feat(queue): 增加非文本模型分布式准入队列

使用 PostgreSQL 统一同步与异步非文本任务的并发准入、持久化等待和 Worker 容量分配，并将生产 API 与独立 Worker 角色拆分。

补充策略管理、共享契约、OpenAPI、Kubernetes 双节点 Worker 清单及跨节点验收脚本；未默认启用任何生产 queue_size 策略。

已在原基线完成 Go、前端、迁移、Shell、Kustomize 与长任务容量验收；合入最新主干后将重新执行发布门禁。

scripts/cluster/bootstrap-platform.sh

@@ -23,6 +23,11 @@ node "$cluster_root/scripts/release-manifest.mjs" validate "$release_manifest" >
 api_image=$(node "$cluster_root/scripts/release-manifest.mjs" get "$release_manifest" images.api)
 web_image=$(node "$cluster_root/scripts/release-manifest.mjs" get "$release_manifest" images.web)
 source_sha=$(node "$cluster_root/scripts/release-manifest.mjs" get "$release_manifest" sourceSha)
+activate_workers=${AI_GATEWAY_ACTIVATE_WORKERS:-true}
+[[ $activate_workers == true || $activate_workers == false ]] || {
+  echo 'AI_GATEWAY_ACTIVATE_WORKERS must be true or false' >&2
+  exit 64
+}
 [[ $api_image =~ ^registry\.cn-shanghai\.aliyuncs\.com/easyaigc/ai-gateway@sha256:[0-9a-f]{64}$ ]]
 [[ $web_image =~ ^registry\.cn-shanghai\.aliyuncs\.com/easyaigc/ai-gateway-web@sha256:[0-9a-f]{64}$ ]]
 
@@ -87,11 +92,28 @@ if [[ $mode == prepare ]]; then
   sed -i.bak 's/^  replicas: 1$/  replicas: 0/' "$working_directory/production/application.yaml"
   rm -f -- "$working_directory/production/application.yaml.bak"
 fi
-if [[ ${AI_GATEWAY_ACTIVATE_WORKERS:-true} == false ]]; then
-  sed -i.bak \
-    's/^  AI_GATEWAY_ASYNC_QUEUE_WORKER_ENABLED: "true"$/  AI_GATEWAY_ASYNC_QUEUE_WORKER_ENABLED: "false"/' \
-    "$working_directory/production/application-config.yaml"
-  rm -f -- "$working_directory/production/application-config.yaml.bak"
+if [[ $activate_workers == false ]]; then
+  awk '
+    /^---$/ {
+      document_kind = ""
+      worker_deployment = 0
+      print
+      next
+    }
+    /^kind: / {
+      document_kind = $2
+    }
+    document_kind == "Deployment" && /^  name: easyai-worker-/ {
+      worker_deployment = 1
+    }
+    worker_deployment && /^  replicas: 1$/ {
+      sub(/1$/, "0")
+    }
+    { print }
+  ' "$working_directory/production/application.yaml" \
+    >"$working_directory/production/application.yaml.next"
+  mv "$working_directory/production/application.yaml.next" \
+    "$working_directory/production/application.yaml"
 fi
 kubectl kustomize "$working_directory/production" >"$working_directory/easyai-production.yaml"
 if rg -q 'sha256:0{64}|image: (easyai-api|easyai-web)$' "$working_directory/easyai-production.yaml"; then
@@ -136,10 +158,12 @@ for file in "${platform_files[@]}"; do
   cluster_scp "$working_directory/$file" "$CLUSTER_NINGBO_HOST:/root/$file"
 done
 
-cluster_ssh "$CLUSTER_NINGBO_HOST" bash -s -- "$source_sha" "$mode" <<'REMOTE'
+cluster_ssh "$CLUSTER_NINGBO_HOST" bash -s -- \
+  "$source_sha" "$mode" "$activate_workers" <<'REMOTE'
 set -euo pipefail
 source_sha=$1
 mode=$2
+activate_workers=$3
 kubectl='k3s kubectl'
 runtime_environment=
 cleanup_bootstrap_files() {
@@ -234,6 +258,10 @@ $kubectl apply --server-side --dry-run=server --force-conflicts \
 $kubectl apply -f /root/easyai-production.yaml >/dev/null
 $kubectl annotate namespace easyai easyai.io/release="$source_sha" --overwrite >/dev/null
 if [[ $mode == activate ]]; then
+  if [[ $activate_workers == true ]]; then
+    $kubectl rollout status deployment/easyai-worker-hongkong -n easyai --timeout=300s
+    $kubectl rollout status deployment/easyai-worker-ningbo -n easyai --timeout=300s
+  fi
   $kubectl rollout status deployment/easyai-api-hongkong -n easyai --timeout=300s
   $kubectl rollout status deployment/easyai-web-hongkong -n easyai --timeout=300s
   $kubectl rollout status deployment/easyai-api-ningbo -n easyai --timeout=300s

scripts/cluster/migrate-production.sh

@@ -114,7 +114,9 @@ rollback_before_commit() {
   if [[ $traffic_committed == false ]]; then
     echo '[cutover] rolling back to the frozen Docker database' >&2
     k3s kubectl scale deployment -n easyai \
-      easyai-api-ningbo easyai-api-hongkong easyai-web-ningbo easyai-web-hongkong \
+      easyai-api-ningbo easyai-api-hongkong \
+      easyai-worker-ningbo easyai-worker-hongkong \
+      easyai-web-ningbo easyai-web-hongkong \
       --replicas=0 >/dev/null 2>&1 || true
     docker start easyai-ai-gateway-api-1 easyai-ai-gateway-web-1 >/dev/null 2>&1 || true
     if [[ -f $old_site_backup ]]; then
@@ -312,7 +314,9 @@ set -euo pipefail
 backup_directory=$(cat /root/.easyai-pre-cutover-backup)
 [[ $backup_directory == /var/backups/easyai-ai-gateway/pre-cutover-* ]]
 k3s kubectl scale deployment -n easyai \
-  easyai-api-ningbo easyai-api-hongkong easyai-web-ningbo easyai-web-hongkong \
+  easyai-api-ningbo easyai-api-hongkong \
+  easyai-worker-ningbo easyai-worker-hongkong \
+  easyai-web-ningbo easyai-web-hongkong \
   --replicas=0 >/dev/null 2>&1 || true
 docker start easyai-ai-gateway-api-1 easyai-ai-gateway-web-1 >/dev/null
 install -m 0644 "$backup_directory/nginx-site.conf" \
@@ -325,7 +329,7 @@ REMOTE
 }
 trap rollback_local_before_commit ERR
 
-echo '[cutover] starting both application sites with River workers frozen'
+echo '[cutover] starting both application sites with dedicated workers frozen'
 AI_GATEWAY_ACTIVATE_WORKERS=false \
   "$script_dir/bootstrap-platform.sh" activate "$release_manifest"
 
@@ -352,12 +356,10 @@ echo '[cutover] traffic committed; enabling Hong Kong worker, then Ningbo worker
 cluster_ssh "$CLUSTER_NINGBO_HOST" 'bash -s' <<'REMOTE'
 set -euo pipefail
 kubectl='k3s kubectl'
-$kubectl set env deployment/easyai-api-hongkong -n easyai \
-  AI_GATEWAY_ASYNC_QUEUE_WORKER_ENABLED=true
-$kubectl rollout status deployment/easyai-api-hongkong -n easyai --timeout=300s
-$kubectl set env deployment/easyai-api-ningbo -n easyai \
-  AI_GATEWAY_ASYNC_QUEUE_WORKER_ENABLED=true
-$kubectl rollout status deployment/easyai-api-ningbo -n easyai --timeout=300s
+$kubectl scale deployment/easyai-worker-hongkong -n easyai --replicas=1
+$kubectl rollout status deployment/easyai-worker-hongkong -n easyai --timeout=300s
+$kubectl scale deployment/easyai-worker-ningbo -n easyai --replicas=1
+$kubectl rollout status deployment/easyai-worker-ningbo -n easyai --timeout=300s
 backup_resource=$(cat <<'EOF' | $kubectl create -f - -o name
 apiVersion: postgresql.cnpg.io/v1
 kind: Backup

scripts/cluster/run-cross-node-file-e2e.sh

@@ -13,43 +13,41 @@ require_commands base64 jq node
 restore_workers() {
   cluster_ssh "$CLUSTER_NINGBO_HOST" 'bash -s' <<'REMOTE' || true
 set -euo pipefail
-k3s kubectl set env deployment/easyai-api-ningbo -n easyai \
-  AI_GATEWAY_ASYNC_QUEUE_WORKER_ENABLED=true >/dev/null
-k3s kubectl set env deployment/easyai-api-hongkong -n easyai \
-  AI_GATEWAY_ASYNC_QUEUE_WORKER_ENABLED=true >/dev/null
-k3s kubectl rollout status deployment/easyai-api-ningbo -n easyai --timeout=300s
-k3s kubectl rollout status deployment/easyai-api-hongkong -n easyai --timeout=300s
+k3s kubectl scale deployment/easyai-worker-ningbo -n easyai --replicas=1 >/dev/null
+k3s kubectl scale deployment/easyai-worker-hongkong -n easyai --replicas=1 >/dev/null
+k3s kubectl rollout status deployment/easyai-worker-ningbo -n easyai --timeout=300s
+k3s kubectl rollout status deployment/easyai-worker-hongkong -n easyai --timeout=300s
 REMOTE
 }
 trap restore_workers EXIT HUP INT TERM
 
 cluster_ssh "$CLUSTER_NINGBO_HOST" 'bash -s' <<'REMOTE'
 set -euo pipefail
-k3s kubectl set env deployment/easyai-api-hongkong -n easyai \
-  AI_GATEWAY_ASYNC_QUEUE_WORKER_ENABLED=true >/dev/null
-k3s kubectl rollout status deployment/easyai-api-hongkong -n easyai --timeout=300s
-k3s kubectl set env deployment/easyai-api-ningbo -n easyai \
-  AI_GATEWAY_ASYNC_QUEUE_WORKER_ENABLED=false >/dev/null
-k3s kubectl rollout status deployment/easyai-api-ningbo -n easyai --timeout=300s
+k3s kubectl scale deployment/easyai-worker-hongkong -n easyai --replicas=1 >/dev/null
+k3s kubectl rollout status deployment/easyai-worker-hongkong -n easyai --timeout=300s
+k3s kubectl scale deployment/easyai-worker-ningbo -n easyai --replicas=0 >/dev/null
 for _ in $(seq 1 60); do
   pods=$(k3s kubectl get pods -n easyai \
-    -l app.kubernetes.io/name=easyai-api,easyai.io/site=ningbo \
+    -l app.kubernetes.io/name=easyai-worker,easyai.io/site=hongkong \
     -o jsonpath='{range .items[*]}{.metadata.name}{"\n"}{end}')
   pod_count=$(wc -w <<<"$pods" | tr -d ' ')
   if [[ $pod_count == 1 ]]; then
     pod=$pods
-    worker_enabled=$(k3s kubectl get pod "$pod" -n easyai \
-      -o jsonpath='{.spec.containers[0].env[?(@.name=="AI_GATEWAY_ASYNC_QUEUE_WORKER_ENABLED")].value}')
+    process_role=$(k3s kubectl get pod "$pod" -n easyai \
+      -o jsonpath='{.spec.containers[0].env[?(@.name=="AI_GATEWAY_PROCESS_ROLE")].value}')
     pod_ready=$(k3s kubectl get pod "$pod" -n easyai \
       -o jsonpath='{.status.containerStatuses[0].ready}')
-    if [[ $worker_enabled == false && $pod_ready == true ]]; then
+    if [[ $process_role == worker && $pod_ready == true ]]; then
       break
     fi
   fi
   sleep 1
 done
-[[ ${pod_count:-0} == 1 && ${worker_enabled:-} == false && ${pod_ready:-} == true ]]
-k3s kubectl logs "$pod" -n easyai | grep -q 'asynchronous queue worker disabled for this process'
+[[ ${pod_count:-0} == 1 && ${process_role:-} == worker && ${pod_ready:-} == true ]]
+[[ $(k3s kubectl get deployment easyai-api-ningbo -n easyai \
+  -o jsonpath='{.spec.template.spec.containers[0].env[?(@.name=="AI_GATEWAY_PROCESS_ROLE")].value}') == api ]]
+[[ $(k3s kubectl get deployment easyai-worker-ningbo -n easyai \
+  -o jsonpath='{.status.readyReplicas}') == 0 ]]
 curl -fsS http://10.77.0.1:31088/api/v1/readyz | grep -q '"ok":true'
 curl -fsS http://10.77.0.2:31089/api/v1/readyz | grep -q '"ok":true'
 REMOTE
@@ -98,7 +96,7 @@ query() {
 }
 attempted_by=$(query -c \
   "SELECT attempted_by::text FROM river_job WHERE id=(SELECT river_job_id FROM gateway_tasks WHERE id='$task_id'::uuid);")
-[[ $attempted_by == *easyai-api-hongkong* ]] || {
+[[ $attempted_by == *easyai-worker-hongkong* ]] || {
   echo "River attempted_by does not identify Hong Kong: $attempted_by" >&2
   exit 1
 }

scripts/cluster/verify-cluster.sh

@@ -95,6 +95,18 @@ kubectl='k3s kubectl'
   --field-selector=status.phase=Running --no-headers | wc -l) -eq 2 ]]
 [[ $($kubectl get pods -n easyai -l app.kubernetes.io/name=easyai-web \
   --field-selector=status.phase=Running --no-headers | wc -l) -eq 2 ]]
+[[ $($kubectl get pods -n easyai -l app.kubernetes.io/name=easyai-worker \
+  --field-selector=status.phase=Running --no-headers | wc -l) -eq 2 ]]
+for site in ningbo hongkong; do
+  [[ $($kubectl get deployment "easyai-api-$site" -n easyai \
+    -o jsonpath='{.status.readyReplicas}') -eq 1 ]]
+  [[ $($kubectl get deployment "easyai-worker-$site" -n easyai \
+    -o jsonpath='{.status.readyReplicas}') -eq 1 ]]
+  [[ $($kubectl get deployment "easyai-api-$site" -n easyai \
+    -o jsonpath='{.spec.template.spec.containers[0].env[?(@.name=="AI_GATEWAY_PROCESS_ROLE")].value}') == api ]]
+  [[ $($kubectl get deployment "easyai-worker-$site" -n easyai \
+    -o jsonpath='{.spec.template.spec.containers[0].env[?(@.name=="AI_GATEWAY_PROCESS_ROLE")].value}') == worker ]]
+done
 images=$($kubectl get deployments -n easyai -o json |
   jq -r '.items[] | select(.metadata.name | startswith("easyai-")) |
     .spec.template.spec.containers[].image')
@@ -112,7 +124,7 @@ printf 'applications=ready\n'
 REMOTE
 )
   [[ $application_state == applications=ready ]]
-  echo 'application_acceptance=PASS api=2 web=2 digests=pinned'
+  echo 'application_acceptance=PASS api=2 worker=2 web=2 roles=isolated digests=pinned'
 
   for edge_ip in "${CLUSTER_NINGBO_HOST#*@}" "${CLUSTER_HONGKONG_HOST#*@}"; do
     health_body=$(curl -fsS --max-time 10 --resolve "ai.51easyai.com:443:$edge_ip" \