feat(deploy): 合并宁波单节点生产拓扑

apps/api/internal/capacitycontroller/controller.go

@@ -140,12 +140,15 @@ func (controller *Controller) reconcile(ctx context.Context) error {
 	sites := make([]SiteResources, 0, 2)
 	kubernetesStates := make(map[string]KubernetesSiteState, 2)
 	for _, site := range []string{"ningbo", "hongkong"} {
+		minReplicas, maxReplicas := controller.siteReplicaBounds(site)
+		if maxReplicas == 0 {
+			continue
+		}
 		state, stateErr := controller.kubernetes.SiteState(ctx, site)
 		if stateErr != nil {
 			return stateErr
 		}
 		kubernetesStates[site] = state
-		minReplicas, maxReplicas := controller.siteReplicaBounds(site)
 		sites = append(sites, SiteResources{
 			Site: site, CurrentReplicas: state.CurrentReplicas,
 			MinReplicas: minReplicas, MaxReplicas: maxReplicas,

apps/api/internal/config/config.go

@@ -198,11 +198,11 @@ func Load() Config {
 		) != "",
 		WorkerAutoscalingEnabled:          env("AI_GATEWAY_WORKER_AUTOSCALING_ENABLED", "false") == "true",
 		WorkerReplicasNingbo:              envOptionalIntValidated("AI_GATEWAY_WORKER_REPLICAS_NINGBO", 1),
-		WorkerReplicasHongkong:            envOptionalIntValidated("AI_GATEWAY_WORKER_REPLICAS_HONGKONG", 1),
+		WorkerReplicasHongkong:            envOptionalIntValidated("AI_GATEWAY_WORKER_REPLICAS_HONGKONG", 0),
 		WorkerMinReplicasNingbo:           envOptionalIntValidated("AI_GATEWAY_WORKER_MIN_REPLICAS_NINGBO", 1),
-		WorkerMinReplicasHongkong:         envOptionalIntValidated("AI_GATEWAY_WORKER_MIN_REPLICAS_HONGKONG", 1),
+		WorkerMinReplicasHongkong:         envOptionalIntValidated("AI_GATEWAY_WORKER_MIN_REPLICAS_HONGKONG", 0),
 		WorkerMaxReplicasNingbo:           envOptionalIntValidated("AI_GATEWAY_WORKER_MAX_REPLICAS_NINGBO", 1),
-		WorkerMaxReplicasHongkong:         envOptionalIntValidated("AI_GATEWAY_WORKER_MAX_REPLICAS_HONGKONG", 1),
+		WorkerMaxReplicasHongkong:         envOptionalIntValidated("AI_GATEWAY_WORKER_MAX_REPLICAS_HONGKONG", 0),
 		WorkerTargetOutstandingPerReplica: envOptionalIntValidated("AI_GATEWAY_WORKER_TARGET_OUTSTANDING_PER_REPLICA", 0),
 		WorkerScaleUpWindowSeconds:        envIntValidated("AI_GATEWAY_WORKER_SCALE_UP_WINDOW_SECONDS", 20),
 		WorkerScaleDownStabilizationSeconds: envIntValidated(

apps/api/internal/config/config_test.go

@@ -217,6 +217,26 @@ func TestWorkerAutoscalingConfiguration(t *testing.T) {
 	}
 }
 
+func TestWorkerTopologyDefaultsToNingboSingleNode(t *testing.T) {
+	for _, name := range []string{
+		"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",
+	} {
+		t.Setenv(name, "")
+	}
+	cfg := Load()
+	if cfg.WorkerReplicasNingbo != 1 || cfg.WorkerMinReplicasNingbo != 1 || cfg.WorkerMaxReplicasNingbo != 1 {
+		t.Fatalf("Ningbo Worker defaults=%d/%d/%d, want 1/1/1", cfg.WorkerReplicasNingbo, cfg.WorkerMinReplicasNingbo, cfg.WorkerMaxReplicasNingbo)
+	}
+	if cfg.WorkerReplicasHongkong != 0 || cfg.WorkerMinReplicasHongkong != 0 || cfg.WorkerMaxReplicasHongkong != 0 {
+		t.Fatalf("Hong Kong Worker defaults=%d/%d/%d, want disabled", cfg.WorkerReplicasHongkong, cfg.WorkerMinReplicasHongkong, cfg.WorkerMaxReplicasHongkong)
+	}
+}
+
 func TestLoadPrefersWorkerGlobalHardLimitAlias(t *testing.T) {
 	t.Setenv("AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT", "48")
 	t.Setenv("AI_GATEWAY_ASYNC_WORKER_GLOBAL_HARD_LIMIT", "96")

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

@@ -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" \

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

@@ -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

deploy/kubernetes/production/application-config.yaml

@@ -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"

deploy/kubernetes/production/application.yaml

@@ -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:

deploy/kubernetes/production/kustomization.yaml

@@ -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.

deploy/kubernetes/production/service-account-rbac.yaml

@@ -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"]

scripts/cluster/common.sh

@@ -24,16 +24,14 @@ load_cluster_env() {
   cluster_ssh_key=${AI_GATEWAY_CLUSTER_SSH_KEY:-$cluster_ssh_key}
 
   : "${AI_GATEWAY_DEPLOY_HOST:?}"
-  : "${AI_GATEWAY_DEPLOY_HOST_2:?}"
-  : "${AI_GATEWAY_DEPLOY_HOST_3:?}"
   : "${ALI_KEY:?}"
   : "${ALI_SECRET:?}"
   : "${ALI_REGION:?}"
   : "${ALI_BUCKET:?}"
 
   CLUSTER_NINGBO_HOST=${AI_GATEWAY_NINGBO_HOST:-$AI_GATEWAY_DEPLOY_HOST}
-  CLUSTER_LOS_ANGELES_HOST=${AI_GATEWAY_LOS_ANGELES_HOST:-$AI_GATEWAY_DEPLOY_HOST_2}
-  CLUSTER_HONGKONG_HOST=${AI_GATEWAY_HONGKONG_HOST:-$AI_GATEWAY_DEPLOY_HOST_3}
+  CLUSTER_LOS_ANGELES_HOST=${AI_GATEWAY_LOS_ANGELES_HOST:-${AI_GATEWAY_DEPLOY_HOST_2:-}}
+  CLUSTER_HONGKONG_HOST=${AI_GATEWAY_HONGKONG_HOST:-${AI_GATEWAY_DEPLOY_HOST_3:-}}
   CLUSTER_WORKER_NODE_4_HOST=${AI_GATEWAY_WORKER_NODE_4_HOST:-}
   CLUSTER_WORKER_NODE_4_NAME=${AI_GATEWAY_WORKER_NODE_4_NAME:-}
   CLUSTER_WORKER_NODE_4_SITE=${AI_GATEWAY_WORKER_NODE_4_SITE:-}
@@ -46,6 +44,7 @@ load_cluster_env() {
   export CLUSTER_WORKER_NODE_4_ENABLED
 
   for host in "$CLUSTER_NINGBO_HOST" "$CLUSTER_HONGKONG_HOST" "$CLUSTER_LOS_ANGELES_HOST"; do
+    [[ -n $host ]] || continue
     [[ $host =~ ^root@[A-Za-z0-9.-]+$ ]] || {
       echo "cluster hosts must use root@host syntax" >&2
       return 1

scripts/cluster/run-production-acceptance.sh

@@ -705,17 +705,22 @@ admin_request() {
       -o 'jsonpath={.spec.clusterIP}')
     service_port=$(remote_kubectl get service api-ningbo-edge -n "$namespace" \
       -o 'jsonpath={.spec.ports[0].port}')
-    fallback_service_ip=$(remote_kubectl get service api-hongkong-edge -n "$namespace" \
-      -o 'jsonpath={.spec.clusterIP}')
-    fallback_service_port=$(remote_kubectl get service api-hongkong-edge -n "$namespace" \
-      -o 'jsonpath={.spec.ports[0].port}')
-    [[ $service_ip =~ ^[0-9.]+$ && $service_port =~ ^[0-9]+$ &&
-       $fallback_service_ip =~ ^[0-9.]+$ && $fallback_service_port =~ ^[0-9]+$ ]] || {
+    [[ $service_ip =~ ^[0-9.]+$ && $service_port =~ ^[0-9]+$ ]] || {
       echo 'acceptance admin Service has an invalid cluster address' >&2
       return 1
     }
     acceptance_admin_base=http://$service_ip:$service_port
-    acceptance_admin_fallback_base=http://$fallback_service_ip:$fallback_service_port
+    if [[ ${single_node_acceptance:-false} != true ]]; then
+      fallback_service_ip=$(remote_kubectl get service api-hongkong-edge -n "$namespace" \
+        -o 'jsonpath={.spec.clusterIP}')
+      fallback_service_port=$(remote_kubectl get service api-hongkong-edge -n "$namespace" \
+        -o 'jsonpath={.spec.ports[0].port}')
+      [[ $fallback_service_ip =~ ^[0-9.]+$ && $fallback_service_port =~ ^[0-9]+$ ]] || {
+        echo 'acceptance fallback admin Service has an invalid cluster address' >&2
+        return 1
+      }
+      acceptance_admin_fallback_base=http://$fallback_service_ip:$fallback_service_port
+    fi
   fi
   token_encoded=$(printf '%s' "$AI_GATEWAY_ACCEPTANCE_ADMIN_TOKEN" |
     openssl base64 -A)
@@ -748,6 +753,10 @@ fi
       echo "acceptance control API transport failed without safe retry: method=$method path=$path" >&2
       return 1
     fi
+    if [[ ${single_node_acceptance:-false} == true ]]; then
+      echo "acceptance control API transport failed on Ningbo: method=$method path=$path" >&2
+      return 1
+    fi
     printf -v remote_command 'bash -c %q -- %q %q %q %q %q %q' \
       "$remote_script" "$acceptance_admin_fallback_base" "$method" "$path" \
       "$token_encoded" "$body_encoded" "$has_body"
@@ -794,15 +803,13 @@ verify_release_cas() {
     echo "production release CAS mismatch: expected=$release_sha current=$current_sha" >&2
     return 1
   }
-  for deployment in easyai-api-ningbo easyai-api-hongkong easyai-worker-ningbo easyai-worker-hongkong; do
+  for deployment in easyai-api-ningbo easyai-worker-ningbo; do
     [[ $(remote_kubectl get deployment "$deployment" -n "$namespace" \
       -o 'jsonpath={.spec.template.spec.containers[0].image}') == "$api_image" ]]
   done
   [[ $verify_capacity == true ]] || return 0
   [[ $(remote_kubectl get deployment easyai-worker-ningbo -n "$namespace" \
     -o 'jsonpath={.spec.replicas}') == "$AI_GATEWAY_ACCEPTANCE_BASE_REPLICAS_NINGBO" ]]
-  [[ $(remote_kubectl get deployment easyai-worker-hongkong -n "$namespace" \
-    -o 'jsonpath={.spec.replicas}') == "$AI_GATEWAY_ACCEPTANCE_BASE_REPLICAS_HONGKONG" ]]
   if [[ $CLUSTER_WORKER_NODE_4_ENABLED == true ]]; then
     local expected_replicas eligible_nodes eligible_count pod_report
     local running_count invalid_count node4_count used_nodes
@@ -943,9 +950,7 @@ snapshot_pre_acceptance_capacity() {
   local output=$report_root/pre-acceptance-capacity.json
   remote_kubectl get \
     deployment/easyai-api-ningbo \
-    deployment/easyai-api-hongkong \
     deployment/easyai-worker-ningbo \
-    deployment/easyai-worker-hongkong \
     deployment/easyai-capacity-controller \
     configmap/easyai-ai-gateway-config \
     configmap/easyai-gateway-release \
@@ -3466,7 +3471,6 @@ apply_single_node_capacity() {
     }}')
   remote_kubectl label node easyai-ningbo easyai.io/worker=true --overwrite >/dev/null
   remote_kubectl scale deployment/easyai-capacity-controller -n "$namespace" --replicas=0 >/dev/null
-  remote_kubectl scale deployment/easyai-worker-hongkong -n "$namespace" --replicas=0 >/dev/null
   remote_kubectl patch configmap easyai-ai-gateway-config -n "$namespace" \
     --type=merge -p "$config_patch" >/dev/null
   remote_kubectl set env deployment/easyai-worker-ningbo -n "$namespace" \

tests/release/cluster-release-helper-test.sh

@@ -41,7 +41,6 @@ kubectl=("$tmp/kubectl")
 NAMESPACE=easyai
 ACTIVE_RELEASE_SHA=aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
 AI_GATEWAY_WORKER_REPLICAS_NINGBO=1
-AI_GATEWAY_WORKER_REPLICAS_HONGKONG=1
 AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT=24
 AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT=48
 AI_GATEWAY_ASYNC_WORKER_GLOBAL_HARD_LIMIT=48
@@ -55,9 +54,7 @@ AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY=24
 AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY=24
 AI_GATEWAY_WORKER_AUTOSCALING_ENABLED=false
 AI_GATEWAY_WORKER_MIN_REPLICAS_NINGBO=1
-AI_GATEWAY_WORKER_MIN_REPLICAS_HONGKONG=1
 AI_GATEWAY_WORKER_MAX_REPLICAS_NINGBO=1
-AI_GATEWAY_WORKER_MAX_REPLICAS_HONGKONG=1
 AI_GATEWAY_WORKER_TARGET_OUTSTANDING_PER_REPLICA=48
 AI_GATEWAY_WORKER_CPU_REQUEST_MILLICORES=500
 AI_GATEWAY_WORKER_MEMORY_REQUEST_MIB=512
@@ -66,7 +63,7 @@ grep -Fq "\"AI_GATEWAY_REVISION=\${ACTIVE_RELEASE_SHA:-}\"" "$helper"
 
 # activate_manifest invokes this function inside an && condition. A failed
 # rollout must still propagate instead of falling through to a stale Ready Pod.
-if rollout_worker_site hongkong 'registry.example.invalid/api@sha256:deadbeef' && true; then
+if rollout_worker_site ningbo 'registry.example.invalid/api@sha256:deadbeef' && true; then
   echo 'failed Worker rollout was reported as successful' >&2
   exit 1
 fi
@@ -80,16 +77,21 @@ deployment_strategy() {
   ' "$application"
 }
 
-for deployment in easyai-worker-ningbo easyai-worker-hongkong; do
-  strategy=$(deployment_strategy "$deployment")
-  grep -Fq '      maxSurge: 0' <<<"$strategy"
-  grep -Fq '      maxUnavailable: 1' <<<"$strategy"
-done
+strategy=$(deployment_strategy easyai-worker-ningbo)
+grep -Fq '      maxSurge: 0' <<<"$strategy"
+grep -Fq '      maxUnavailable: 1' <<<"$strategy"
 
-for deployment in easyai-api-ningbo easyai-api-hongkong; do
-  strategy=$(deployment_strategy "$deployment")
-  grep -Fq '      maxSurge: 1' <<<"$strategy"
-  grep -Fq '      maxUnavailable: 0' <<<"$strategy"
-done
+strategy=$(deployment_strategy easyai-api-ningbo)
+grep -Fq '      maxSurge: 1' <<<"$strategy"
+grep -Fq '      maxUnavailable: 0' <<<"$strategy"
+
+grep -Fq 'name: easyai-api-hongkong' "$root/deploy/kubernetes/production/kustomization.yaml"
+grep -Fq 'name: easyai-worker-hongkong' "$root/deploy/kubernetes/production/kustomization.yaml"
+grep -Fq 'name: easyai-web-hongkong' "$root/deploy/kubernetes/production/kustomization.yaml"
+if grep -Fq 'AI_GATEWAY_WORKER_REPLICAS_HONGKONG=' \
+  "$root/deploy/kubernetes/easyai-ai-gateway-cluster-release.conf.example"; then
+  echo 'Hong Kong Worker capacity remains in the single-node release environment' >&2
+  exit 1
+fi
 
 echo 'cluster_release_helper_tests=PASS'