feat(deploy): 切换宁波单节点生产拓扑

移除已停用的香港 API、Web、Worker 和边缘 Service 渲染结果，将 Worker 与容量控制器固定到宁波单实例，并在发布成功后清理旧香港资源。\n\n容量控制器忽略最大副本为零的站点，集群环境允许不再声明香港主机；单节点 production acceptance 不再依赖香港 Service、Deployment 或 SSH 回退。\n\n验证：Go 全量测试与 go vet；迁移安全检查；Kustomize 单节点渲染；ShellCheck；cluster release、production acceptance 和 manual release 测试。
2026-08-03 09:59:34 +08:00
parent 5e9a7888ae
commit 4ff85d96e0
12 changed files with 156 additions and 125 deletions
@@ -15,8 +15,7 @@ source "$config_file"
 : "${PUBLIC_BASE_URL:=https://ai.51easyai.com}"
 : "${K3S_BIN:=/usr/local/bin/k3s}"
 : "${DESIRED_STATE_DIR:=/usr/local/share/easyai-ai-gateway-release/production}"
-: "${AI_GATEWAY_WORKER_REPLICAS_NINGBO:=0}"
-: "${AI_GATEWAY_WORKER_REPLICAS_HONGKONG:=2}"
+: "${AI_GATEWAY_WORKER_REPLICAS_NINGBO:=1}"
 : "${AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT:=24}"
 : "${AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT:=48}"
 : "${AI_GATEWAY_ASYNC_WORKER_GLOBAL_HARD_LIMIT:=$AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT}"
@@ -33,10 +32,8 @@ source "$config_file"
 : "${AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY:=24}"
 : "${AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY:=24}"
 : "${AI_GATEWAY_WORKER_AUTOSCALING_ENABLED:=false}"
-: "${AI_GATEWAY_WORKER_MIN_REPLICAS_NINGBO:=0}"
-: "${AI_GATEWAY_WORKER_MIN_REPLICAS_HONGKONG:=1}"
-: "${AI_GATEWAY_WORKER_MAX_REPLICAS_NINGBO:=0}"
-: "${AI_GATEWAY_WORKER_MAX_REPLICAS_HONGKONG:=2}"
+: "${AI_GATEWAY_WORKER_MIN_REPLICAS_NINGBO:=1}"
+: "${AI_GATEWAY_WORKER_MAX_REPLICAS_NINGBO:=1}"
 : "${AI_GATEWAY_WORKER_TARGET_OUTSTANDING_PER_REPLICA:=$((AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT * 2))}"
 : "${AI_GATEWAY_WORKER_SCALE_UP_WINDOW_SECONDS:=20}"
 : "${AI_GATEWAY_WORKER_SCALE_DOWN_STABILIZATION_SECONDS:=600}"
@@ -45,7 +42,7 @@ source "$config_file"
 : "${AI_GATEWAY_NODE_MEMORY_HARD_PERCENT:=85}"
 : "${AI_GATEWAY_NODE_CPU_TARGET_PERCENT:=70}"
 : "${AI_GATEWAY_POSTGRES_CONNECTION_BUDGET:=150}"
-: "${AI_GATEWAY_POSTGRES_NON_WORKER_CONNECTION_BUDGET:=$((AI_GATEWAY_API_DATABASE_MAX_CONNS * 2 + 8))}"
+: "${AI_GATEWAY_POSTGRES_NON_WORKER_CONNECTION_BUDGET:=$((AI_GATEWAY_API_DATABASE_MAX_CONNS + 4))}"
 [[ $RELEASES_DIR == /* && $NAMESPACE =~ ^[a-z0-9-]+$ ]] || {
  echo 'invalid cluster release configuration' >&2
  exit 1
@@ -58,11 +55,8 @@ source "$config_file"
 }
 for replica_value in \
  "$AI_GATEWAY_WORKER_REPLICAS_NINGBO" \
-  "$AI_GATEWAY_WORKER_REPLICAS_HONGKONG" \
  "$AI_GATEWAY_WORKER_MIN_REPLICAS_NINGBO" \
-  "$AI_GATEWAY_WORKER_MIN_REPLICAS_HONGKONG" \
-  "$AI_GATEWAY_WORKER_MAX_REPLICAS_NINGBO" \
-  "$AI_GATEWAY_WORKER_MAX_REPLICAS_HONGKONG"; do
+  "$AI_GATEWAY_WORKER_MAX_REPLICAS_NINGBO"; do
  [[ $replica_value =~ ^[0-9]+$ ]] || {
    echo 'Worker replica configuration must contain non-negative integers' >&2
    exit 1
@@ -110,8 +104,7 @@ done
  exit 1
 }
 (( AI_GATEWAY_WORKER_REPLICAS_NINGBO <= 16 &&
-   AI_GATEWAY_WORKER_REPLICAS_HONGKONG <= 16 &&
-   AI_GATEWAY_WORKER_REPLICAS_NINGBO + AI_GATEWAY_WORKER_REPLICAS_HONGKONG >= 1 &&
+   AI_GATEWAY_WORKER_REPLICAS_NINGBO >= 1 &&
   AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT <= 256 &&
   AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT <= 512 &&
   AI_GATEWAY_ASYNC_WORKER_GLOBAL_HARD_LIMIT <= 512 &&
@@ -133,17 +126,14 @@ done
   AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY <= 1024 &&
   AI_GATEWAY_WORKER_MIN_REPLICAS_NINGBO <= AI_GATEWAY_WORKER_REPLICAS_NINGBO &&
   AI_GATEWAY_WORKER_REPLICAS_NINGBO <= AI_GATEWAY_WORKER_MAX_REPLICAS_NINGBO &&
-   AI_GATEWAY_WORKER_MIN_REPLICAS_HONGKONG <= AI_GATEWAY_WORKER_REPLICAS_HONGKONG &&
-   AI_GATEWAY_WORKER_REPLICAS_HONGKONG <= AI_GATEWAY_WORKER_MAX_REPLICAS_HONGKONG &&
   AI_GATEWAY_WORKER_MAX_REPLICAS_NINGBO <= 16 &&
-   AI_GATEWAY_WORKER_MAX_REPLICAS_HONGKONG <= 16 &&
   AI_GATEWAY_NODE_MEMORY_TARGET_PERCENT < AI_GATEWAY_NODE_MEMORY_HARD_PERCENT &&
   AI_GATEWAY_NODE_MEMORY_HARD_PERCENT <= 95 &&
   AI_GATEWAY_NODE_CPU_TARGET_PERCENT <= 90 &&
   AI_GATEWAY_POSTGRES_CONNECTION_BUDGET <= 150 &&
   AI_GATEWAY_POSTGRES_NON_WORKER_CONNECTION_BUDGET < AI_GATEWAY_POSTGRES_CONNECTION_BUDGET &&
   AI_GATEWAY_POSTGRES_NON_WORKER_CONNECTION_BUDGET +
-     (AI_GATEWAY_WORKER_REPLICAS_NINGBO + AI_GATEWAY_WORKER_REPLICAS_HONGKONG) *
+     AI_GATEWAY_WORKER_REPLICAS_NINGBO *
       AI_GATEWAY_DATABASE_MAX_CONNS <= AI_GATEWAY_POSTGRES_CONNECTION_BUDGET )) || {
  echo 'worker capacity configuration exceeds the release safety bounds' >&2
  exit 1
@@ -169,7 +159,6 @@ manifest_get() {
 capacity_config_hash() {
  printf '%s\n' \
    "$AI_GATEWAY_WORKER_REPLICAS_NINGBO" \
-    "$AI_GATEWAY_WORKER_REPLICAS_HONGKONG" \
    "$AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT" \
    "$AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT" \
    "$AI_GATEWAY_ASYNC_WORKER_GLOBAL_HARD_LIMIT" \
@@ -187,9 +176,7 @@ capacity_config_hash() {
    "$AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY" \
    "$AI_GATEWAY_WORKER_AUTOSCALING_ENABLED" \
    "$AI_GATEWAY_WORKER_MIN_REPLICAS_NINGBO" \
-    "$AI_GATEWAY_WORKER_MIN_REPLICAS_HONGKONG" \
    "$AI_GATEWAY_WORKER_MAX_REPLICAS_NINGBO" \
-    "$AI_GATEWAY_WORKER_MAX_REPLICAS_HONGKONG" \
    "$AI_GATEWAY_WORKER_TARGET_OUTSTANDING_PER_REPLICA" \
    "$AI_GATEWAY_WORKER_SCALE_UP_WINDOW_SECONDS" \
    "$AI_GATEWAY_WORKER_SCALE_DOWN_STABILIZATION_SECONDS" \
@@ -207,11 +194,8 @@ record_capacity_config() {
 	local runtime_patch
 	runtime_patch=$(jq -nc \
 		--arg replicasNingbo "$AI_GATEWAY_WORKER_REPLICAS_NINGBO" \
-		--arg replicasHongkong "$AI_GATEWAY_WORKER_REPLICAS_HONGKONG" \
 		--arg minNingbo "$AI_GATEWAY_WORKER_MIN_REPLICAS_NINGBO" \
-		--arg minHongkong "$AI_GATEWAY_WORKER_MIN_REPLICAS_HONGKONG" \
 		--arg maxNingbo "$AI_GATEWAY_WORKER_MAX_REPLICAS_NINGBO" \
-		--arg maxHongkong "$AI_GATEWAY_WORKER_MAX_REPLICAS_HONGKONG" \
 		--arg autoscaling "$AI_GATEWAY_WORKER_AUTOSCALING_ENABLED" \
 		--arg instanceLimit "$AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT" \
 		--arg globalLimit "$AI_GATEWAY_ASYNC_WORKER_GLOBAL_HARD_LIMIT" \
@@ -229,11 +213,11 @@ record_capacity_config() {
 		--arg postgresBudget "$AI_GATEWAY_POSTGRES_CONNECTION_BUDGET" \
 		'{data:{
 			AI_GATEWAY_WORKER_REPLICAS_NINGBO:$replicasNingbo,
-			AI_GATEWAY_WORKER_REPLICAS_HONGKONG:$replicasHongkong,
+			AI_GATEWAY_WORKER_REPLICAS_HONGKONG:null,
 			AI_GATEWAY_WORKER_MIN_REPLICAS_NINGBO:$minNingbo,
-			AI_GATEWAY_WORKER_MIN_REPLICAS_HONGKONG:$minHongkong,
+			AI_GATEWAY_WORKER_MIN_REPLICAS_HONGKONG:null,
 			AI_GATEWAY_WORKER_MAX_REPLICAS_NINGBO:$maxNingbo,
-			AI_GATEWAY_WORKER_MAX_REPLICAS_HONGKONG:$maxHongkong,
+			AI_GATEWAY_WORKER_MAX_REPLICAS_HONGKONG:null,
 			AI_GATEWAY_WORKER_AUTOSCALING_ENABLED:$autoscaling,
 			AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT:$instanceLimit,
 			AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT:$globalLimit,
@@ -259,8 +243,8 @@ record_capacity_config() {
  "${kubectl[@]}" annotate namespace "$NAMESPACE" \
    "easyai.io/config-hash=$config_hash" --overwrite >/dev/null
  for deployment in \
-    easyai-api-ningbo easyai-api-hongkong \
-    easyai-worker-ningbo easyai-worker-hongkong \
+    easyai-api-ningbo \
+    easyai-worker-ningbo \
    easyai-capacity-controller; do
    "${kubectl[@]}" annotate deployment "$deployment" -n "$NAMESPACE" \
      "easyai.io/config-hash=$config_hash" --overwrite >/dev/null
@@ -299,11 +283,6 @@ verify_site() {
  local api_port
  local web_port
  case $site in
-    hongkong)
-      address=10.77.0.2
-      api_port=31089
-      web_port=31179
-      ;;
    ningbo)
      address=10.77.0.1
      api_port=31088
@@ -316,10 +295,19 @@ verify_site() {
  wait_for_url "$site Web" "http://$address:$web_port/" 'EasyAI AI Gateway'
 }

+retire_hongkong_site() {
+  "${kubectl[@]}" delete deployment \
+    easyai-api-hongkong easyai-worker-hongkong easyai-web-hongkong \
+    -n "$NAMESPACE" --ignore-not-found=true --wait=true >/dev/null
+  "${kubectl[@]}" delete service \
+    api-hongkong-edge web-hongkong-edge \
+    -n "$NAMESPACE" --ignore-not-found=true --wait=true >/dev/null
+}
+
 apply_desired_state() {
  local api_image=$1
  local web_image=$2
-  local capacity_controller_replicas=${3:-2}
+  local capacity_controller_replicas=${3:-1}
  local api_repository=${api_image%@*}
  local api_digest=${api_image#*@}
  local web_repository=${web_image%@*}
@@ -333,7 +321,7 @@ apply_desired_state() {
    "$working_directory/kustomization.yaml"
  if [[ $capacity_controller_replicas == 0 ]]; then
    sed -i \
-      '/^  name: easyai-capacity-controller$/,/^---$/s/^  replicas: 2$/  replicas: 0/' \
+      '/^  name: easyai-capacity-controller$/,/^---$/s/^  replicas: 1$/  replicas: 0/' \
      "$working_directory/application.yaml"
  fi
  "${kubectl[@]}" apply --server-side --dry-run=server --force-conflicts \
@@ -395,7 +383,6 @@ rollout_worker_site() {
  local replicas ready_replicas
  case $site in
    ningbo) replicas=$AI_GATEWAY_WORKER_REPLICAS_NINGBO ;;
-    hongkong) replicas=$AI_GATEWAY_WORKER_REPLICAS_HONGKONG ;;
    *) return 1 ;;
  esac
  "${kubectl[@]}" set env "deployment/easyai-worker-$site" -n "$NAMESPACE" \
@@ -414,11 +401,8 @@ rollout_worker_site() {
    "AI_GATEWAY_REVISION=${ACTIVE_RELEASE_SHA:-}" \
    "AI_GATEWAY_WORKER_AUTOSCALING_ENABLED=$AI_GATEWAY_WORKER_AUTOSCALING_ENABLED" \
    "AI_GATEWAY_WORKER_REPLICAS_NINGBO=$AI_GATEWAY_WORKER_REPLICAS_NINGBO" \
-    "AI_GATEWAY_WORKER_REPLICAS_HONGKONG=$AI_GATEWAY_WORKER_REPLICAS_HONGKONG" \
    "AI_GATEWAY_WORKER_MIN_REPLICAS_NINGBO=$AI_GATEWAY_WORKER_MIN_REPLICAS_NINGBO" \
-    "AI_GATEWAY_WORKER_MIN_REPLICAS_HONGKONG=$AI_GATEWAY_WORKER_MIN_REPLICAS_HONGKONG" \
    "AI_GATEWAY_WORKER_MAX_REPLICAS_NINGBO=$AI_GATEWAY_WORKER_MAX_REPLICAS_NINGBO" \
-    "AI_GATEWAY_WORKER_MAX_REPLICAS_HONGKONG=$AI_GATEWAY_WORKER_MAX_REPLICAS_HONGKONG" \
    "AI_GATEWAY_WORKER_TARGET_OUTSTANDING_PER_REPLICA=$AI_GATEWAY_WORKER_TARGET_OUTSTANDING_PER_REPLICA" || return 1
  "${kubectl[@]}" set resources "deployment/easyai-worker-$site" -n "$NAMESPACE" \
    --containers=worker \
@@ -446,11 +430,8 @@ rollout_capacity_controller() {
  "${kubectl[@]}" set env deployment/easyai-capacity-controller -n "$NAMESPACE" \
    "AI_GATEWAY_WORKER_AUTOSCALING_ENABLED=$AI_GATEWAY_WORKER_AUTOSCALING_ENABLED" \
    "AI_GATEWAY_WORKER_REPLICAS_NINGBO=$AI_GATEWAY_WORKER_REPLICAS_NINGBO" \
-    "AI_GATEWAY_WORKER_REPLICAS_HONGKONG=$AI_GATEWAY_WORKER_REPLICAS_HONGKONG" \
    "AI_GATEWAY_WORKER_MIN_REPLICAS_NINGBO=$AI_GATEWAY_WORKER_MIN_REPLICAS_NINGBO" \
-    "AI_GATEWAY_WORKER_MIN_REPLICAS_HONGKONG=$AI_GATEWAY_WORKER_MIN_REPLICAS_HONGKONG" \
    "AI_GATEWAY_WORKER_MAX_REPLICAS_NINGBO=$AI_GATEWAY_WORKER_MAX_REPLICAS_NINGBO" \
-    "AI_GATEWAY_WORKER_MAX_REPLICAS_HONGKONG=$AI_GATEWAY_WORKER_MAX_REPLICAS_HONGKONG" \
    "AI_GATEWAY_WORKER_TARGET_OUTSTANDING_PER_REPLICA=$AI_GATEWAY_WORKER_TARGET_OUTSTANDING_PER_REPLICA" \
    "AI_GATEWAY_WORKER_SCALE_UP_WINDOW_SECONDS=$AI_GATEWAY_WORKER_SCALE_UP_WINDOW_SECONDS" \
    "AI_GATEWAY_WORKER_SCALE_DOWN_STABILIZATION_SECONDS=$AI_GATEWAY_WORKER_SCALE_DOWN_STABILIZATION_SECONDS" \
@@ -469,7 +450,7 @@ rollout_capacity_controller() {
      "capacity-controller=$api_image" || return 1
  fi
  "${kubectl[@]}" scale deployment/easyai-capacity-controller \
-    -n "$NAMESPACE" --replicas=2 || return 1
+    -n "$NAMESPACE" --replicas=1 || return 1
  "${kubectl[@]}" rollout status deployment/easyai-capacity-controller \
    -n "$NAMESPACE" --timeout=300s || return 1
 }
@@ -493,6 +474,7 @@ spec:
  restartPolicy: Never
  serviceAccountName: easyai-capacity-controller
  nodeSelector:
+    easyai.io/site: ningbo
    easyai.io/workload: "true"
  securityContext:
    runAsNonRoot: true
@@ -615,20 +597,17 @@ apply_capacity_config() {
    echo 'current release SHA is unavailable for capacity update' >&2
    return 1
  }
-  rollout_worker_site hongkong ""
  rollout_worker_site ningbo ""
-  rollout_api_capacity_site hongkong
  rollout_api_capacity_site ningbo
  if [[ $capacity_controller_enabled == true ]]; then
    rollout_capacity_controller ""
  else
    "${kubectl[@]}" scale deployment/easyai-capacity-controller -n "$NAMESPACE" --replicas=0
  fi
-  verify_site hongkong
  verify_site ningbo
  wait_for_url 'public readiness' "$PUBLIC_BASE_URL/api/v1/readyz" '"ok":true'
  record_capacity_config
-  echo "production_capacity=PASS config_hash=$(capacity_config_hash) ningbo_replicas=$AI_GATEWAY_WORKER_REPLICAS_NINGBO hongkong_replicas=$AI_GATEWAY_WORKER_REPLICAS_HONGKONG instance_limit=$AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT global_limit=$AI_GATEWAY_ASYNC_WORKER_GLOBAL_HARD_LIMIT admission_microbatch=$AI_GATEWAY_ASYNC_ADMISSION_MICROBATCH_SIZE 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=$AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY worker_memory_request_mib=$AI_GATEWAY_WORKER_MEMORY_REQUEST_MIB worker_cpu_request_millicores=$AI_GATEWAY_WORKER_CPU_REQUEST_MILLICORES autoscaling=$AI_GATEWAY_WORKER_AUTOSCALING_ENABLED"
+  echo "production_capacity=PASS config_hash=$(capacity_config_hash) site=ningbo replicas=$AI_GATEWAY_WORKER_REPLICAS_NINGBO instance_limit=$AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT global_limit=$AI_GATEWAY_ASYNC_WORKER_GLOBAL_HARD_LIMIT admission_microbatch=$AI_GATEWAY_ASYNC_ADMISSION_MICROBATCH_SIZE worker_database_pool=$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 worker_memory_request_mib=$AI_GATEWAY_WORKER_MEMORY_REQUEST_MIB worker_cpu_request_millicores=$AI_GATEWAY_WORKER_CPU_REQUEST_MILLICORES autoscaling=$AI_GATEWAY_WORKER_AUTOSCALING_ENABLED"
 }

 rollout_site() {
@@ -786,10 +765,10 @@ record_current() {
    "easyai.io/config-hash=$(capacity_config_hash)" \
    --overwrite >/dev/null
  for deployment in \
-    easyai-api-ningbo easyai-api-hongkong \
-    easyai-worker-ningbo easyai-worker-hongkong \
+    easyai-api-ningbo \
+    easyai-worker-ningbo \
    easyai-capacity-controller \
-    easyai-web-ningbo easyai-web-hongkong; do
+    easyai-web-ningbo; do
    "${kubectl[@]}" annotate deployment "$deployment" -n "$NAMESPACE" \
      "easyai.io/release=$source_sha" \
      "easyai.io/config-hash=$(capacity_config_hash)" \
@@ -936,11 +915,9 @@ activate_manifest() {
  fi

 	if { [[ $api_changed != true ]] ||
-       { rollout_worker_site hongkong "$api_image" &&
-         rollout_worker_site ningbo "$api_image" &&
+       { rollout_worker_site ningbo "$api_image" &&
         { [[ $action == rollback ]] ||
           rollout_capacity_controller "$api_image"; }; }; } &&
-     rollout_site hongkong "$api_image" "$web_image" "$api_changed" "$web_changed" &&
     rollout_site ningbo "$api_image" "$web_image" "$api_changed" "$web_changed" &&
     wait_for_url 'public readiness' "$PUBLIC_BASE_URL/api/v1/readyz" '"ok":true'; then
    :
@@ -953,6 +930,7 @@ activate_manifest() {
 		return 1
 	fi

+	retire_hongkong_site
 	record_capacity_config
 	record_current "$manifest" "$action" "$started_at"
  rm -f -- "$deployment_snapshot" "$current_file"
@@ -972,10 +950,8 @@ case ${1:-} in
    node "$manifest_tool" validate "$2" >/dev/null
    api_image=$(manifest_get "$2" images.api)
    web_image=$(manifest_get "$2" images.web)
-    for site in ningbo hongkong; do
-      [[ $("${kubectl[@]}" get deployment "easyai-worker-$site" -n "$NAMESPACE" \
-        -o jsonpath='{.spec.template.spec.containers[?(@.name=="worker")].image}') == "$api_image" ]]
-    done
+    [[ $("${kubectl[@]}" get deployment easyai-worker-ningbo -n "$NAMESPACE" \
+      -o jsonpath='{.spec.template.spec.containers[?(@.name=="worker")].image}') == "$api_image" ]]
    [[ $("${kubectl[@]}" get deployment easyai-capacity-controller -n "$NAMESPACE" \
      -o jsonpath='{.spec.template.spec.containers[?(@.name=="capacity-controller")].image}') == "$api_image" ]]
    [[ $("${kubectl[@]}" get deployment easyai-api-ningbo -n "$NAMESPACE" \
@@ -6,8 +6,7 @@ DESIRED_STATE_DIR=/usr/local/share/easyai-ai-gateway-release/production

 # Worker 容量只需修改本发布配置并执行获授权的 `capacity`，不再修改 Go 代码。
 # 副本数属于 Deployment 配置，不能由容器内部环境变量改变；发布工具读取这两个变量并写入 replicas。
-AI_GATEWAY_WORKER_REPLICAS_NINGBO=0
-AI_GATEWAY_WORKER_REPLICAS_HONGKONG=2
+AI_GATEWAY_WORKER_REPLICAS_NINGBO=1
 AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT=24
 AI_GATEWAY_ASYNC_ADMISSION_MICROBATCH_SIZE=8
 AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT=48
@@ -24,10 +23,8 @@ AI_GATEWAY_DATABASE_MAX_CONN_IDLE_SECONDS=300
 AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY=24
 AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY=24
 AI_GATEWAY_WORKER_AUTOSCALING_ENABLED=false
-AI_GATEWAY_WORKER_MIN_REPLICAS_NINGBO=0
-AI_GATEWAY_WORKER_MIN_REPLICAS_HONGKONG=1
-AI_GATEWAY_WORKER_MAX_REPLICAS_NINGBO=0
-AI_GATEWAY_WORKER_MAX_REPLICAS_HONGKONG=2
+AI_GATEWAY_WORKER_MIN_REPLICAS_NINGBO=1
+AI_GATEWAY_WORKER_MAX_REPLICAS_NINGBO=1
 AI_GATEWAY_WORKER_TARGET_OUTSTANDING_PER_REPLICA=48
 AI_GATEWAY_WORKER_SCALE_UP_WINDOW_SECONDS=20
 AI_GATEWAY_WORKER_SCALE_DOWN_STABILIZATION_SECONDS=600
@@ -36,5 +33,5 @@ AI_GATEWAY_NODE_MEMORY_TARGET_PERCENT=75
 AI_GATEWAY_NODE_MEMORY_HARD_PERCENT=85
 AI_GATEWAY_NODE_CPU_TARGET_PERCENT=70
 AI_GATEWAY_POSTGRES_CONNECTION_BUDGET=150
-# 2 个 API 池（2×32）加 2 个 capacity-controller 池（2×4）。
-AI_GATEWAY_POSTGRES_NON_WORKER_CONNECTION_BUDGET=72
+# 1 个 API 池（32）加 1 个 capacity-controller 池（4）。
+AI_GATEWAY_POSTGRES_NON_WORKER_CONNECTION_BUDGET=36
@@ -46,12 +46,9 @@ data:
  AI_GATEWAY_ASYNC_WORKER_GLOBAL_HARD_LIMIT: "48"
  AI_GATEWAY_ASYNC_WORKER_REFRESH_INTERVAL_SECONDS: "5"
  AI_GATEWAY_WORKER_AUTOSCALING_ENABLED: "false"
-  AI_GATEWAY_WORKER_REPLICAS_NINGBO: "0"
-  AI_GATEWAY_WORKER_REPLICAS_HONGKONG: "2"
-  AI_GATEWAY_WORKER_MIN_REPLICAS_NINGBO: "0"
-  AI_GATEWAY_WORKER_MIN_REPLICAS_HONGKONG: "1"
-  AI_GATEWAY_WORKER_MAX_REPLICAS_NINGBO: "0"
-  AI_GATEWAY_WORKER_MAX_REPLICAS_HONGKONG: "2"
+  AI_GATEWAY_WORKER_REPLICAS_NINGBO: "1"
+  AI_GATEWAY_WORKER_MIN_REPLICAS_NINGBO: "1"
+  AI_GATEWAY_WORKER_MAX_REPLICAS_NINGBO: "1"
  AI_GATEWAY_WORKER_TARGET_OUTSTANDING_PER_REPLICA: "48"
  AI_GATEWAY_WORKER_SCALE_UP_WINDOW_SECONDS: "20"
  AI_GATEWAY_WORKER_SCALE_DOWN_STABILIZATION_SECONDS: "600"
@@ -158,7 +158,7 @@ metadata:
    app.kubernetes.io/component: capacity-controller
    app.kubernetes.io/part-of: easyai-ai-gateway
 spec:
-  replicas: 2
+  replicas: 1
  strategy:
    type: RollingUpdate
    rollingUpdate:
@@ -176,22 +176,9 @@ spec:
    spec:
      serviceAccountName: easyai-capacity-controller
      terminationGracePeriodSeconds: 30
-      affinity:
-        podAntiAffinity:
-          requiredDuringSchedulingIgnoredDuringExecution:
-            - topologyKey: easyai.io/site
-              labelSelector:
-                matchLabels:
-                  app.kubernetes.io/name: easyai-capacity-controller
      nodeSelector:
+        easyai.io/site: ningbo
        easyai.io/workload: "true"
-      topologySpreadConstraints:
-        - maxSkew: 1
-          topologyKey: easyai.io/site
-          whenUnsatisfiable: DoNotSchedule
-          labelSelector:
-            matchLabels:
-              app.kubernetes.io/name: easyai-capacity-controller
      securityContext:
        runAsNonRoot: true
        runAsUser: 10001
@@ -225,7 +212,7 @@ spec:
            - name: AI_GATEWAY_WORKER_DATABASE_MAX_CONNS
              value: "32"
            - name: AI_GATEWAY_POSTGRES_NON_WORKER_CONNECTION_BUDGET
-              value: "72"
+              value: "36"
            - name: POD_NAMESPACE
              valueFrom:
                fieldRef:
@@ -438,7 +425,7 @@ metadata:
    app.kubernetes.io/component: worker
    app.kubernetes.io/part-of: easyai-ai-gateway
 spec:
-  replicas: 0
+  replicas: 1
  strategy:
    type: RollingUpdate
    rollingUpdate:
@@ -8,6 +8,52 @@ resources:
  - application.yaml
  - database.yaml
  - network-policy.yaml
+patches:
+  - target:
+      kind: Deployment
+      name: easyai-api-hongkong
+    patch: |-
+      $patch: delete
+      apiVersion: apps/v1
+      kind: Deployment
+      metadata:
+        name: easyai-api-hongkong
+  - target:
+      kind: Deployment
+      name: easyai-worker-hongkong
+    patch: |-
+      $patch: delete
+      apiVersion: apps/v1
+      kind: Deployment
+      metadata:
+        name: easyai-worker-hongkong
+  - target:
+      kind: Deployment
+      name: easyai-web-hongkong
+    patch: |-
+      $patch: delete
+      apiVersion: apps/v1
+      kind: Deployment
+      metadata:
+        name: easyai-web-hongkong
+  - target:
+      kind: Service
+      name: api-hongkong-edge
+    patch: |-
+      $patch: delete
+      apiVersion: v1
+      kind: Service
+      metadata:
+        name: api-hongkong-edge
+  - target:
+      kind: Service
+      name: web-hongkong-edge
+    patch: |-
+      $patch: delete
+      apiVersion: v1
+      kind: Service
+      metadata:
+        name: web-hongkong-edge
 images:
  # Deployment tooling must replace both sentinel digests with release-manifest
  # digests before server-side apply.
@@ -48,11 +48,11 @@ metadata:
 rules:
  - apiGroups: ["apps"]
    resources: ["deployments"]
-    resourceNames: ["easyai-worker-ningbo", "easyai-worker-hongkong"]
+    resourceNames: ["easyai-worker-ningbo"]
    verbs: ["get"]
  - apiGroups: ["apps"]
    resources: ["deployments/scale"]
-    resourceNames: ["easyai-worker-ningbo", "easyai-worker-hongkong"]
+    resourceNames: ["easyai-worker-ningbo"]
    verbs: ["get", "update", "patch"]
  - apiGroups: [""]
    resources: ["pods"]