1

README.md

@@ -72,7 +72,7 @@ export NAMESPACE=prod
 
 改 JSON → push `master` / `test` → workflow 按分支写入对应 Consul 并 `rollout restart`。
 
-手动：
+手动（在 runner 机上；跨集群会自动 port-forward）：
 
 ```bash
 # prod

scripts/export-from-consul.sh

@@ -35,10 +35,34 @@ json_ok_str() {
     jq empty <<<"$raw"
   fi
 }
-if [[ -z "${CONSUL_HTTP_ADDR:-}" ]]; then
+# Consul：跨集群时 ClusterIP 不可达 → port-forward
-  CIP="$(kubectl -n "$NS" get svc consul -o jsonpath='{.spec.clusterIP}')"
+CONSUL_PF_PID=""
-  CONSUL_HTTP_ADDR="http://${CIP}:8500"
+cleanup_pf() {
+  if [[ -n "${CONSUL_PF_PID:-}" ]]; then
+    kill "$CONSUL_PF_PID" 2>/dev/null || true
+    wait "$CONSUL_PF_PID" 2>/dev/null || true
+    CONSUL_PF_PID=""
   fi
+}
+trap cleanup_pf EXIT
+
+if [[ -z "${CONSUL_HTTP_ADDR:-}" ]]; then
+  LOCAL_PORT=18510
+  echo "==> port-forward svc/consul → 127.0.0.1:${LOCAL_PORT} (ns=${NS})"
+  kubectl -n "$NS" port-forward svc/consul "${LOCAL_PORT}:8500" >/tmp/consul-pf-export-${NS}.log 2>&1 &
+  CONSUL_PF_PID=$!
+  ok=0
+  for _ in $(seq 1 40); do
+    if curl -sf --connect-timeout 1 "http://127.0.0.1:${LOCAL_PORT}/v1/status/leader" >/dev/null 2>&1; then
+      ok=1
+      break
+    fi
+    sleep 0.5
+  done
+  [[ "$ok" == "1" ]] || { echo "FAIL: Consul port-forward 未就绪"; cat /tmp/consul-pf-export-${NS}.log 2>/dev/null || true; exit 1; }
+  CONSUL_HTTP_ADDR="http://127.0.0.1:${LOCAL_PORT}"
+fi
+echo "==> Consul ${CONSUL_HTTP_ADDR}"
 
 if [[ -z "${CONSUL_HTTP_TOKEN:-}" ]]; then
   raw="$(kubectl -n "$NS" get secret consul-acl -o jsonpath='{.data.CONSUL_HTTP_TOKEN}' 2>/dev/null || true)"

scripts/sync-to-consul.sh

@@ -147,9 +147,54 @@ load_secret JWT_SECRET_KEY prod-credentials JWT_SECRET_KEY
 : "${ELASTIC_PASSWORD:?缺少 ELASTIC_PASSWORD}"
 : "${JWT_SECRET_KEY:?缺少 JWT_SECRET_KEY（请创建 secret/jwt-secret 或写入 prod-credentials）}"
 
+# Consul：runner 可能在另一集群节点上，ClusterIP 不可达 → 一律 port-forward 到本机
+CONSUL_PF_PID=""
+cleanup() {
+  if [[ -n "${CONSUL_PF_PID:-}" ]]; then
+    kill "$CONSUL_PF_PID" 2>/dev/null || true
+    wait "$CONSUL_PF_PID" 2>/dev/null || true
+    CONSUL_PF_PID=""
+  fi
+  [[ -n "${TMP:-}" && -d "${TMP:-}" ]] && rm -rf "$TMP"
+}
+trap cleanup EXIT
+
 if [[ -z "$CONSUL_HTTP_ADDR" ]]; then
-  CIP="$(kubectl -n "$NS" get svc consul -o jsonpath='{.spec.clusterIP}')"
+  if ! kubectl -n "$NS" get svc consul >/dev/null 2>&1; then
-  CONSUL_HTTP_ADDR="http://${CIP}:8500"
+    echo "FAIL: 无 svc/${NS}/consul（KUBECONFIG=${KUBECONFIG}）"
+    exit 1
+  fi
+  # 避开常用端口；失败则换一个
+  LOCAL_PORT=""
+  for p in 18500 18501 18502 18503 18504; do
+    if ! ss -ltn 2>/dev/null | grep -q ":${p} " && ! netstat -ltn 2>/dev/null | grep -q ":${p} "; then
+      LOCAL_PORT=$p
+      break
+    fi
+  done
+  LOCAL_PORT="${LOCAL_PORT:-18500}"
+  echo "==> port-forward svc/consul → 127.0.0.1:${LOCAL_PORT} (ns=${NS})"
+  kubectl -n "$NS" port-forward svc/consul "${LOCAL_PORT}:8500" >/tmp/consul-pf-${NS}.log 2>&1 &
+  CONSUL_PF_PID=$!
+  ok=0
+  for _ in $(seq 1 40); do
+    if ! kill -0 "$CONSUL_PF_PID" 2>/dev/null; then
+      echo "FAIL: port-forward 已退出，见 /tmp/consul-pf-${NS}.log"
+      cat /tmp/consul-pf-${NS}.log 2>/dev/null || true
+      exit 1
+    fi
+    if curl -sf --connect-timeout 1 "http://127.0.0.1:${LOCAL_PORT}/v1/status/leader" >/dev/null 2>&1; then
+      ok=1
+      break
+    fi
+    sleep 0.5
+  done
+  if [[ "$ok" != "1" ]]; then
+    echo "FAIL: Consul port-forward 未就绪，见 /tmp/consul-pf-${NS}.log"
+    cat /tmp/consul-pf-${NS}.log 2>/dev/null || true
+    exit 1
+  fi
+  CONSUL_HTTP_ADDR="http://127.0.0.1:${LOCAL_PORT}"
 fi
 echo "==> Consul ${CONSUL_HTTP_ADDR}"
 
@@ -179,7 +224,6 @@ fi
 [[ ${#TARGETS[@]} -gt 0 ]] || { echo "FAIL: 无同步目标"; exit 1; }
 
 TMP="$(mktemp -d)"
-trap 'rm -rf "$TMP"' EXIT
 
 put_kv() {
   local key="$1" file="$2"