perf(postgres): 回收突发连接并等待验收副作用

失败 Run 的未提交任务已取消，但退款 outbox 尚未完成时下一轮会继承 API 连接高水位和 canceled acquire 增量，污染容量验收。

新增 AI_GATEWAY_DATABASE_MAX_CONN_IDLE_SECONDS，生产配置 30 秒回收突发空闲连接；验收中止后等待旧任务、退款、回调和数据库连接全部收敛后才允许新负载启动。

验证：Go 全量测试、gofmt、bash -n、ShellCheck、kubectl kustomize 和 git diff --check 通过。

deploy/kubernetes/easyai-ai-gateway-cluster-release

@@ -21,6 +21,7 @@ source "$config_file"
 : "${AI_GATEWAY_DATABASE_MAX_CONNS:=32}"
 : "${AI_GATEWAY_API_DATABASE_MAX_CONNS:=64}"
 : "${AI_GATEWAY_DATABASE_MIN_IDLE_CONNS:=4}"
+: "${AI_GATEWAY_DATABASE_MAX_CONN_IDLE_SECONDS:=30}"
 : "${AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY:=24}"
 : "${AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY:=24}"
 [[ $RELEASES_DIR == /* && $NAMESPACE =~ ^[a-z0-9-]+$ ]] || {
@@ -34,6 +35,7 @@ for capacity_value in \
   "$AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT" \
   "$AI_GATEWAY_DATABASE_MAX_CONNS" \
   "$AI_GATEWAY_API_DATABASE_MAX_CONNS" \
+  "$AI_GATEWAY_DATABASE_MAX_CONN_IDLE_SECONDS" \
   "$AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY" \
   "$AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY"; do
   [[ $capacity_value =~ ^[1-9][0-9]*$ ]] || {
@@ -53,6 +55,7 @@ done
    AI_GATEWAY_API_DATABASE_MAX_CONNS <= 256 &&
    AI_GATEWAY_DATABASE_MIN_IDLE_CONNS <= AI_GATEWAY_DATABASE_MAX_CONNS &&
    AI_GATEWAY_DATABASE_MIN_IDLE_CONNS <= AI_GATEWAY_API_DATABASE_MAX_CONNS &&
+   AI_GATEWAY_DATABASE_MAX_CONN_IDLE_SECONDS <= 3600 &&
    AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY <= 256 &&
    AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY <= 1024 )) || {
   echo 'worker capacity configuration exceeds the release safety bounds' >&2
@@ -138,6 +141,7 @@ rollout_worker_site() {
     "AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT=$AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT" \
     "AI_GATEWAY_DATABASE_MAX_CONNS=$AI_GATEWAY_DATABASE_MAX_CONNS" \
     "AI_GATEWAY_DATABASE_MIN_IDLE_CONNS=$AI_GATEWAY_DATABASE_MIN_IDLE_CONNS" \
+    "AI_GATEWAY_DATABASE_MAX_CONN_IDLE_SECONDS=$AI_GATEWAY_DATABASE_MAX_CONN_IDLE_SECONDS" \
     "AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY=$AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY" \
     "AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY=$AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY"
   "${kubectl[@]}" scale "deployment/easyai-worker-$site" -n "$NAMESPACE" \
@@ -158,7 +162,8 @@ rollout_api_capacity_site() {
   "${kubectl[@]}" set env "deployment/easyai-api-$site" -n "$NAMESPACE" \
     "AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT=$AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT" \
     "AI_GATEWAY_DATABASE_MAX_CONNS=$AI_GATEWAY_API_DATABASE_MAX_CONNS" \
-    "AI_GATEWAY_DATABASE_MIN_IDLE_CONNS=$AI_GATEWAY_DATABASE_MIN_IDLE_CONNS"
+    "AI_GATEWAY_DATABASE_MIN_IDLE_CONNS=$AI_GATEWAY_DATABASE_MIN_IDLE_CONNS" \
+    "AI_GATEWAY_DATABASE_MAX_CONN_IDLE_SECONDS=$AI_GATEWAY_DATABASE_MAX_CONN_IDLE_SECONDS"
   "${kubectl[@]}" rollout status "deployment/easyai-api-$site" -n "$NAMESPACE" --timeout=300s
 }
 
@@ -170,7 +175,7 @@ apply_capacity_config() {
   verify_site hongkong
   verify_site ningbo
   wait_for_url 'public readiness' "$PUBLIC_BASE_URL/api/v1/readyz" '"ok":true'
-  echo "production_capacity=PASS ningbo_replicas=$AI_GATEWAY_WORKER_REPLICAS_NINGBO hongkong_replicas=$AI_GATEWAY_WORKER_REPLICAS_HONGKONG instance_limit=$AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT global_limit=$AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT worker_database_pool=$AI_GATEWAY_DATABASE_MAX_CONNS api_database_pool=$AI_GATEWAY_API_DATABASE_MAX_CONNS database_min_idle=$AI_GATEWAY_DATABASE_MIN_IDLE_CONNS media_concurrency=$AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY"
+  echo "production_capacity=PASS ningbo_replicas=$AI_GATEWAY_WORKER_REPLICAS_NINGBO hongkong_replicas=$AI_GATEWAY_WORKER_REPLICAS_HONGKONG instance_limit=$AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT global_limit=$AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT worker_database_pool=$AI_GATEWAY_DATABASE_MAX_CONNS api_database_pool=$AI_GATEWAY_API_DATABASE_MAX_CONNS database_min_idle=$AI_GATEWAY_DATABASE_MIN_IDLE_CONNS database_max_idle_seconds=$AI_GATEWAY_DATABASE_MAX_CONN_IDLE_SECONDS media_concurrency=$AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY"
 }
 
 rollout_site() {
@@ -183,7 +188,8 @@ 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_DATABASE_MAX_CONNS=$AI_GATEWAY_API_DATABASE_MAX_CONNS" \
-      "AI_GATEWAY_DATABASE_MIN_IDLE_CONNS=$AI_GATEWAY_DATABASE_MIN_IDLE_CONNS"
+      "AI_GATEWAY_DATABASE_MIN_IDLE_CONNS=$AI_GATEWAY_DATABASE_MIN_IDLE_CONNS" \
+      "AI_GATEWAY_DATABASE_MAX_CONN_IDLE_SECONDS=$AI_GATEWAY_DATABASE_MAX_CONN_IDLE_SECONDS"
     "${kubectl[@]}" set image "deployment/easyai-api-$site" -n "$NAMESPACE" \
       "api=$api_image"
     "${kubectl[@]}" rollout status "deployment/easyai-api-$site" -n "$NAMESPACE" --timeout=300s

deploy/kubernetes/easyai-ai-gateway-cluster-release.conf.example

@@ -12,5 +12,6 @@ AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT=48
 AI_GATEWAY_DATABASE_MAX_CONNS=32
 AI_GATEWAY_API_DATABASE_MAX_CONNS=64
 AI_GATEWAY_DATABASE_MIN_IDLE_CONNS=4
+AI_GATEWAY_DATABASE_MAX_CONN_IDLE_SECONDS=30
 AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY=24
 AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY=24

deploy/kubernetes/production/application.yaml

@@ -56,6 +56,8 @@ spec:
               value: "64"
             - name: AI_GATEWAY_DATABASE_MIN_IDLE_CONNS
               value: "4"
+            - name: AI_GATEWAY_DATABASE_MAX_CONN_IDLE_SECONDS
+              value: "30"
             - name: AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY
               value: "16"
             - name: AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY
@@ -187,6 +189,8 @@ spec:
               value: "64"
             - name: AI_GATEWAY_DATABASE_MIN_IDLE_CONNS
               value: "4"
+            - name: AI_GATEWAY_DATABASE_MAX_CONN_IDLE_SECONDS
+              value: "30"
             - name: AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY
               value: "16"
             - name: AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY
@@ -318,6 +322,8 @@ spec:
               value: "32"
             - name: AI_GATEWAY_DATABASE_MIN_IDLE_CONNS
               value: "4"
+            - name: AI_GATEWAY_DATABASE_MAX_CONN_IDLE_SECONDS
+              value: "30"
             - name: AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT
               value: "24"
             - name: AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY
@@ -451,6 +457,8 @@ spec:
               value: "32"
             - name: AI_GATEWAY_DATABASE_MIN_IDLE_CONNS
               value: "4"
+            - name: AI_GATEWAY_DATABASE_MAX_CONN_IDLE_SECONDS
+              value: "30"
             - name: AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT
               value: "24"
             - name: AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY