BCAI/easyai-ai-gateway
3
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

perf(worker): 解耦准入调度与远端执行

Browse Source 
原因：线上 P24 同构验收确认两台香港 Worker 的全局容量为 48，但跨地域逐行准入事务每轮只能形成 8 个活跃租约，队列最老等待超过 15 分钟。\n\n影响：新增可配置的异步准入 dispatcher 角色；生产两地 API 负责准入和过期回收，Worker 仅执行 River job。当前主库同站点 API 可低延迟填满容量，主库切换后另一地 API 通过既有数据库锁安全接管；未配置环境变量时保持原 Worker 一体化行为。\n\n验证：Go 全量测试、go vet、gofmt、真实 PostgreSQL 1000 任务双进程回归、前端 lint/test/build、Kubernetes server-side dry-run、Compose、ShellCheck、cluster/manual release tests 全部通过。
This commit is contained in:
wangbo
2026-08-01 23:24:30 +08:00
parent 9c093974a1
commit a4046c2b81
10 changed files with 96 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files
deploy/kubernetes/easyai-ai-gateway-cluster-release
+5
View File
@@ -403,6 +403,7 @@ rollout_worker_site() {
"AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT=$AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT" \
"AI_GATEWAY_ASYNC_WORKER_GLOBAL_HARD_LIMIT=$AI_GATEWAY_ASYNC_WORKER_GLOBAL_HARD_LIMIT" \
"AI_GATEWAY_ASYNC_ADMISSION_MICROBATCH_SIZE=$AI_GATEWAY_ASYNC_ADMISSION_MICROBATCH_SIZE" \
"AI_GATEWAY_ASYNC_ADMISSION_DISPATCHER_ENABLED=false" \
"AI_GATEWAY_DATABASE_MAX_CONNS=$AI_GATEWAY_DATABASE_MAX_CONNS" \
"AI_GATEWAY_DATABASE_CRITICAL_MAX_CONNS=$AI_GATEWAY_DATABASE_CRITICAL_MAX_CONNS" \
"AI_GATEWAY_DATABASE_RIVER_MAX_CONNS=$AI_GATEWAY_DATABASE_RIVER_MAX_CONNS" \
@@ -433,6 +434,8 @@ rollout_worker_site() {
-n "$NAMESPACE" --timeout=300s || return 1
[[ $("${kubectl[@]}" get deployment "easyai-worker-$site" -n "$NAMESPACE" \
-o jsonpath='{.spec.template.spec.containers[?(@.name=="worker")].env[?(@.name=="AI_GATEWAY_PROCESS_ROLE")].value}') == worker ]] || return 1
[[ $("${kubectl[@]}" get deployment "easyai-worker-$site" -n "$NAMESPACE" \
-o jsonpath='{.spec.template.spec.containers[?(@.name=="worker")].env[?(@.name=="AI_GATEWAY_ASYNC_ADMISSION_DISPATCHER_ENABLED")].value}') == false ]] || return 1
ready_replicas=$("${kubectl[@]}" get deployment "easyai-worker-$site" -n "$NAMESPACE" \
-o json | jq -r '.status.readyReplicas // 0') || return 1
[[ $ready_replicas == "$replicas" ]] || return 1
@@ -592,6 +595,7 @@ rollout_api_capacity_site() {
"${kubectl[@]}" set env "deployment/easyai-api-$site" -n "$NAMESPACE" \
"AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT=$AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT" \
"AI_GATEWAY_ASYNC_WORKER_GLOBAL_HARD_LIMIT=$AI_GATEWAY_ASYNC_WORKER_GLOBAL_HARD_LIMIT" \
"AI_GATEWAY_ASYNC_ADMISSION_DISPATCHER_ENABLED=true" \
"AI_GATEWAY_DATABASE_MAX_CONNS=$AI_GATEWAY_API_DATABASE_MAX_CONNS" \
"AI_GATEWAY_DATABASE_CRITICAL_MAX_CONNS=$AI_GATEWAY_DATABASE_CRITICAL_MAX_CONNS" \
"AI_GATEWAY_DATABASE_RIVER_MAX_CONNS=$AI_GATEWAY_API_DATABASE_RIVER_MAX_CONNS" \
@@ -637,6 +641,7 @@ rollout_site() {
"${kubectl[@]}" set env "deployment/easyai-api-$site" -n "$NAMESPACE" \
"AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT=$AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT" \
"AI_GATEWAY_ASYNC_WORKER_GLOBAL_HARD_LIMIT=$AI_GATEWAY_ASYNC_WORKER_GLOBAL_HARD_LIMIT" \
"AI_GATEWAY_ASYNC_ADMISSION_DISPATCHER_ENABLED=true" \
"AI_GATEWAY_DATABASE_MAX_CONNS=$AI_GATEWAY_API_DATABASE_MAX_CONNS" \
"AI_GATEWAY_DATABASE_CRITICAL_MAX_CONNS=$AI_GATEWAY_DATABASE_CRITICAL_MAX_CONNS" \
"AI_GATEWAY_DATABASE_RIVER_MAX_CONNS=$AI_GATEWAY_API_DATABASE_RIVER_MAX_CONNS" \
deploy/kubernetes/production/application.yaml
+8
View File
@@ -52,6 +52,8 @@ spec:
value: api
- name: AI_GATEWAY_ASYNC_QUEUE_WORKER_ENABLED
value: "false"
- name: AI_GATEWAY_ASYNC_ADMISSION_DISPATCHER_ENABLED
value: "true"
- name: AI_GATEWAY_DATABASE_MAX_CONNS
value: "32"
- name: AI_GATEWAY_DATABASE_CRITICAL_MAX_CONNS
@@ -330,6 +332,8 @@ spec:
value: api
- name: AI_GATEWAY_ASYNC_QUEUE_WORKER_ENABLED
value: "false"
- name: AI_GATEWAY_ASYNC_ADMISSION_DISPATCHER_ENABLED
value: "true"
- name: AI_GATEWAY_DATABASE_MAX_CONNS
value: "32"
- name: AI_GATEWAY_DATABASE_CRITICAL_MAX_CONNS
@@ -501,6 +505,8 @@ spec:
value: worker
- name: AI_GATEWAY_ASYNC_QUEUE_WORKER_ENABLED
value: "true"
- name: AI_GATEWAY_ASYNC_ADMISSION_DISPATCHER_ENABLED
value: "false"
- name: AI_GATEWAY_DATABASE_MAX_CONNS
value: "32"
- name: AI_GATEWAY_DATABASE_CRITICAL_MAX_CONNS
@@ -676,6 +682,8 @@ spec:
value: worker
- name: AI_GATEWAY_ASYNC_QUEUE_WORKER_ENABLED
value: "true"
- name: AI_GATEWAY_ASYNC_ADMISSION_DISPATCHER_ENABLED
value: "false"
- name: AI_GATEWAY_DATABASE_MAX_CONNS
value: "32"
- name: AI_GATEWAY_DATABASE_CRITICAL_MAX_CONNS