feat(codex): add optional minimum quota reserve for pool scheduling

apps/aether-gateway/src/handlers/admin/provider/pool/config.rs

@@ -411,6 +411,7 @@ pub(crate) fn admin_provider_pool_config_from_config_value(
             unschedulable_rules: Vec::new(),
             lru_enabled: false,
             skip_exhausted_accounts: false,
+            reserve_minimum_quota: false,
             sticky_session_ttl_seconds: 3600,
             latency_window_seconds: 3600,
             latency_sample_limit: 50,
@@ -446,6 +447,10 @@ pub(crate) fn admin_provider_pool_config_from_config_value(
             .get("skip_exhausted_accounts")
             .and_then(Value::as_bool)
             .unwrap_or(false),
+        reserve_minimum_quota: pool_advanced
+            .get("reserve_minimum_quota")
+            .and_then(Value::as_bool)
+            .unwrap_or(false),
         sticky_session_ttl_seconds: pool_advanced
             .get("sticky_session_ttl_seconds")
             .and_then(json_u64)
@@ -574,6 +579,22 @@ mod tests {
         let config = admin_provider_pool_config(&provider).expect("pool config should exist");
 
         assert!(!config.skip_exhausted_accounts);
+        assert!(!config.reserve_minimum_quota);
+    }
+
+    #[test]
+    fn parses_reserve_minimum_quota_independently_of_skip_exhausted_accounts() {
+        for enabled in [false, true] {
+            let provider = sample_provider(json!({
+                "pool_advanced": {
+                    "reserve_minimum_quota": enabled,
+                    "skip_exhausted_accounts": false
+                }
+            }));
+            let config = admin_provider_pool_config(&provider).expect("pool config should exist");
+            assert_eq!(config.reserve_minimum_quota, enabled);
+            assert!(!config.skip_exhausted_accounts);
+        }
     }
 
     #[test]

apps/aether-gateway/src/handlers/admin/provider/pool/runtime/writes.rs

@@ -651,6 +651,7 @@ mod tests {
             unschedulable_rules: Vec::new(),
             lru_enabled: true,
             skip_exhausted_accounts: false,
+            reserve_minimum_quota: false,
             sticky_session_ttl_seconds: 120,
             latency_window_seconds: 600,
             latency_sample_limit: 10,

apps/aether-gateway/src/handlers/admin/provider/pool_admin/payloads.rs

@@ -1117,10 +1117,16 @@ pub(super) fn build_admin_pool_key_payload(
     let health_score = admin_pool_health_score(key);
     let circuit_breaker_open = false;
     let auth_semantics = provider_key_auth_semantics(key, provider_type);
-    let account_quota_exhausted = pool_config
-        .as_ref()
-        .is_some_and(|config| config.skip_exhausted_accounts)
-        && admin_provider_pool_pure::admin_pool_key_account_quota_exhausted(key, provider_type);
+    let account_quota_exhausted = pool_config.as_ref().is_some_and(|config| {
+        (config.skip_exhausted_accounts
+            && admin_provider_pool_pure::admin_pool_key_account_quota_exhausted(key, provider_type))
+            || (config.reserve_minimum_quota
+                && admin_provider_pool_pure::admin_pool_key_minimum_quota_reached(
+                    key,
+                    provider_type,
+                    None,
+                ))
+    });
     let auth_config = state.parse_catalog_auth_config_json(key);
     let oauth_expires_at =
         admin_pool_derive_oauth_expires_at(provider_type, key, auth_config.as_ref());

apps/aether-gateway/src/handlers/admin/provider/pool_admin/read_routes/keys.rs

@@ -488,9 +488,16 @@ pub(super) fn admin_pool_key_visible_status_filter(
     ) {
         return status;
     }
-    if pool_config.is_some_and(|config| config.skip_exhausted_accounts)
-        && admin_provider_pool_pure::admin_pool_key_account_quota_exhausted(key, provider_type)
-    {
+    if pool_config.is_some_and(|config| {
+        (config.skip_exhausted_accounts
+            && admin_provider_pool_pure::admin_pool_key_account_quota_exhausted(key, provider_type))
+            || (config.reserve_minimum_quota
+                && admin_provider_pool_pure::admin_pool_key_minimum_quota_reached(
+                    key,
+                    provider_type,
+                    None,
+                ))
+    }) {
         return "quota_exhausted";
     }
     if !key.is_active {

apps/aether-gateway/src/handlers/admin/provider/query/models/model_test.rs

@@ -1340,6 +1340,7 @@ async fn provider_query_apply_pool_scheduler_to_test_candidates(
         provider.id.clone(),
         provider_query_ai_pool_runtime_state(&runtime),
     );
+    let reserve_minimum_quota = pool_config.reserve_minimum_quota;
     let pool_config =
         provider_query_ai_pool_scheduling_config(pool_config, provider.provider_type.as_str());
     let inputs = keys
@@ -1351,6 +1352,14 @@ async fn provider_query_apply_pool_scheduler_to_test_candidates(
                 effective_model: effective_model.to_string(),
                 scheduler_skip_reason: None,
             };
+            let mut key_context =
+                provider_query_pool_catalog_key_context(state, &key, &provider.provider_type);
+            key_context.quota_exhausted |= reserve_minimum_quota
+                && admin_provider_pool_pure::admin_pool_key_minimum_quota_reached(
+                    &key,
+                    &provider.provider_type,
+                    Some(effective_model),
+                );
             AiPoolCandidateInput {
                 facts: AiPoolCandidateFacts {
                     provider_id: provider.id.clone(),
@@ -1362,11 +1371,7 @@ async fn provider_query_apply_pool_scheduler_to_test_candidates(
                     key_internal_priority: key.internal_priority,
                 },
                 pool_config: Some(pool_config.clone()),
-                key_context: provider_query_pool_catalog_key_context(
-                    state,
-                    &key,
-                    &provider.provider_type,
-                ),
+                key_context,
                 candidate,
             }
         })

apps/aether-gateway/src/handlers/admin/provider/shared/support.rs

@@ -65,6 +65,7 @@ pub(crate) struct AdminProviderPoolConfig {
     pub(crate) unschedulable_rules: Vec<AdminProviderPoolUnschedulableRule>,
     pub(crate) lru_enabled: bool,
     pub(crate) skip_exhausted_accounts: bool,
+    pub(crate) reserve_minimum_quota: bool,
     pub(crate) sticky_session_ttl_seconds: u64,
     pub(crate) latency_window_seconds: u64,
     pub(crate) latency_sample_limit: u64,