From fbdfe22757cc8a4057cf598005fdac58e0ca5a18 Mon Sep 17 00:00:00 2001 From: Zyy Date: Sat, 31 Oct 2026 21:29:50 +0800 Subject: [PATCH] feat(settings): update default concurrency policy Co-authored-by: Codex --- .../tests/postgres/runtime_settings.rs | 3 +- backend/migrations/.frozen-sha256 | 2 +- backend/migrations/0035_cpr_link.sql | 13 -------- ...0035_cpr_link_and_concurrency_defaults.sql | 32 +++++++++++++++++++ docs/api.md | 6 ++-- 5 files changed, 38 insertions(+), 18 deletions(-) delete mode 100644 backend/migrations/0035_cpr_link.sql create mode 100644 backend/migrations/0035_cpr_link_and_concurrency_defaults.sql diff --git a/backend/crates/gateway-store/tests/postgres/runtime_settings.rs b/backend/crates/gateway-store/tests/postgres/runtime_settings.rs index 8f4665ed3..b4542d8be 100644 --- a/backend/crates/gateway-store/tests/postgres/runtime_settings.rs +++ b/backend/crates/gateway-store/tests/postgres/runtime_settings.rs @@ -514,12 +514,13 @@ async fn concurrency_queue_settings_round_trip_into_the_runtime_snapshot() { let before = repository.load_runtime_settings().await.unwrap(); assert_eq!( ( + before.values.max_concurrent_per_account, before.values.max_waiting_per_key, before.values.max_waiting_per_account, before.values.concurrency_wait_timeout_seconds, before.values.openai_guardian_reserved_concurrency, ), - (0, 0, 30, 0) + (3, 10, 10, 30, 5) ); let mut update = settings_with_margin(1234); update.values.max_concurrent_per_account = 7; diff --git a/backend/migrations/.frozen-sha256 b/backend/migrations/.frozen-sha256 index 7cedf9653..83b8509a2 100644 --- a/backend/migrations/.frozen-sha256 +++ b/backend/migrations/.frozen-sha256 @@ -32,4 +32,4 @@ b49832e205405de7ef078f3b65c0c19f2c030b8984d42411598738b6b6c146bf 0031_proxy_and a0e409402611dbc76bb71393b8eda7c717756c18974dbabca305fdcb4b115f31 0032_usage_and_scheduling.sql 592b80972af3bd8b3527b42da3703ca1106ecaf8fd5273c676189ca28fcbd067 0033_proxy_and_reset_credits.sql 9c15a6b6c1f9eab8ad478772f98f8fc4a997b825d58d12cffe1a15579c5785b3 0034_account_error_protection.sql -573e35299924e62d8e5e1867d69b00756bd6f53f03516d8d3791a4688e9b793d 0035_cpr_link.sql +bd73379456839e8a12d7591ad2a839901af0f5ba27871248569abdfa26bc83ef 0035_cpr_link_and_concurrency_defaults.sql diff --git a/backend/migrations/0035_cpr_link.sql b/backend/migrations/0035_cpr_link.sql deleted file mode 100644 index 2762c609e..000000000 --- a/backend/migrations/0035_cpr_link.sql +++ /dev/null @@ -1,13 +0,0 @@ --- 互联公开权限随 Key 保存 -alter table client_api_keys - add column peer_sharing_json jsonb not null - default '{"enabled":false,"sharePool":false,"shareQuota":false,"sharePlans":false}'; - --- 各额度来源共用当前快照，协议入口另行限制响应大小 -alter table provider_accounts drop constraint provider_accounts_quota_ck; -alter table provider_accounts add constraint provider_accounts_quota_ck check ( - provider_quota_json is null or ( - jsonb_typeof(provider_quota_json) = 'object' - and octet_length(provider_quota_json::text) <= 8388608 - ) -); diff --git a/backend/migrations/0035_cpr_link_and_concurrency_defaults.sql b/backend/migrations/0035_cpr_link_and_concurrency_defaults.sql new file mode 100644 index 000000000..da63099ea --- /dev/null +++ b/backend/migrations/0035_cpr_link_and_concurrency_defaults.sql @@ -0,0 +1,32 @@ +-- 互联公开权限随 Key 保存 +alter table client_api_keys + add column peer_sharing_json jsonb not null + default '{"enabled":false,"sharePool":false,"shareQuota":false,"sharePlans":false}'; + +-- 各额度来源共用当前快照，协议入口另行限制响应大小 +alter table provider_accounts drop constraint provider_accounts_quota_ck; +alter table provider_accounts add constraint provider_accounts_quota_ck check ( + provider_quota_json is null or ( + jsonb_typeof(provider_quota_json) = 'object' + and octet_length(provider_quota_json::text) <= 8388608 + ) +); + +-- 初始并发策略，已保存的运行设置保持原值 +alter table runtime_settings + alter column max_concurrent_per_account set default 3, + alter column max_waiting_per_key set default 10, + alter column max_waiting_per_account set default 10, + alter column openai_guardian_reserved_concurrency set default 5; + +update runtime_settings +set max_concurrent_per_account = 3, + max_waiting_per_key = 10, + max_waiting_per_account = 10, + openai_guardian_reserved_concurrency = 5 +where config_revision = 1 + and max_concurrent_per_account = 5 + and max_waiting_per_key = 0 + and max_waiting_per_account = 0 + and concurrency_wait_timeout_seconds = 30 + and openai_guardian_reserved_concurrency = 0; diff --git a/docs/api.md b/docs/api.md index 1df56291a..49fd4ea59 100644 --- a/docs/api.md +++ b/docs/api.md @@ -1915,11 +1915,11 @@ UA、平台参数及自动 Git 上下文版本由 [OpenAI 上游客户端身份] ### 并发与排队 -`maxConcurrentPerAccount` 是默认账号并发上限，初始值为 5，取值 0～4294967295；`0` 表示不限制。 +`maxConcurrentPerAccount` 是默认账号并发上限，初始值为 3，取值 0～4294967295；`0` 表示不限制。 账号的 `concurrencyLimit: null` 继承该默认值，单独设置的正数上限仍优先生效。 无限并发仍统计在途请求，并遵守账号可用性与 Client Key 限制 -`maxWaitingPerKey` 与 `maxWaitingPerAccount` 是全局统一的普通排队容量，取值 0～1,000，默认 0（关闭）； +`maxWaitingPerKey` 与 `maxWaitingPerAccount` 是全局统一的普通排队容量，取值 0～1,000，默认 10，`0` 表示关闭； 每个 Key、每个账号各自独立计数，没有单对象覆盖字段。执行并发为 5、最大排队数为 5 时， 该对象最多容纳 5 个执行请求与 5 个等待请求。Key 并发为 0（不限）时跳过 Key 排队。 OpenAI 严格亲和下后代线程的会话账号等待不随普通账号排队关闭，见 [Client Key 等待规则](#7-client-key)。 @@ -1927,7 +1927,7 @@ OpenAI 严格亲和下后代线程的会话账号等待不随普通账号排队 切换账号或内部重试不重新计时；若插件设置了请求总时限，等待也计入该时限。排队超时不用于中断已开始的上游生成。 设置更新请求须包含这三个字段，新请求使用更新后的快照 -`openaiGuardianReservedConcurrency`（默认 0，取值 0～4,294,967,295）设置 Codex Guardian 自动审批的每账号独立并发，保存后对新请求生效。 +`openaiGuardianReservedConcurrency`（默认 5，取值 0～4,294,967,295）设置 Codex Guardian 自动审批的每账号独立并发，保存后对新请求生效。 Guardian 以 `subagent_kind` 或 `client_metadata.x-openai-subagent` 值 `guardian` 识别。取值 R 大于 0 时， 普通请求保留账号原上限 L，Guardian 单独计数并最多运行 R 个，两类请求互不占用名额，也不借用对方空位。 例如 L 为 10、R 为 3 时，可以同时运行 10 个普通请求和 3 个审批请求；普通并发不限时，审批仍受 R 限制。