);
     second.key_global_priority_by_format = Some(serde_json::json!({"openai:chat": 2}));
 

apps/aether-gateway/src/scheduler/candidate/tests/support.rs

@@ -40,12 +40,14 @@ pub(super) fn sample_row() -> StoredMinimalCandidateSelectionRow {
                 priority: 1,
                 api_formats: Some(vec!["openai:chat".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             },
             StoredProviderModelMapping {
                 name: "gpt-4.1-responses".to_string(),
                 priority: 1,
                 api_formats: Some(vec!["openai:responses".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             },
         ]),
         model_supports_streaming: None,