Merge pull request #852 from zhefox/main

fix(codex): restore recovered quota status and stabilize lifecycle tests

Cargo.lock

@@ -524,6 +524,7 @@ dependencies = [
  "aether-data-contracts",
  "aether-pool-core",
  "aether-provider-transport",
+ "chrono",
  "serde_json",
  "url",
  "uuid",

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

@@ -4005,7 +4005,7 @@ mod tests {
     async fn pool_key_cursor_reserve_minimum_quota_filters_pages_and_sticky_hits() {
         for reserve_enabled in [false, true] {
             for sticky in [false, true] {
-                for used_percent in [99.0, 98.0] {
+                for used_percent in [99.0, 98.0, 83.0] {
                     let provider_config = Some(json!({
                         "pool_advanced": {
                             "reserve_minimum_quota": reserve_enabled,
@@ -4016,8 +4016,27 @@ mod tests {
                         sample_codex_pool_provider("provider-pool", 0, provider_config.clone());
                     let endpoint = sample_codex_pool_endpoint("provider-pool", "endpoint-1");
                     let mut reserved = sample_codex_pool_key("provider-pool", "key-low");
+                    reserved.status_snapshot = Some(json!({
+                        "quota": {
+                            "provider_type": "codex",
+                            "updated_at": 100,
+                            "allowed": false,
+                            "exhausted": true,
+                            "code": "exhausted",
+                            "windows": [{
+                                "code": "weekly",
+                                "scope": "account",
+                                "used_ratio": 1.0,
+                                "reset_at": 4_102_444_800u64
+                            }]
+                        }
+                    }));
                     reserved.upstream_metadata = Some(json!({
-                        "codex": {"primary_used_percent": used_percent}
+                        "codex": {
+                            "updated_at": 200,
+                            "primary_used_percent": used_percent,
+                            "primary_reset_at": 4_102_444_800u64
+                        }
                     }));
                     let ready = sample_codex_pool_key("provider-pool", "key-ready");
                     let rows = vec![

apps/aether-gateway/src/handlers/shared/catalog.rs

@@ -21,6 +21,7 @@ use aether_crypto::{
 use aether_data_contracts::repository::provider_catalog::StoredProviderCatalogKey;
 use aether_provider_pool::{
     grok_pool_tier_from_quota_bucket, grok_supported_quota_windows_for_tier,
+    provider_pool_codex_metadata_has_account_quota,
 };
 use aether_scheduler_core::provider_key_circuit_payload_is_active_open_at;
 use serde_json::{json, Map, Value};
@@ -1111,9 +1112,8 @@ fn build_codex_quota_status_snapshot(
     source: &str,
 ) -> Option<Value> {
     let metadata = provider_quota_metadata_bucket(upstream_metadata, "codex")?;
-    let observed_at_unix_secs = metadata
+    let observed_at_unix_secs = provider_quota_timestamp_unix_secs(metadata.get("observed_at"))
-        .get("updated_at")
+        .or_else(|| provider_quota_timestamp_unix_secs(metadata.get("updated_at")));
-        .and_then(admin_provider_quota_pure::coerce_json_u64);
     let plan_type = metadata
         .get("plan_type")
         .and_then(Value::as_str)
@@ -2502,17 +2502,19 @@ fn codex_upstream_metadata_is_at_least_as_fresh(
     let Some(metadata) = provider_quota_metadata_bucket(upstream_metadata, "codex") else {
         return false;
     };
-    let Some(metadata_updated_at) = metadata
+    // Identity, reset-credit, and model-only updates do not replace the
-        .get("updated_at")
+    // account's quota observation, even when their timestamp is newer.
-        .and_then(admin_provider_quota_pure::coerce_json_u64)
+    if !provider_pool_codex_metadata_has_account_quota(metadata) {
+        return false;
+    }
+    let Some(metadata_updated_at) = provider_quota_timestamp_unix_secs(metadata.get("observed_at"))
+        .or_else(|| provider_quota_timestamp_unix_secs(metadata.get("updated_at")))
     else {
         return false;
     };
     let snapshot_updated_at = quota_snapshot.and_then(|quota| {
-        quota
+        provider_quota_timestamp_unix_secs(quota.get("observed_at"))
-            .get("updated_at")
+            .or_else(|| provider_quota_timestamp_unix_secs(quota.get("updated_at")))
-            .or_else(|| quota.get("observed_at"))
-            .and_then(admin_provider_quota_pure::coerce_json_u64)
     });
 
     snapshot_updated_at.is_none_or(|updated_at| metadata_updated_at >= updated_at)
@@ -2611,6 +2613,19 @@ pub(crate) fn provider_key_status_snapshot_payload(
     let mut snapshot = provider_key_status_snapshot_object(Some(&payload))
         .or_else(|| default_provider_key_status_snapshot().as_object().cloned())
         .unwrap_or_default();
+    // Legacy snapshots can retain an exhausted summary after a window reset or
+    // newer quota observation. Use the same decision as scheduling so the
+    // account list and its status filter do not keep displaying that stale block.
+    if provider_type.trim().eq_ignore_ascii_case("codex")
+        && !aether_provider_pool::provider_pool_key_account_quota_exhausted(key, provider_type)
+    {
+        if let Some(quota) = snapshot.get_mut("quota").and_then(Value::as_object_mut) {
+            quota.insert("exhausted".to_string(), json!(false));
+            if quota.get("code").and_then(Value::as_str) == Some("exhausted") {
+                quota.insert("code".to_string(), json!("ok"));
+            }
+        }
+    }
     snapshot.insert(
         "oauth".to_string(),
         build_provider_key_oauth_status_snapshot(key),
@@ -4213,6 +4228,116 @@ mod tests {
         );
     }
 
+    #[test]
+    fn provider_key_status_snapshot_payload_clears_stale_codex_exhaustion_summary() {
+        let mut key = sample_catalog_key();
+        key.status_snapshot = Some(json!({
+            "quota": {
+                "provider_type": "codex",
+                "code": "exhausted",
+                "exhausted": true,
+                "windows": [{
+                    "code": "weekly",
+                    "scope": "account",
+                    "used_ratio": 0.83,
+                    "remaining_ratio": 0.17,
+                    "reset_at": 4_102_444_800u64
+                }]
+            }
+        }));
+        let payload = provider_key_status_snapshot_payload(&key, "codex");
+        assert_eq!(payload["quota"]["code"], "ok");
+        assert_eq!(payload["quota"]["exhausted"], false);
+        assert!(payload["quota"]["label"].is_null());
+
+        // An explicit current upstream refusal is not a stale percentage summary.
+        key.status_snapshot.as_mut().unwrap()["quota"]["allowed"] = json!(false);
+        let payload = provider_key_status_snapshot_payload(&key, "codex");
+        assert_eq!(payload["quota"]["code"], "exhausted");
+        assert_eq!(payload["quota"]["exhausted"], true);
+
+        // Missing capacity evidence must not clear an exhausted summary either.
+        key.status_snapshot = Some(json!({"quota": {
+            "provider_type": "codex", "code": "exhausted", "exhausted": true
+        }}));
+        let payload = provider_key_status_snapshot_payload(&key, "codex");
+        assert_eq!(payload["quota"]["code"], "exhausted");
+        assert_eq!(payload["quota"]["exhausted"], true);
+    }
+
+    #[test]
+    fn provider_key_status_snapshot_payload_refreshes_codex_timestamp_formats() {
+        for updated_at in [
+            json!(1_900_000_000u64),
+            json!(1_900_000_000_000u64),
+            json!("2030-03-17T17:46:40Z"),
+        ] {
+            let mut key = sample_catalog_key();
+            key.upstream_metadata = Some(json!({
+                "codex": {
+                    "updated_at": updated_at,
+                    "primary_used_percent": 83.0,
+                    "primary_reset_at": 4_102_444_800u64
+                }
+            }));
+            key.status_snapshot = Some(json!({
+                "quota": {
+                    "provider_type": "codex",
+                    "updated_at": 1_899_999_000u64,
+                    "code": "exhausted",
+                    "exhausted": true,
+                    "allowed": false,
+                    "windows": [{
+                        "code": "weekly",
+                        "scope": "account",
+                        "used_ratio": 1.0,
+                        "remaining_ratio": 0.0,
+                        "reset_at": 4_102_444_800u64
+                    }]
+                }
+            }));
+            let payload = provider_key_status_snapshot_payload(&key, "codex");
+            assert_eq!(payload["quota"]["code"], "ok");
+            assert_eq!(payload["quota"]["updated_at"], 1_900_000_000u64);
+            assert_eq!(payload["quota"]["windows"][0]["used_ratio"], 0.83);
+            assert!(payload["quota"]["allowed"].is_null());
+        }
+    }
+
+    #[test]
+    fn provider_key_status_snapshot_payload_preserves_codex_account_quota_on_unrelated_updates() {
+        for patch in [
+            json!({"plan_type": "pro"}),
+            json!({"spark_primary_used_percent": 83.0}),
+            json!({"credits_unlimited": false}),
+            json!({"windows": [{"code": "weekly", "reset_at": 4_102_444_800u64}]}),
+        ] {
+            let mut metadata = patch;
+            metadata["updated_at"] = json!("2030-03-17T17:46:40Z");
+            let mut key = sample_catalog_key();
+            key.upstream_metadata = Some(json!({"codex": metadata}));
+            key.status_snapshot = Some(json!({"quota": {
+                "provider_type": "codex", "updated_at": 200,
+                "code": "exhausted", "exhausted": true, "allowed": false,
+                "windows": [{"code": "weekly", "scope": "account", "used_ratio": 1.0,
+                    "remaining_ratio": 0.0, "reset_at": 4_102_444_800u64}]
+            }}));
+            let payload = provider_key_status_snapshot_payload(&key, "codex");
+            assert_eq!(payload["quota"]["code"], "exhausted", "{metadata}");
+            assert_eq!(payload["quota"]["exhausted"], true, "{metadata}");
+            assert_eq!(payload["quota"]["allowed"], false, "{metadata}");
+            assert_eq!(payload["quota"]["updated_at"], 200, "{metadata}");
+            assert_eq!(
+                payload["quota"]["windows"][0]["code"], "weekly",
+                "{metadata}"
+            );
+            assert_eq!(
+                payload["quota"]["windows"][0]["used_ratio"], 1.0,
+                "{metadata}"
+            );
+        }
+    }
+
     #[test]
     fn provider_key_status_snapshot_payload_restores_complete_codex_cache() {
         let mut key = sample_catalog_key();

apps/aether-gateway/src/tests/control/admin/pool.rs

@@ -2436,6 +2436,134 @@ async fn gateway_reserve_minimum_quota_marks_and_filters_codex_pool_keys() {
     }
 }
 
+#[tokio::test]
+async fn gateway_reserve_minimum_quota_recovers_after_config_update_with_cached_catalog() {
+    let mut provider = sample_provider("provider-codex", "codex", 10);
+    provider.provider_type = "codex".to_string();
+    provider.config = Some(json!({
+        "pool_advanced": {"reserve_minimum_quota": true, "skip_exhausted_accounts": true}
+    }));
+    let mut key = sample_key(
+        "key-codex",
+        "provider-codex",
+        "openai:responses",
+        "oauth-placeholder",
+    );
+    key.auth_type = "oauth".to_string();
+    key.status_snapshot = Some(json!({
+        "quota": {
+            "provider_type": "codex", "updated_at": 100,
+            "code": "exhausted", "exhausted": true, "allowed": false,
+            "windows": [{
+                "code": "weekly", "scope": "account", "used_ratio": 1.0,
+                "remaining_ratio": 0.0, "reset_at": 4_102_444_800u64
+            }]
+        }
+    }));
+    key.upstream_metadata = Some(json!({
+        "codex": {"updated_at": 200, "primary_used_percent": 99.0,
+                  "primary_reset_at": 4_102_444_800u64}
+    }));
+    let repository = Arc::new(InMemoryProviderCatalogReadRepository::seed(
+        vec![provider.clone()],
+        Vec::new(),
+        vec![key],
+    ));
+    let state = AppState::new()
+        .expect("gateway should build")
+        .with_data_state_for_tests(
+            GatewayDataState::with_provider_catalog_repository_for_tests(Arc::clone(&repository))
+                .with_cached_provider_catalog_reader_for_tests(repository),
+        );
+    for reserve_enabled in [true, false, true] {
+        provider.config.as_mut().unwrap()["pool_advanced"]["reserve_minimum_quota"] =
+            json!(reserve_enabled);
+        state
+            .update_provider_catalog_provider(&provider)
+            .await
+            .expect("provider should update");
+        for status in ["all", "quota_exhausted", "available"] {
+            let response = local_admin_pool_response(
+                &state,
+                http::Method::GET,
+                &format!("/api/admin/pool/provider-codex/keys?status={status}"),
+                None,
+            )
+            .await;
+            assert_eq!(response.status(), StatusCode::OK);
+            let payload: serde_json::Value = serde_json::from_slice(
+                &to_bytes(response.into_body(), usize::MAX)
+                    .await
+                    .expect("body should read"),
+            )
+            .expect("json should parse");
+            let keys = payload["keys"].as_array().expect("keys should be array");
+            let visible = status == "all" || (status == "quota_exhausted") == reserve_enabled;
+            assert_eq!(keys.len(), usize::from(visible));
+            if visible {
+                assert_eq!(
+                    keys[0]["scheduling_reason"] == "account_quota_exhausted",
+                    reserve_enabled
+                );
+                assert_eq!(keys[0]["status_snapshot"]["quota"]["code"], "ok");
+            }
+        }
+    }
+}
+
+#[tokio::test]
+async fn gateway_codex_pool_filter_clears_stale_exhausted_summary_with_remaining_quota() {
+    let mut provider = sample_provider("provider-codex", "codex", 10);
+    provider.provider_type = "codex".to_string();
+    provider.config = Some(json!({"pool_advanced": {
+        "reserve_minimum_quota": false, "skip_exhausted_accounts": true
+    }}));
+    let mut key = sample_key(
+        "key-codex",
+        "provider-codex",
+        "openai:responses",
+        "oauth-placeholder",
+    );
+    key.auth_type = "oauth".to_string();
+    key.status_snapshot = Some(json!({"quota": {
+        "provider_type": "codex", "code": "exhausted", "exhausted": true,
+        "windows": [{"code": "weekly", "scope": "account", "used_ratio": 0.83,
+            "remaining_ratio": 0.17, "reset_at": 4_102_444_800u64}]
+    }}));
+    let state = AppState::new()
+        .expect("gateway should build")
+        .with_data_state_for_tests(GatewayDataState::with_provider_catalog_reader_for_tests(
+            Arc::new(InMemoryProviderCatalogReadRepository::seed(
+                vec![provider],
+                Vec::new(),
+                vec![key],
+            )),
+        ));
+    for status in ["all", "available", "quota_exhausted"] {
+        let response = local_admin_pool_response(
+            &state,
+            http::Method::GET,
+            &format!("/api/admin/pool/provider-codex/keys?status={status}"),
+            None,
+        )
+        .await;
+        assert_eq!(response.status(), StatusCode::OK);
+        let payload: serde_json::Value = serde_json::from_slice(
+            &to_bytes(response.into_body(), usize::MAX)
+                .await
+                .expect("body should read"),
+        )
+        .expect("json should parse");
+        let keys = payload["keys"].as_array().expect("keys should be array");
+        assert_eq!(keys.len(), usize::from(status != "quota_exhausted"));
+        if let Some(key) = keys.first() {
+            assert_eq!(key["scheduling_status"], "available");
+            assert_eq!(key["status_snapshot"]["quota"]["code"], "ok");
+            assert!(key["account_quota"].as_str().unwrap().contains("17.0%"));
+        }
+    }
+}
+
 #[tokio::test]
 async fn gateway_lists_inherited_fixed_provider_api_formats_for_pool_keys() {
     let mut provider = sample_provider("provider-codex", "codex", 10).with_transport_fields(
@@ -2738,8 +2866,10 @@ async fn gateway_treats_stale_codex_exhausted_snapshot_as_available_when_windows
     assert_eq!(keys[0]["scheduling_reason"], json!("available"));
     assert_eq!(
         keys[0]["account_quota"],
-        json!("周剩余 100.0% (7天0小时后重置) | 5H剩余 100.0% (5小时0分钟后重置)")
+        json!("周剩余 100.0% | 5H剩余 100.0%")
     );
+    assert_eq!(keys[0]["status_snapshot"]["quota"]["code"], "ok");
+    assert_eq!(keys[0]["status_snapshot"]["quota"]["exhausted"], false);
 }
 
 #[tokio::test]

crates/aether-provider/pool/Cargo.toml

@@ -11,6 +11,7 @@ aether-contracts.workspace = true
 aether-data-contracts.workspace = true
 aether-pool-core.workspace = true
 aether-provider-transport.workspace = true
+chrono.workspace = true
 serde_json.workspace = true
 url.workspace = true
 uuid.workspace = true

crates/aether-provider/pool/src/lib.rs

@@ -38,11 +38,12 @@ pub use providers::{
     XAI_BILLING_PATH, XAI_USER_PATH,
 };
 pub use quota::{
-    provider_pool_key_account_quota_exhausted, provider_pool_key_minimum_quota_reached,
+    provider_pool_codex_metadata_has_account_quota, provider_pool_key_account_quota_exhausted,
-    provider_pool_key_model_quota_exhausted, provider_pool_key_model_quota_hard_blocked,
+    provider_pool_key_minimum_quota_reached, provider_pool_key_model_quota_exhausted,
-    provider_pool_key_quota_hard_blocked, provider_pool_key_scheduling_label,
+    provider_pool_key_model_quota_hard_blocked, provider_pool_key_quota_hard_blocked,
-    provider_pool_member_quota_snapshot, provider_pool_quota_metadata_provider_type,
+    provider_pool_key_scheduling_label, provider_pool_member_quota_snapshot,
-    provider_pool_quota_metadata_updated_at, provider_pool_quota_snapshot_updated_at,
+    provider_pool_quota_metadata_provider_type, provider_pool_quota_metadata_updated_at,
+    provider_pool_quota_snapshot_updated_at,
 };
 pub use quota_refresh::ProviderPoolQuotaRequestSpec;
 pub use service::ProviderPoolService;
@@ -1600,6 +1601,245 @@ mod tests {
         assert!(!signals.quota_exhausted);
     }
 
+    #[test]
+    fn codex_newer_flat_quota_metadata_clears_stale_exhausted_snapshot() {
+        let now = std::time::SystemTime::now()
+            .duration_since(std::time::UNIX_EPOCH)
+            .expect("system time should be after unix epoch")
+            .as_secs();
+        let service = ProviderPoolService::with_builtin_adapters();
+        // Headers and quota refreshes persist the flat metadata shape while the
+        // derived snapshot may still describe the previous quota observation.
+        for used_percent in [83.0, 99.0, 100.0] {
+            let mut key = sample_key(Some(json!({
+                "codex": {
+                    "updated_at": now,
+                    "primary_used_percent": used_percent,
+                    "primary_reset_at": now + 3600
+                }
+            })));
+            key.status_snapshot = Some(json!({
+                "quota": {
+                    "provider_type": "codex",
+                    "observed_at": now - 60,
+                    "updated_at": now - 60,
+                    "code": "exhausted",
+                    "exhausted": true,
+                    "allowed": false,
+                    "limit_reached": true,
+                    "reset_at": now + 3600,
+                    "windows": [{
+                        "code": "weekly",
+                        "scope": "account",
+                        "used_ratio": 1.0,
+                        "remaining_ratio": 0.0,
+                        "is_exhausted": true,
+                        "reset_at": now + 3600
+                    }]
+                }
+            }));
+            for model in [None, Some("gpt-5.4"), Some("o3")] {
+                let signals = service.member_signals("codex", &key, None, model);
+                assert_eq!(signals.quota_exhausted, used_percent >= 100.0);
+                assert!(!signals.quota_hard_blocked);
+                assert_eq!(
+                    provider_pool_key_minimum_quota_reached(&key, "codex", model),
+                    used_percent >= 99.0
+                );
+            }
+        }
+    }
+
+    #[test]
+    fn codex_account_and_model_quota_agree_on_explicit_allow_and_deny_flags() {
+        let service = ProviderPoolService::with_builtin_adapters();
+        for (flags, exhausted) in [
+            (json!({ "allowed": true }), false),
+            (json!({ "limit_reached": false }), false),
+            (json!({ "allowed": true, "limit_reached": true }), true),
+            (json!({ "allowed": false, "limit_reached": false }), true),
+        ] {
+            for use_windows in [false, true] {
+                let mut metadata = flags.clone();
+                metadata["updated_at"] = json!(300);
+                if use_windows {
+                    metadata["windows"] = json!([{ "code": "weekly", "used_ratio": 1.0 }]);
+                } else {
+                    metadata["primary_used_percent"] = json!(100.0);
+                }
+                let key = sample_key(Some(json!({ "codex": metadata })));
+                for model in [None, Some("gpt-5.4"), Some("o3")] {
+                    assert_eq!(
+                        service
+                            .member_signals("codex", &key, None, model)
+                            .quota_exhausted,
+                        exhausted,
+                        "flags={flags}, use_windows={use_windows}, model={model:?}"
+                    );
+                    // The opt-in reserve still protects a numerically full
+                    // window even when the upstream reports it as allowed.
+                    assert!(provider_pool_key_minimum_quota_reached(
+                        &key, "codex", model
+                    ));
+                }
+            }
+        }
+    }
+
+    #[test]
+    fn codex_model_quota_honors_latest_account_refusal_without_blocking_spark() {
+        let service = ProviderPoolService::with_builtin_adapters();
+        for include_window in [false, true] {
+            let mut metadata = json!({
+                "updated_at": 300,
+                "allowed": false,
+                "limit_reached": true,
+                "spark_primary_used_percent": 17.0
+            });
+            if include_window {
+                metadata["primary_used_percent"] = json!(83.0);
+            }
+            let mut key = sample_key(Some(json!({ "codex": metadata })));
+            for include_snapshot in [false, true] {
+                if include_snapshot {
+                    key.status_snapshot = Some(json!({
+                        "quota": {
+                            "provider_type": "codex",
+                            "observed_at": 200,
+                            "exhausted": false,
+                            "windows": [{ "code": "weekly", "used_ratio": 0.83 }]
+                        }
+                    }));
+                }
+                assert!(
+                    service
+                        .member_signals("codex", &key, None, Some("gpt-5.4"))
+                        .quota_exhausted
+                );
+                assert!(
+                    !service
+                        .member_signals("codex", &key, None, Some("gpt-5.3-codex-spark"))
+                        .quota_exhausted
+                );
+            }
+        }
+    }
+
+    #[test]
+    fn codex_quota_source_freshness_supports_all_timestamp_formats() {
+        let service = ProviderPoolService::with_builtin_adapters();
+        for observed_at in [
+            json!(1_700_000_200_u64),
+            json!(1_700_000_200_000_u64),
+            json!("1700000200000"),
+            json!("2023-11-14T22:16:40Z"),
+        ] {
+            for use_windows in [false, true] {
+                let metadata = if use_windows {
+                    json!({
+                        "updated_at": observed_at,
+                        "windows": [{ "code": "weekly", "used_ratio": 0.83 }]
+                    })
+                } else {
+                    json!({ "updated_at": observed_at, "primary_used_percent": 83.0 })
+                };
+                let mut key = sample_key(Some(json!({ "codex": metadata })));
+                key.status_snapshot = Some(json!({
+                    "quota": {
+                        "provider_type": "codex",
+                        "observed_at": "2023-11-14T22:15:00Z",
+                        "exhausted": true,
+                        "allowed": false,
+                        "windows": [{ "code": "weekly", "used_ratio": 1.0 }]
+                    }
+                }));
+                for model in [None, Some("gpt-5.4")] {
+                    let signals = service.member_signals("codex", &key, None, model);
+                    assert!(!signals.quota_exhausted, "timestamp={observed_at}");
+                    assert!(!signals.quota_hard_blocked, "timestamp={observed_at}");
+                    assert!(!provider_pool_key_minimum_quota_reached(
+                        &key, "codex", model
+                    ));
+                }
+
+                // Swap freshness while retaining the same observations. An
+                // older usable bucket cannot erase a later exhausted snapshot.
+                key.status_snapshot.as_mut().unwrap()["quota"]["observed_at"] =
+                    json!("2023-11-14T22:18:20Z");
+                assert!(provider_pool_key_account_quota_exhausted(&key, "codex"));
+                assert!(provider_pool_key_quota_hard_blocked(&key, "codex"));
+                assert_eq!(
+                    provider_pool_key_model_quota_exhausted(&key, "codex", "gpt-5.4"),
+                    Some(true)
+                );
+            }
+        }
+    }
+
+    #[test]
+    fn codex_account_quota_recovery_requires_new_account_observation() {
+        for metadata in [
+            json!({ "updated_at": 100, "primary_used_percent": 83.0 }),
+            json!({ "primary_used_percent": 83.0 }),
+            json!({ "updated_at": 300, "plan_type": "plus" }),
+            json!({ "updated_at": 300, "credits_unlimited": false }),
+            json!({ "updated_at": 300, "spark_primary_used_percent": 83.0 }),
+            json!({ "updated_at": 300, "windows": [{ "code": "weekly" }] }),
+        ] {
+            let mut key = sample_key(Some(json!({ "codex": metadata })));
+            key.status_snapshot = Some(json!({
+                "quota": {
+                    "provider_type": "codex",
+                    "observed_at": 200,
+                    "exhausted": true,
+                    "allowed": false,
+                    "limit_reached": true,
+                    "windows": [{ "code": "weekly", "used_ratio": 1.0 }]
+                }
+            }));
+            assert!(provider_pool_key_account_quota_exhausted(&key, "codex"));
+            assert!(provider_pool_key_quota_hard_blocked(&key, "codex"));
+            assert_eq!(
+                provider_pool_key_model_quota_exhausted(&key, "codex", "gpt-5.4"),
+                Some(true)
+            );
+        }
+
+        let mut key = sample_key(Some(json!({
+            "codex": {
+                "updated_at": 300,
+                "primary_used_percent": 83.0,
+                "allowed": false,
+                "limit_reached": true
+            }
+        })));
+        key.status_snapshot = Some(json!({
+            "quota": { "provider_type": "codex", "updated_at": 200, "exhausted": false }
+        }));
+        assert!(provider_pool_key_account_quota_exhausted(&key, "codex"));
+        assert!(provider_pool_key_quota_hard_blocked(&key, "codex"));
+    }
+
+    #[test]
+    fn codex_account_updates_do_not_override_independent_model_quota() {
+        for (spark_ratio, account_percent) in [(0.83, 100.0), (1.0, 83.0)] {
+            let mut key = sample_key(Some(json!({
+                "codex": { "updated_at": 300, "primary_used_percent": account_percent }
+            })));
+            key.status_snapshot = Some(json!({
+                "quota": {
+                    "provider_type": "codex",
+                    "updated_at": 200,
+                    "windows": [{ "code": "spark_5h", "used_ratio": spark_ratio }]
+                }
+            }));
+            assert_eq!(
+                provider_pool_key_model_quota_exhausted(&key, "codex", "gpt-5.3-codex-spark"),
+                Some(spark_ratio >= 1.0)
+            );
+        }
+    }
+
     #[test]
     fn codex_explicit_quota_block_is_hard_until_reset() {
         let now = std::time::SystemTime::now()

crates/aether-provider/pool/src/providers/codex.rs

@@ -10,10 +10,11 @@ use crate::provider::{
     ProviderPoolMemberInput,
 };
 use crate::quota::{
-    provider_pool_current_unix_secs, provider_pool_json_bool, provider_pool_json_f64,
+    provider_pool_codex_metadata_has_account_quota, provider_pool_current_unix_secs,
-    provider_pool_member_quota_snapshot, provider_pool_metadata_bucket,
+    provider_pool_json_bool, provider_pool_json_f64, provider_pool_member_quota_snapshot,
-    provider_pool_model_quota_exhausted, provider_pool_quota_snapshot_exhausted_decision,
+    provider_pool_metadata_bucket, provider_pool_model_quota_exhausted,
-    provider_pool_reset_deadline_elapsed, provider_pool_timestamp_unix_secs,
+    provider_pool_quota_snapshot_exhausted_decision, provider_pool_reset_deadline_elapsed,
+    provider_pool_source_account_quota_exhausted, provider_pool_timestamp_unix_secs,
 };
 use crate::quota_refresh::ProviderPoolQuotaRequestSpec;
 
@@ -54,6 +55,9 @@ impl ProviderPoolAdapter for CodexProviderPoolAdapter {
         }) {
             return exhausted;
         }
+        if let Some(bucket) = codex_newer_account_quota_metadata(input.key, input.provider_type) {
+            return quota_exhausted_from_bucket(bucket);
+        }
         if let Some(quota_snapshot) =
             provider_pool_member_quota_snapshot(input.key, input.provider_type)
         {
@@ -108,15 +112,12 @@ fn codex_explicit_quota_block_active(
     key: &aether_data_contracts::repository::provider_catalog::StoredProviderCatalogKey,
     provider_type: &str,
 ) -> bool {
+    if let Some(bucket) = codex_newer_account_quota_metadata(key, provider_type) {
+        return codex_explicit_quota_block_from_bucket(bucket);
+    }
     let Some(quota_snapshot) = provider_pool_member_quota_snapshot(key, provider_type) else {
         return provider_pool_metadata_bucket(key.upstream_metadata.as_ref(), provider_type)
-            .is_some_and(|bucket| {
+            .is_some_and(codex_explicit_quota_block_from_bucket);
-                (provider_pool_json_bool(bucket.get("allowed")) == Some(false)
-                    || provider_pool_json_bool(bucket.get("limit_reached")) == Some(true))
-                    && !["primary", "secondary"]
-                        .into_iter()
-                        .any(|prefix| codex_window_reset_elapsed(bucket, prefix))
-            });
     };
     let explicitly_blocked = provider_pool_json_bool(quota_snapshot.get("allowed")) == Some(false)
         || provider_pool_json_bool(quota_snapshot.get("limit_reached")) == Some(true);
@@ -133,6 +134,35 @@ fn codex_explicit_quota_block_active(
     })
 }
 
+fn codex_explicit_quota_block_from_bucket(bucket: &Map<String, Value>) -> bool {
+    (provider_pool_json_bool(bucket.get("allowed")) == Some(false)
+        || provider_pool_json_bool(bucket.get("limit_reached")) == Some(true))
+        && !["primary", "secondary"]
+            .into_iter()
+            .any(|prefix| codex_window_reset_elapsed(bucket, prefix))
+}
+
+/// A successful refresh can update raw metadata before the status snapshot.
+/// Do not keep an old account-level block once a newer quota observation exists.
+/// Identity-only or model-only updates cannot clear an account quota decision.
+fn codex_newer_account_quota_metadata<'a>(
+    key: &'a aether_data_contracts::repository::provider_catalog::StoredProviderCatalogKey,
+    provider_type: &str,
+) -> Option<&'a Map<String, Value>> {
+    let snapshot = provider_pool_member_quota_snapshot(key, provider_type)?;
+    let metadata = provider_pool_metadata_bucket(key.upstream_metadata.as_ref(), provider_type)?;
+    if !provider_pool_codex_metadata_has_account_quota(metadata) {
+        return None;
+    }
+    let metadata_observed_at = provider_pool_timestamp_unix_secs(metadata.get("observed_at"))
+        .or_else(|| provider_pool_timestamp_unix_secs(metadata.get("updated_at")))?;
+    let snapshot_observed_at = provider_pool_timestamp_unix_secs(snapshot.get("observed_at"))
+        .or_else(|| provider_pool_timestamp_unix_secs(snapshot.get("updated_at")));
+    snapshot_observed_at
+        .is_none_or(|observed_at| metadata_observed_at >= observed_at)
+        .then_some(metadata)
+}
+
 fn build_codex_wham_headers(
     resolved_oauth_auth: Option<(String, String)>,
     decrypted_api_key: Option<&str>,
@@ -322,11 +352,14 @@ pub(crate) fn quota_exhausted_from_bucket(bucket: &Map<String, Value>) -> bool {
     if provider_pool_json_bool(bucket.get("credits_unlimited")) == Some(true) {
         return false;
     }
+    let account_windows_exhausted = provider_pool_source_account_quota_exhausted(bucket);
     let has_window_data = provider_pool_json_f64(bucket.get("primary_used_percent")).is_some()
-        || provider_pool_json_f64(bucket.get("secondary_used_percent")).is_some();
+        || provider_pool_json_f64(bucket.get("secondary_used_percent")).is_some()
+        || account_windows_exhausted.is_some();
     if !has_window_data && provider_pool_json_bool(bucket.get("has_credits")) == Some(false) {
         return true;
     }
-    codex_window_used_percent_exhausted(bucket, "primary")
+    account_windows_exhausted == Some(true)
+        || codex_window_used_percent_exhausted(bucket, "primary")
         || codex_window_used_percent_exhausted(bucket, "secondary")
 }

crates/aether-provider/pool/src/quota.rs

@@ -87,6 +87,7 @@ fn provider_pool_quota_reaches_reserve(
     provider_model_name: Option<&str>,
     reserve_ratio: f64,
 ) -> Option<bool> {
+    let is_codex = provider_type.trim().eq_ignore_ascii_case("codex");
     let requested = provider_model_name.map(provider_pool_identifier_tokens);
     if reserve_ratio <= 0.0 && requested.as_ref().is_some_and(|tokens| tokens.is_empty()) {
         return None;
@@ -103,12 +104,19 @@ fn provider_pool_quota_reaches_reserve(
     let mut resolved = None::<(Option<u64>, bool)>;
     let mut resolved_specific_bucket = false;
     for source in sources.into_iter().flatten() {
+        let observed_at = provider_pool_timestamp_unix_secs(source.get("observed_at"))
+            .or_else(|| provider_pool_timestamp_unix_secs(source.get("updated_at")));
+        let account_signal = (is_codex && reserve_ratio <= 0.0)
+            .then(|| provider_pool_codex_account_quota_signal(source, observed_at))
+            .flatten();
         let mut windows = provider_pool_collect_quota_windows(source);
-        if reserve_ratio > 0.0 && provider_type.trim().eq_ignore_ascii_case("codex") {
+        if is_codex {
             // Raw refresh metadata can be newer than the materialized windows.
             // Include all four legacy slots before selecting the model bucket.
             for prefix in ["primary", "secondary", "spark_primary", "spark_secondary"] {
-                let Some(used_percent) = source.get(&format!("{prefix}_used_percent")) else {
+                let Some(used_percent) =
+                    provider_pool_json_f64(source.get(&format!("{prefix}_used_percent")))
+                else {
                     continue;
                 };
                 if provider_pool_json_f64(source.get(&format!("{prefix}_window_minutes")))
@@ -118,7 +126,7 @@ fn provider_pool_quota_reaches_reserve(
                 }
                 let mut window = Map::from_iter([
                     ("code".to_string(), json!(prefix)),
-                    ("used_percent".to_string(), used_percent.clone()),
+                    ("used_percent".to_string(), json!(used_percent)),
                 ]);
                 for field in [
                     "reset_at",
@@ -132,12 +140,21 @@ fn provider_pool_quota_reaches_reserve(
                 }
                 windows.push(window);
             }
+            windows.retain(provider_pool_quota_window_has_observation);
+            if let Some(exhausted) = account_signal {
+                if !windows.iter().any(provider_pool_window_is_generic) {
+                    // A flags-only quota refresh is still a newer account
+                    // observation, but must not override a model-specific bucket.
+                    windows.push(Map::from_iter([
+                        ("code".to_string(), json!("account")),
+                        ("is_exhausted".to_string(), json!(exhausted)),
+                    ]));
+                }
+            }
         }
         if windows.is_empty() {
             continue;
         }
-        let observed_at = provider_pool_timestamp_unix_secs(source.get("observed_at"))
-            .or_else(|| provider_pool_timestamp_unix_secs(source.get("updated_at")));
         let model_matches = windows
             .iter()
             .filter(|window| {
@@ -155,7 +172,7 @@ fn provider_pool_quota_reaches_reserve(
                 provider_pool_explicit_model_windows_exhausted(model_matches, observed_at)
             };
             if resolved.is_none()
-                || (reserve_ratio > 0.0 && !resolved_specific_bucket)
+                || (is_codex && !resolved_specific_bucket)
                 || provider_pool_should_replace_model_quota_resolution(
                     resolved.as_ref().and_then(|(observed_at, _)| *observed_at),
                     observed_at,
@@ -183,7 +200,7 @@ fn provider_pool_quota_reaches_reserve(
             let exhausted =
                 provider_pool_any_window_exhausted(family_matches, observed_at, reserve_ratio);
             if resolved.is_none()
-                || (reserve_ratio > 0.0 && !resolved_specific_bucket)
+                || (is_codex && !resolved_specific_bucket)
                 || provider_pool_should_replace_model_quota_resolution(
                     resolved.as_ref().and_then(|(observed_at, _)| *observed_at),
                     observed_at,
@@ -197,7 +214,7 @@ fn provider_pool_quota_reaches_reserve(
 
         // A newer account observation does not update an independent model
         // bucket. Only compare freshness between applicable model sources.
-        if reserve_ratio > 0.0 && resolved_specific_bucket {
+        if is_codex && resolved_specific_bucket {
             continue;
         }
 
@@ -211,8 +228,9 @@ fn provider_pool_quota_reaches_reserve(
             .filter(|window| provider_pool_window_is_generic(window))
             .collect::<Vec<_>>();
         if !generic_matches.is_empty() {
-            let exhausted =
+            let exhausted = account_signal.unwrap_or_else(|| {
-                provider_pool_any_window_exhausted(generic_matches, observed_at, reserve_ratio);
+                provider_pool_any_window_exhausted(generic_matches, observed_at, reserve_ratio)
+            });
             if resolved.is_none()
                 || provider_pool_should_replace_model_quota_resolution(
                     resolved.as_ref().and_then(|(observed_at, _)| *observed_at),
@@ -227,6 +245,37 @@ fn provider_pool_quota_reaches_reserve(
     resolved.map(|(_, exhausted)| exhausted)
 }
 
+fn provider_pool_codex_account_quota_signal(
+    source: &Map<String, Value>,
+    observed_at: Option<u64>,
+) -> Option<bool> {
+    let allowed = provider_pool_json_bool(source.get("allowed"));
+    let limit_reached = provider_pool_json_bool(source.get("limit_reached"));
+    if allowed == Some(false) || limit_reached == Some(true) {
+        let reset_elapsed = provider_pool_current_unix_secs().is_some_and(|now| {
+            provider_pool_reset_deadline_elapsed(source, observed_at, now)
+                || ["primary", "secondary"].into_iter().any(|prefix| {
+                    let window = [
+                        "reset_at",
+                        "next_reset_at",
+                        "reset_seconds",
+                        "reset_after_seconds",
+                    ]
+                    .into_iter()
+                    .filter_map(|field| {
+                        source
+                            .get(&format!("{prefix}_{field}"))
+                            .map(|value| (field.to_string(), value.clone()))
+                    })
+                    .collect::<Map<_, _>>();
+                    provider_pool_reset_deadline_elapsed(&window, observed_at, now)
+                })
+        });
+        return (!reset_elapsed).then_some(true);
+    }
+    (allowed == Some(true) || limit_reached == Some(false)).then_some(false)
+}
+
 fn provider_pool_should_replace_model_quota_resolution(
     previous_observed_at: Option<u64>,
     next_observed_at: Option<u64>,
@@ -282,6 +331,69 @@ fn provider_pool_any_window_exhausted(
     })
 }
 
+fn provider_pool_quota_window_has_observation(window: &Map<String, Value>) -> bool {
+    ["is_exhausted", "exhausted"]
+        .into_iter()
+        .any(|field| provider_pool_json_bool(window.get(field)).is_some())
+        || [
+            "used_ratio",
+            "usage_ratio",
+            "used_percent",
+            "remaining_ratio",
+            "remaining_fraction",
+            "remaining_percent",
+        ]
+        .into_iter()
+        .any(|field| provider_pool_json_f64(window.get(field)).is_some())
+        || (provider_pool_json_f64(
+            window
+                .get("remaining")
+                .or_else(|| window.get("remaining_value")),
+        )
+        .is_some()
+            && provider_pool_json_f64(
+                window
+                    .get("limit")
+                    .or_else(|| window.get("limit_value"))
+                    .or_else(|| window.get("total")),
+            )
+            .is_some_and(|limit| limit > 0.0))
+}
+
+pub(crate) fn provider_pool_source_account_quota_exhausted(
+    source: &Map<String, Value>,
+) -> Option<bool> {
+    let windows = provider_pool_collect_quota_windows(source);
+    let account_windows = windows
+        .iter()
+        .filter(|window| provider_pool_window_is_generic(window))
+        .filter(|window| provider_pool_quota_window_has_observation(window))
+        .collect::<Vec<_>>();
+    if account_windows.is_empty() {
+        return None;
+    }
+    let observed_at = provider_pool_timestamp_unix_secs(source.get("observed_at"))
+        .or_else(|| provider_pool_timestamp_unix_secs(source.get("updated_at")));
+    Some(provider_pool_any_window_exhausted(
+        account_windows,
+        observed_at,
+        0.0,
+    ))
+}
+
+/// Whether Codex metadata contains an account quota observation rather than an
+/// identity-only update or an independent model quota bucket.
+pub fn provider_pool_codex_metadata_has_account_quota(source: &Map<String, Value>) -> bool {
+    ["primary_used_percent", "secondary_used_percent"]
+        .into_iter()
+        .any(|field| provider_pool_json_f64(source.get(field)).is_some())
+        || provider_pool_source_account_quota_exhausted(source).is_some()
+        || ["allowed", "limit_reached", "has_credits"]
+            .into_iter()
+            .any(|field| provider_pool_json_bool(source.get(field)).is_some())
+        || provider_pool_json_bool(source.get("credits_unlimited")) == Some(true)
+}
+
 /// Whether an applicable Codex window has at most 1% remaining. Missing quota
 /// data and windows whose reset has elapsed do not trigger this opt-in guard.
 pub fn provider_pool_key_minimum_quota_reached(
@@ -700,7 +812,16 @@ pub(crate) fn provider_pool_json_f64(value: Option<&Value>) -> Option<f64> {
 }
 
 pub(crate) fn provider_pool_timestamp_unix_secs(value: Option<&Value>) -> Option<u64> {
-    let mut timestamp = provider_pool_json_f64(value)?;
+    let mut timestamp = match provider_pool_json_f64(value) {
+        Some(timestamp) => timestamp,
+        None => {
+            return chrono::DateTime::parse_from_rfc3339(value?.as_str()?.trim())
+                .ok()?
+                .timestamp()
+                .try_into()
+                .ok();
+        }
+    };
     if timestamp <= 0.0 {
         return None;
     }

crates/aether-usage/runtime/src/runtime.rs

@@ -10055,6 +10055,24 @@ mod tests {
             2,
             "the later direct caller should receive its own bounded write attempt"
         );
+        // Direct persistence can finish before the submission worker joins the
+        // barrier handoff and accounts for its completed slot.
+        timeout(Duration::from_secs(1), async {
+            loop {
+                let snapshot = runtime.metrics_snapshot();
+                let submission = &runtime.lifecycle_submission.state;
+                if snapshot.terminal_submission_pending == 0
+                    && snapshot.ordered_lifecycle_pending == 0
+                    && snapshot.lifecycle_submission_pending == 0
+                    && submission.admission.available_permits() == submission.capacity
+                {
+                    break;
+                }
+                sleep(Duration::from_millis(1)).await;
+            }
+        })
+        .await
+        .expect("failed terminal submission accounting and admission should drain");
         let snapshot = runtime.metrics_snapshot();
         assert_eq!(snapshot.terminal_submission_pending, 0);
         assert_eq!(snapshot.ordered_lifecycle_pending, 0);
@@ -10162,6 +10180,7 @@ mod tests {
             .await;
 
         assert_eq!(remaining_policy_panics.load(Ordering::Acquire), 0);
+        {
             let records = records.lock().expect("records lock");
             assert_eq!(
                 records
@@ -10179,7 +10198,25 @@ mod tests {
                     == 1,
                 "only the later healthy attempt should persist for the panicked request"
             );
-        drop(records);
+        }
+        // The final direct attempt also submits a barrier whose worker may
+        // account for completion after the persistence call has returned.
+        timeout(Duration::from_secs(1), async {
+            loop {
+                let snapshot = runtime.metrics_snapshot();
+                let submission = &runtime.lifecycle_submission.state;
+                if snapshot.terminal_submission_pending == 0
+                    && snapshot.ordered_lifecycle_pending == 0
+                    && snapshot.lifecycle_submission_pending == 0
+                    && submission.admission.available_permits() == submission.capacity
+                {
+                    break;
+                }
+                sleep(Duration::from_millis(1)).await;
+            }
+        })
+        .await
+        .expect("panicked terminal submission accounting and admission should drain");
         let snapshot = runtime.metrics_snapshot();
         assert_eq!(snapshot.terminal_submission_pending, 0);
         assert_eq!(snapshot.ordered_lifecycle_pending, 0);

frontend/src/api/__tests__/providers-pool-cache.spec.ts

@@ -0,0 +1,76 @@
+import { beforeEach, describe, expect, it, vi } from 'vitest'
+
+const { getMock, patchMock } = vi.hoisted(() => ({ getMock: vi.fn(), patchMock: vi.fn() }))
+
+vi.mock('@/api/client', () => ({ default: { get: getMock, patch: patchMock } }))
+
+import { getPoolOverview, listPoolKeys, listPoolScores } from '@/api/endpoints/pool'
+import { getProvider, updateProvider } from '@/api/endpoints/providers'
+import { cache } from '@/utils/cache'
+
+const options = { cacheTtlMs: 30_000 }
+const provider = { id: 'codex', pool_advanced: { reserve_minimum_quota: false } }
+
+function deferred<T>() {
+  let resolve!: (value: T) => void
+  const promise = new Promise<T>((resolvePromise) => { resolve = resolvePromise })
+  return { promise, resolve }
+}
+
+beforeEach(() => {
+  cache.clear()
+  getMock.mockReset()
+  patchMock.mockReset()
+  patchMock.mockResolvedValue({ data: provider })
+})
+
+describe('provider pool settings cache invalidation', () => {
+  it('refreshes every cached key page, score page and overview after changing pool settings', async () => {
+    getMock.mockResolvedValue({ data: { version: 'old' } })
+    await getPoolOverview(options)
+    await listPoolKeys('codex', {}, options)
+    await listPoolKeys('codex', { page: 2, status: 'quota_exhausted' }, options)
+    await listPoolScores('codex', {}, options)
+    await listPoolKeys('codex-other', {}, options)
+    await updateProvider('codex', { pool_advanced: { reserve_minimum_quota: false } })
+    getMock.mockResolvedValue({ data: { version: 'new' } })
+
+    await expect(getPoolOverview(options)).resolves.toEqual({ version: 'new' })
+    await expect(listPoolKeys('codex', {}, options)).resolves.toEqual({ version: 'new' })
+    await expect(listPoolKeys('codex', { page: 2, status: 'quota_exhausted' }, options))
+      .resolves.toEqual({ version: 'new' })
+    await expect(listPoolScores('codex', {}, options)).resolves.toEqual({ version: 'new' })
+    await expect(listPoolKeys('codex-other', {}, options)).resolves.toEqual({ version: 'old' })
+  })
+
+  it('does not reuse or cache a key request started before the settings were saved', async () => {
+    const oldResponse = deferred<{ data: { version: string } }>()
+    const newResponse = deferred<{ data: { version: string } }>()
+    getMock.mockReturnValueOnce(oldResponse.promise).mockReturnValueOnce(newResponse.promise)
+    const oldRequest = listPoolKeys('codex', { page: 2 }, options)
+    await updateProvider('codex', { pool_advanced: { reserve_minimum_quota: false } })
+    const newRequest = listPoolKeys('codex', { page: 2 }, options)
+    expect(getMock).toHaveBeenCalledTimes(2)
+
+    oldResponse.resolve({ data: { version: 'old' } })
+    await oldRequest
+    const deduped = listPoolKeys('codex', { page: 2 }, options)
+    expect(getMock).toHaveBeenCalledTimes(2)
+    newResponse.resolve({ data: { version: 'new' } })
+    await expect(newRequest).resolves.toEqual({ version: 'new' })
+    await expect(deduped).resolves.toEqual({ version: 'new' })
+    await expect(listPoolKeys('codex', { page: 2 }, options)).resolves.toEqual({ version: 'new' })
+  })
+
+  it('does not reuse a provider detail request started before a successful save', async () => {
+    const oldResponse = deferred<{ data: typeof provider }>()
+    getMock.mockReturnValueOnce(oldResponse.promise).mockResolvedValueOnce({ data: provider })
+    const oldRequest = getProvider('codex')
+    await updateProvider('codex', { pool_advanced: { reserve_minimum_quota: false } })
+    const newRequest = getProvider('codex')
+    expect(getMock).toHaveBeenCalledTimes(2)
+    oldResponse.resolve({ data: { ...provider, pool_advanced: { reserve_minimum_quota: true } } })
+    await oldRequest
+    await expect(newRequest).resolves.toMatchObject(provider)
+  })
+})

frontend/src/api/endpoints/providers.ts

@@ -1,5 +1,5 @@
 import client from '../client'
-import { buildCacheKey, cachedRequest, dedupedRequest } from '@/utils/cache'
+import { buildCacheKey, cache, cachedRequest, dedupedRequest } from '@/utils/cache'
 import type {
   ClaudeCodeAdvancedConfig,
   FailoverRulesConfig,
@@ -142,6 +142,15 @@ export async function updateProvider(
   requestOptions?: ProviderRequestOptions,
 ): Promise<ProviderWithEndpointsSummary> {
   const response = await client.patch<ProviderWithEndpointsSummary>(`/api/admin/providers/${providerId}`, data, requestOptions)
+  cache.delete(`providers:detail:${providerId}`)
+  if ('pool_advanced' in data) {
+    cache.delete('pool:overview')
+    for (const kind of ['keys', 'scores']) {
+      const prefix = `pool:${kind}:${providerId}`
+      cache.delete(prefix)
+      cache.deleteByPrefix(`${prefix}:`)
+    }
+  }
   return normalizeProviderSummary(response.data)
 }
 

frontend/src/utils/cache.ts

@@ -61,6 +61,14 @@ class MemoryCache {
     this.keyVersions.set(key, (this.keyVersions.get(key) ?? 0) + 1)
   }
 
+  /** 删除匹配前缀的缓存及进行中请求，旧响应不再回填缓存。 */
+  deleteByPrefix(prefix: string): void {
+    const keys = new Set([...this.cache.keys(), ...this.inFlight.keys()])
+    for (const key of keys) {
+      if (key.startsWith(prefix)) this.delete(key)
+    }
+  }
+
   /**
    * 清空所有缓存
    */

frontend/src/views/admin/PoolManagement.vue

@@ -1395,6 +1395,8 @@ async function loadOverview(options: { cacheTtlMs?: number, silent?: boolean } =
 
 async function handleSchedulingSaved(updatedProvider: ProviderWithEndpointsSummary) {
   if (!selectedProviderId.value || updatedProvider.id !== selectedProviderId.value) return
+  // 保存前发出的详情读取不得覆盖这次保存返回的新配置。
+  providerDataRequestId += 1
   // 优先回写保存接口返回值，避免弹窗立即重开时读到旧配置。
   if (selectedProviderData.value) {
     Object.assign(selectedProviderData.value, updatedProvider)
@@ -1403,7 +1405,7 @@ async function handleSchedulingSaved(updatedProvider: ProviderWithEndpointsSumma
   }
   showSchedulingDialog.value = false
   showAdvancedDialog.value = false
-  await loadOverview({ silent: true })
+  await Promise.all([loadKeys({ silent: true }), loadOverview({ silent: true })])
 }
 
 // --- Provider Selection ---

frontend/src/views/admin/__tests__/PoolManagement.codex-cycle-stats.spec.ts

@@ -340,12 +340,24 @@ vi.mock('@/components/ui/refresh-button.vue', async () => {
 })
 
 vi.mock('@/features/pool/components/PoolSchedulingDialog.vue', async () => {
-  const { defineComponent } = await import('vue')
+  const { defineComponent, h } = await import('vue')
   return {
     default: defineComponent({
       name: 'PoolSchedulingDialogStub',
-      setup() {
+      props: {
-        return () => null
+        providerId: { type: String, required: true },
+        currentConfig: { type: Object, default: null },
+      },
+      emits: ['saved'],
+      setup(props, { emit }) {
+        return () => h('button', {
+          'data-testid': 'save-pool-scheduling',
+          'data-reserve': String(props.currentConfig?.reserve_minimum_quota ?? false),
+          onClick: () => emit('saved', {
+            id: props.providerId,
+            pool_advanced: { ...props.currentConfig, reserve_minimum_quota: false },
+          }),
+        }, '保存调度测试')
       },
     }),
   }
@@ -1278,4 +1290,80 @@ describe('PoolManagement Codex cycle stats mode', () => {
 
     expect(disabledRoot.querySelector('[data-testid="pool-demand-metrics-button"]')).toBeNull()
   })
+
+  it('reloads account scheduling state immediately after disabling minimum quota reserve', async () => {
+    const key = createPoolKey('codex', {
+      scheduling_status: 'blocked',
+      scheduling_reason: 'account_quota_exhausted',
+      scheduling_label: '额度耗尽',
+    })
+    endpointMocks.getPoolOverview.mockResolvedValue({ items: [createOverview('codex')] })
+    endpointMocks.listPoolKeys.mockResolvedValue(createKeyPage(key))
+    endpointMocks.getProvider.mockResolvedValue(createProvider('codex', {
+      pool_advanced: { reserve_minimum_quota: true },
+    }))
+    const root = mountPoolManagement()
+    await settle()
+    expect(root.querySelector('tbody')?.textContent).toContain('额度耗尽')
+    endpointMocks.listPoolKeys.mockClear()
+    endpointMocks.listPoolKeys.mockResolvedValue(createKeyPage({
+      ...key,
+      scheduling_status: 'available',
+      scheduling_reason: 'available',
+      scheduling_label: '可用',
+    }))
+
+    root.querySelector<HTMLButtonElement>('[data-testid="save-pool-scheduling"]')?.click()
+    await settle()
+
+    expect(endpointMocks.listPoolKeys).toHaveBeenCalledWith(
+      'codex-provider', expect.anything(), { cacheTtlMs: 0 },
+    )
+    expect(root.querySelector('tbody')?.textContent).not.toContain('额度耗尽')
+    expect(root.querySelector('tbody')?.textContent).toContain('可用')
+  })
+
+  it('shows 17% remaining without marking a schedulable account exhausted when reserve is enabled', async () => {
+    endpointMocks.getPoolOverview.mockResolvedValue({ items: [createOverview('codex')] })
+    endpointMocks.getProvider.mockResolvedValue(createProvider('codex', {
+      pool_advanced: { reserve_minimum_quota: true },
+    }))
+    endpointMocks.listPoolKeys.mockResolvedValue(createKeyPage(createPoolKey('codex', {
+      scheduling_status: 'available',
+      scheduling_reason: 'available',
+      scheduling_label: '可用',
+      status_snapshot: {
+        account: { code: 'ok', blocked: false },
+        quota: {
+          code: 'ok',
+          exhausted: false,
+          provider_type: 'codex',
+          windows: [{ code: '5h', used_ratio: 0.83, remaining_ratio: 0.17 }],
+        },
+        oauth: { code: 'none' },
+      },
+    })))
+    const root = mountPoolManagement()
+    await settle()
+
+    expect(root.querySelector('[data-testid="pool-quota-meter-text"]')?.textContent).toBe('17.0%')
+    expect(root.querySelector('tbody')?.textContent).toContain('可用')
+    expect(root.querySelector('tbody')?.textContent).not.toContain('额度耗尽')
+  })
+
+  it('does not replace saved scheduling settings with an older provider detail response', async () => {
+    let resolveOldProvider!: (provider: ReturnType<typeof createProvider>) => void
+    endpointMocks.getProvider.mockReturnValue(new Promise((resolve) => { resolveOldProvider = resolve }))
+    endpointMocks.getPoolOverview.mockResolvedValue({ items: [createOverview('codex')] })
+    endpointMocks.listPoolKeys.mockResolvedValue(createKeyPage(createPoolKey()))
+    const root = mountPoolManagement()
+    await settle()
+
+    root.querySelector<HTMLButtonElement>('[data-testid="save-pool-scheduling"]')?.click()
+    await settle()
+    resolveOldProvider(createProvider('codex', { pool_advanced: { reserve_minimum_quota: true } }))
+    await settle()
+
+    expect(root.querySelector('[data-testid="save-pool-scheduling"]')?.getAttribute('data-reserve')).toBe('false')
+  })
 })