fix(pool): honor quota exhaustion toggle and default ignore to off

apps/aether-gateway/src/dispatch/pool_scheduler.rs

@@ -326,7 +326,9 @@ fn active_probe_member_is_unschedulable_for_request(
         return true;
     }
     key_context.is_some_and(|context| {
-        context.account_blocked || context.quota_exhausted || context.quota_hard_blocked
+        context.account_blocked
+            || (pool_config.skip_exhausted_accounts && context.quota_exhausted)
+            || context.quota_hard_blocked
     })
 }
 
@@ -1529,12 +1531,14 @@ async fn read_pool_catalog_key_contexts_by_id(
                 provider_type,
                 provider_model_name,
             );
-            context.quota_exhausted |= reserve_minimum_quota_key_ids.contains(&key.id)
+            let reserve_reached = reserve_minimum_quota_key_ids.contains(&key.id)
                 && admin_provider_pool_pure::admin_pool_key_minimum_quota_reached(
                     &key,
                     provider_type,
                     provider_model_name,
                 );
+            context.quota_exhausted |= reserve_reached;
+            context.quota_hard_blocked |= reserve_reached;
             (key.id.clone(), context)
         })
         .collect::<BTreeMap<_, _>>();
@@ -2303,6 +2307,31 @@ mod tests {
         );
     }
 
+    #[test]
+    fn pool_scheduler_allows_quota_exhausted_key_when_ignored() {
+        let exhausted = sample_eligible_candidate(
+            "provider-pool",
+            "endpoint-1",
+            "key-exhausted",
+            10,
+            Some(json!({ "pool_advanced": { "ignore_exhausted_accounts": true } })),
+        );
+        let contexts = BTreeMap::from([(
+            "key-exhausted".to_string(),
+            PoolCatalogKeyContext {
+                quota_exhausted: true,
+                ..PoolCatalogKeyContext::default()
+            },
+        )]);
+        let (scheduled, skipped) = apply_local_execution_pool_scheduler_with_runtime_map(
+            vec![exhausted],
+            &BTreeMap::new(),
+            &contexts,
+        );
+        assert_eq!(scheduled.len(), 1);
+        assert!(skipped.is_empty());
+    }
+
     #[test]
     fn pool_scheduler_skips_quota_exhausted_key_when_flag_is_false() {
         let ready = sample_eligible_candidate(

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

@@ -410,7 +410,7 @@ pub(crate) fn admin_provider_pool_config_from_config_value(
             }],
             unschedulable_rules: Vec::new(),
             lru_enabled: false,
-            skip_exhausted_accounts: false,
+            skip_exhausted_accounts: true,
             reserve_minimum_quota: false,
             sticky_session_ttl_seconds: 3600,
             latency_window_seconds: 3600,
@@ -444,9 +444,10 @@ pub(crate) fn admin_provider_pool_config_from_config_value(
         scheduling_presets,
         unschedulable_rules,
         skip_exhausted_accounts: pool_advanced
-            .get("skip_exhausted_accounts")
+            .get("ignore_exhausted_accounts")
             .and_then(Value::as_bool)
-            .unwrap_or(false),
+            .map(|ignore| !ignore)
+            .unwrap_or(true),
         reserve_minimum_quota: pool_advanced
             .get("reserve_minimum_quota")
             .and_then(Value::as_bool)
@@ -574,11 +575,25 @@ mod tests {
     }
 
     #[test]
-    fn defaults_skip_exhausted_accounts_to_false() {
+    fn parses_ignore_exhausted_accounts_with_default_off() {
+        for ignore in [false, true] {
+            let provider = sample_provider(json!({
+                "pool_advanced": {
+                    "ignore_exhausted_accounts": ignore,
+                    "skip_exhausted_accounts": true
+                }
+            }));
+            let config = admin_provider_pool_config(&provider).expect("pool config should exist");
+            assert_eq!(config.skip_exhausted_accounts, !ignore);
+        }
+    }
+
+    #[test]
+    fn defaults_skip_exhausted_accounts_to_true() {
         let provider = sample_provider(json!({ "pool_advanced": {} }));
         let config = admin_provider_pool_config(&provider).expect("pool config should exist");
 
-        assert!(!config.skip_exhausted_accounts);
+        assert!(config.skip_exhausted_accounts);
         assert!(!config.reserve_minimum_quota);
     }
 
@@ -593,7 +608,7 @@ mod tests {
             }));
             let config = admin_provider_pool_config(&provider).expect("pool config should exist");
             assert_eq!(config.reserve_minimum_quota, enabled);
-            assert!(!config.skip_exhausted_accounts);
+            assert!(config.skip_exhausted_accounts);
         }
     }
 

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

@@ -385,9 +385,10 @@ async fn read_provider_pool_state_map(
                 .and_then(|value| value.get("pool_advanced"));
             let skip_exhausted_accounts = pool_advanced
                 .and_then(serde_json::Value::as_object)
-                .and_then(|value| value.get("skip_exhausted_accounts"))
+                .and_then(|value| value.get("ignore_exhausted_accounts"))
                 .and_then(serde_json::Value::as_bool)
-                .unwrap_or(false);
+                .map(|ignore| !ignore)
+                .unwrap_or(true);
             let reserve_minimum_quota = pool_advanced
                 .and_then(serde_json::Value::as_object)
                 .and_then(|value| value.get("reserve_minimum_quota"))

crates/aether-pool-core/src/scheduler.rs

@@ -18,8 +18,6 @@ pub struct PoolSchedulingPreset {
 pub struct PoolSchedulingConfig {
     pub scheduling_presets: Vec<PoolSchedulingPreset>,
     pub lru_enabled: bool,
-    /// Retained for configuration/API compatibility. Active quota exhaustion is
-    /// always an admission block; reset-aware adapters decide when it clears.
     pub skip_exhausted_accounts: bool,
     pub cost_limit_per_key_tokens: Option<u64>,
 }
@@ -227,14 +225,9 @@ fn schedule_pool_group<Candidate>(
             continue;
         }
 
-        // A quota snapshot is an account-level admission signal, not merely a
+        if item.key_context.quota_hard_blocked
-        // ranking hint. Continuing to schedule a member whose quota is known to
+            || (pool_config.skip_exhausted_accounts && item.key_context.quota_exhausted)
-        // be exhausted causes a request-wide retry storm (the upstream returns
+        {
-        // 429 for every attempt). `quota_hard_blocked` remains available for
-        // providers that can distinguish an explicit permanent block, but every
-        // active exhaustion must be removed from the request's candidate set;
-        // reset-aware provider adapters clear the signal once capacity returns.
-        if item.key_context.quota_hard_blocked || item.key_context.quota_exhausted {
             skipped.push(PoolSkippedCandidate {
                 candidate: item.candidate,
                 skip_reason: POOL_ACCOUNT_EXHAUSTED_SKIP_REASON,
@@ -936,7 +929,30 @@ mod tests {
     }
 
     #[test]
-    fn pool_scheduler_skips_exhausted_accounts_even_when_legacy_flag_is_false() {
+    fn pool_scheduler_allows_exhausted_accounts_when_ignored_but_keeps_hard_blocks() {
+        let mut exhausted =
+            sample_candidate("provider-pool", "endpoint-1", "key-exhausted", 10, true);
+        exhausted.key_context.quota_exhausted = true;
+        exhausted
+            .pool_config
+            .as_mut()
+            .unwrap()
+            .skip_exhausted_accounts = false;
+        let mut blocked = exhausted.clone();
+        blocked.candidate = "key-blocked".to_string();
+        blocked.facts.key_id = "key-blocked".to_string();
+        blocked.key_context.quota_hard_blocked = true;
+
+        let outcome = run_pool_scheduler(vec![exhausted, blocked], &BTreeMap::new(), "seed");
+
+        assert_eq!(outcome.candidates.len(), 1);
+        assert_eq!(outcome.candidates[0].candidate, "key-exhausted");
+        assert_eq!(outcome.skipped_candidates.len(), 1);
+        assert_eq!(outcome.skipped_candidates[0].candidate, "key-blocked");
+    }
+
+    #[test]
+    fn pool_scheduler_skips_exhausted_accounts_by_default() {
         let ready = sample_candidate("provider-pool", "endpoint-1", "key-ready", 10, true);
         let mut exhausted =
             sample_candidate("provider-pool", "endpoint-1", "key-exhausted", 10, true);
@@ -1535,7 +1551,7 @@ mod tests {
         let pool_config = pool_enabled.then(|| PoolSchedulingConfig {
             scheduling_presets: Vec::new(),
             lru_enabled: true,
-            skip_exhausted_accounts: false,
+            skip_exhausted_accounts: true,
             cost_limit_per_key_tokens: None,
         });
         PoolCandidateInput {

frontend/src/api/endpoints/types/provider.ts

@@ -836,6 +836,7 @@ export interface PoolAdvancedConfig {
   sticky_session_ttl_seconds?: number | null
   load_threshold_percent?: number | null
   skip_exhausted_accounts?: boolean | null
+  ignore_exhausted_accounts?: boolean | null
   // Codex only: treat remaining quota <= 1% as exhausted (default false).
   reserve_minimum_quota?: boolean
   // 旧字段（兼容读取）

frontend/src/features/pool/components/PoolAdvancedDialog.vue

@@ -591,7 +591,7 @@ const form = ref({
   account_self_check_concurrency: null as number | null | undefined,
   auto_remove_banned_keys: false,
   auto_remove_quota_exhausted_keys: false,
-  skip_exhausted_accounts: false,
+  ignore_exhausted_accounts: false,
 })
 
 interface ClaudeFormState {
@@ -630,8 +630,8 @@ function getHealthToggleValue(key: PoolHealthToggleKey): boolean {
       return form.value.auto_remove_banned_keys
     case 'auto_remove_quota_exhausted_keys':
       return form.value.auto_remove_quota_exhausted_keys
-    case 'skip_exhausted_accounts':
+    case 'ignore_exhausted_accounts':
-      return form.value.skip_exhausted_accounts
+      return form.value.ignore_exhausted_accounts
   }
 }
 
@@ -649,8 +649,8 @@ function updateHealthToggleValue(key: PoolHealthToggleKey, value: boolean): void
     case 'auto_remove_quota_exhausted_keys':
       form.value.auto_remove_quota_exhausted_keys = value
       return
-    case 'skip_exhausted_accounts':
+    case 'ignore_exhausted_accounts':
-      form.value.skip_exhausted_accounts = value
+      form.value.ignore_exhausted_accounts = value
   }
 }
 
@@ -686,7 +686,7 @@ watch([() => props.modelValue, () => props.providerId], ([open]) => {
     account_self_check_concurrency: cfg?.account_self_check_concurrency ?? null,
     auto_remove_banned_keys: cfg?.auto_remove_banned_keys ?? false,
     auto_remove_quota_exhausted_keys: cfg?.auto_remove_quota_exhausted_keys ?? false,
-    skip_exhausted_accounts: cfg?.skip_exhausted_accounts ?? false,
+    ignore_exhausted_accounts: cfg?.ignore_exhausted_accounts ?? false,
   }
 
   const cc = props.currentClaudeConfig
@@ -754,7 +754,7 @@ async function handleSave() {
         : undefined,
       auto_remove_banned_keys: form.value.auto_remove_banned_keys,
       auto_remove_quota_exhausted_keys: form.value.auto_remove_quota_exhausted_keys,
-      skip_exhausted_accounts: form.value.skip_exhausted_accounts,
+      ignore_exhausted_accounts: form.value.ignore_exhausted_accounts,
     })
 
     const payload: Parameters<typeof updateProvider>[1] = {

frontend/src/features/pool/utils/__tests__/poolAdvancedDialog.spec.ts

@@ -12,7 +12,7 @@ describe('poolAdvancedDialog', () => {
       'account_self_check_enabled',
       'auto_remove_banned_keys',
       'auto_remove_quota_exhausted_keys',
-      'skip_exhausted_accounts',
+      'ignore_exhausted_accounts',
     ])
   })
 
@@ -39,9 +39,9 @@ describe('poolAdvancedDialog', () => {
         description: '探测到黑色“额度耗尽”账号后自动从号池移除。',
       },
       {
-        key: 'skip_exhausted_accounts',
+        key: 'ignore_exhausted_accounts',
-        label: '跳过额度耗尽账号',
+        label: '忽略额度耗尽',
-        description: '当 Codex / Kiro 账号额度已耗尽时，直接标记为不可调度并在请求侧跳过。',
+        description: '默认关闭，额度耗尽账号不参与调度。开启后忽略额度耗尽状态，仍保留封禁、冷却和最低额度保留等限制。',
       },
     ])
   })

frontend/src/features/pool/utils/poolAdvancedDialog.ts

@@ -3,7 +3,7 @@ export type PoolHealthToggleKey =
   | 'account_self_check_enabled'
   | 'auto_remove_banned_keys'
   | 'auto_remove_quota_exhausted_keys'
-  | 'skip_exhausted_accounts'
+  | 'ignore_exhausted_accounts'
 
 export interface PoolHealthToggleCard {
   key: PoolHealthToggleKey
@@ -39,9 +39,9 @@ export function buildPoolHealthToggleCards(): PoolHealthToggleCard[] {
       description: '探测到黑色“额度耗尽”账号后自动从号池移除。',
     },
     {
-      key: 'skip_exhausted_accounts',
+      key: 'ignore_exhausted_accounts',
-      label: '跳过额度耗尽账号',
+      label: '忽略额度耗尽',
-      description: '当 Codex / Kiro 账号额度已耗尽时，直接标记为不可调度并在请求侧跳过。',
+      description: '默认关闭，额度耗尽账号不参与调度。开启后忽略额度耗尽状态，仍保留封禁、冷却和最低额度保留等限制。',
     },
   ]
 }