diff --git a/backend/crates/gateway-protocol/src/openai/events.rs b/backend/crates/gateway-protocol/src/openai/events.rs index 985ab1f7e..e067a51e2 100644 --- a/backend/crates/gateway-protocol/src/openai/events.rs +++ b/backend/crates/gateway-protocol/src/openai/events.rs @@ -328,7 +328,7 @@ pub fn parse_rate_limit_headers(headers: &[(String, String)]) -> Option Option>(); + // 生图等响应只通过 active-limit 标明通用 x-codex 窗口的归属 + // 有独立具名头时保留各桶原始事实，避免把真正的 Codex 窗口一起迁移 + if let Some(active_limit) = active_limit.as_ref() + && active_limit != "codex" + && !limits.contains_key(active_limit) + && let Some(mut details) = limits.remove("codex") + { + details.limit_id.clone_from(active_limit); + limits.insert(active_limit.clone(), details); + } let credits = parse_credits_from_lookup(&normalized); let plan_type = lookup_non_empty(&normalized, "x-codex-plan-type"); let promo_message = lookup_non_empty(&normalized, "x-codex-promo-message"); diff --git a/backend/crates/gateway-protocol/tests/openai/events.rs b/backend/crates/gateway-protocol/tests/openai/events.rs index ec40e6493..46120233b 100644 --- a/backend/crates/gateway-protocol/tests/openai/events.rs +++ b/backend/crates/gateway-protocol/tests/openai/events.rs @@ -540,6 +540,7 @@ fn parse_rate_limit_headers_should_extract_primary_and_secondary_windows() { #[test] fn parse_rate_limit_headers_should_preserve_dynamic_limit_and_account_metadata() { let headers = vec![ + ("x-codex-primary-used-percent".to_owned(), "95".to_owned()), ( "x-codex-other-primary-used-percent".to_owned(), "63.5".to_owned(), @@ -556,6 +557,13 @@ fn parse_rate_limit_headers_should_preserve_dynamic_limit_and_account_metadata() ]; let parsed = parse_rate_limit_headers(&headers).expect("rate-limit headers"); + assert_eq!( + parsed.limits["codex"] + .primary + .expect("core window") + .used_percent, + 95.0 + ); let details = parsed.limits.get("codex_other").expect("dynamic limit"); let credits = parsed.credits.expect("credits"); @@ -579,6 +587,38 @@ fn parse_rate_limit_headers_should_preserve_dynamic_limit_and_account_metadata() ); } +#[test] +fn parse_rate_limit_headers_should_assign_generic_windows_to_the_active_limit() { + let headers = [ + ("x-codex-active-limit", "imagegen_premium"), + ("x-codex-primary-used-percent", "0"), + ("x-codex-primary-window-minutes", "1440"), + ("x-codex-primary-reset-at", "1900000000"), + ("x-codex-secondary-used-percent", "0"), + ("x-codex-secondary-window-minutes", "0"), + ("x-codex-secondary-reset-at", ""), + ] + .map(|(name, value)| (name.to_owned(), value.to_owned())); + + let parsed = parse_rate_limit_headers(&headers).expect("image rate-limit headers"); + let image = parsed.limits.get("imagegen_premium").expect("image limit"); + assert_eq!(image.limit_id, "imagegen_premium"); + assert_eq!(image.key_source, RateLimitKeySource::LimitId); + assert_eq!( + image.primary, + Some(RateLimitWindow { + used_percent: 0.0, + window_minutes: Some(1440), + reset_at: Some(1_900_000_000), + }) + ); + assert!(!parsed.limits.contains_key("codex")); + assert_eq!( + parse_rate_limit_headers(&rate_limits_to_header_pairs(&parsed)), + Some(parsed) + ); +} + #[test] fn parse_rate_limit_headers_should_preserve_a_code_review_window() { let headers = vec![ diff --git a/backend/crates/providers/openai/tests/provider/contract/mod.rs b/backend/crates/providers/openai/tests/provider/contract/mod.rs index 3701ab5db..de94d5593 100644 --- a/backend/crates/providers/openai/tests/provider/contract/mod.rs +++ b/backend/crates/providers/openai/tests/provider/contract/mod.rs @@ -2888,6 +2888,102 @@ async fn image_endpoints_bypass_only_the_text_catalog_and_preserve_the_current_c server.verify().await; } +#[tokio::test] +async fn image_quota_headers_keep_the_codex_weekly_window_and_access_state() { + let store = Arc::new(MemoryAccountStore::default()); + let account_id = "acct_provider_contract"; + create_account(&store, account_id).await; + let account = store.account(account_id).expect("account"); + let server = MockServer::start().await; + let (provider, quota) = provider_and_quota_with_affinity_and_base_url_and_leases( + &store, + Arc::new(MemorySessionAffinity::default()), + server.uri(), + Arc::new(TestLeaseCoordinator::default()), + 0, + ); + let core_headers = [ + ("x-codex-active-limit", "codex"), + ("x-codex-primary-used-percent", "95"), + ("x-codex-primary-window-minutes", "10080"), + ("x-codex-primary-reset-at", "1900604800"), + ("x-codex-allowed", "true"), + ] + .map(|(name, value)| (name.to_owned(), value.to_owned())); + quota + .synchronize_passive_headers(&account, &core_headers) + .await + .expect("initial weekly quota"); + let initial = quota.read_account(account.id()).await.unwrap().unwrap(); + let response = br#"{"data":[{"b64_json":"AAEC"}]}"#; + // 生图响应仅用 active-limit 标明归属，窗口仍使用通用 x-codex 头 + Mock::given(method("POST")) + .and(path("/codex/images/generations")) + .respond_with( + ResponseTemplate::new(200) + .insert_header("x-codex-active-limit", "imagegen_premium") + .insert_header("x-codex-primary-used-percent", "0") + .insert_header("x-codex-primary-window-minutes", "1440") + .insert_header("x-codex-primary-reset-at", "1900000000") + .insert_header("x-codex-secondary-used-percent", "0") + .insert_header("x-codex-secondary-window-minutes", "0") + .insert_header("x-codex-secondary-reset-at", "") + .set_body_raw(response.as_slice(), "application/json"), + ) + .expect(1) + .mount(&server) + .await; + let payload = RawJsonPayload::new( + "openai", + Bytes::from_static(br#"{"model":"gpt-image-2.6-flare","prompt":"a tree"}"#), + ) + .unwrap(); + let operation = Operation::GenerateImage(ImageRequest::from_raw_json( + ImageRequestKind::Generation, + payload, + )); + let mut stream = provider + .execute( + planned_provider_endpoint_request("openai", operation), + context("req_image_quota", CancellationToken::new()), + ) + .await + .expect("image stream"); + let mut delivered = None; + while let Some(event) = stream.next().await { + let event = event.expect("image event"); + if let Some(body) = event.wire_event().and_then(|wire| wire.raw_json_body()) { + delivered = Some(body.clone()); + } + } + assert_eq!(delivered.as_deref(), Some(response.as_slice())); + + let snapshot = quota.read_account(account.id()).await.unwrap().unwrap(); + let core = snapshot + .windows() + .iter() + .find(|window| window.is_account_wide()) + .unwrap(); + let image = snapshot + .windows() + .iter() + .find(|window| window.source() == "imagegen_premium") + .unwrap(); + assert_eq!(core, &initial.windows()[0]); + assert_eq!(image.window_seconds(), Some(86_400)); + assert_eq!(image.used_percent(), Some(0.0)); + assert!(!image.is_account_wide()); + assert_eq!(store.account(account_id).unwrap().quota(), initial.quota()); + + quota + .synchronize_passive_headers(&account, &core_headers) + .await + .expect("next Codex quota observation"); + let refreshed = quota.read_account(account.id()).await.unwrap().unwrap(); + assert_eq!(refreshed.windows(), snapshot.windows()); + server.verify().await; +} + #[tokio::test] async fn image_endpoints_should_report_usage_before_delivering_the_unchanged_body() { let response = br#"{ "created":1778832973,"data":[{"b64_json":"AAEC"}],"quality":"medium","size":"1024x1536","usage":{"input_tokens":1474,"input_tokens_details":{"image_tokens":1457,"text_tokens":17,"cached_tokens":100},"output_tokens":1372,"output_tokens_details":{"image_tokens":1372,"text_tokens":0},"total_tokens":2846},"future":9007199254740993 }"#; diff --git a/docs/architecture.md b/docs/architecture.md index 717795197..67ec9a5b8 100644 --- a/docs/architecture.md +++ b/docs/architecture.md @@ -1029,6 +1029,10 @@ credential 与 quota 是两组独立事实：credential refresh 不等于 quota OpenAI Provider 将其中明确的套餐变更与额度原子提交，共用凭据版本和观察时间保护； 空值及 `unknown` 不覆盖套餐，同族泛化值保留具体子类型。Token 刷新保留提交时的账号资料 +OpenAI 响应头的额度窗口由 `gateway-protocol` 按上游计量项归属解析；仅通过 `x-codex-active-limit` +标明具名额度时，通用 `x-codex-*` 窗口归入该计量项。Provider 按计量项独立保存，生图等具名额度 +不覆盖默认 Codex 窗口，也不改变账号级套餐访问结论 + OpenAI Provider 统一解释点数可用性，主动查询、耗尽恢复与失败回读复用同一规则。 Store 在账号事务内保存点数放行依据与快照；关闭自动使用时撤销该依据，迟到观测不能绕过开关。 放行条件与默认值见[账号额度 API](api.md#查询与账号字段)。