@celerispay/hazelcast-client 3.12.7-5 → 3.12.7-6

package/lib/invocation/ClusterService.d.ts

@@ -98,9 +98,20 @@ export declare class ClusterService {
     private onHeartbeatStopped(connection);
     private triggerFailover();
     /**
-     * Handles single-node cluster reset - treats node restart as fresh cluster
+     * Handles single-node node loss. Tears down all state immediately and starts
+     * a TCP probe loop. Only once the port accepts a raw TCP connection do we
+     * attempt the full Hazelcast handshake — eliminating the race between
+     * destroyConnection() re-adding failedConnections and our attempts to clear them.
      */
     private handleSingleNodeClusterReset();
+    /**
+     * Probes the single-node address with a raw TCP socket every probeIntervalMs.
+     * When the port responds (connect event), destroys the probe socket and
+     * calls connectToCluster() to perform the full Hazelcast authentication.
+     * If connectToCluster() fails (server still warming up), retries the probe.
+     * Stops if the client is shut down.
+     */
+    private startSingleNodeProbeLoop(address);
     /**
      * Handles multi-node failover - preserves existing logic
      */

package/lib/invocation/ClusterService.js

@@ -15,6 +15,7 @@
  * limitations under the License.
  */
 Object.defineProperty(exports, "__esModule", { value: true });
+var net = require("net");
 var Promise = require("bluebird");
 var ClientAddMembershipListenerCodec_1 = require("../codec/ClientAddMembershipListenerCodec");
 var ClientInfo_1 = require("../ClientInfo");
@@ -264,48 +265,78 @@ var ClusterService = /** @class */ (function () {
         }
     };
     /**
-     * Handles single-node cluster reset - treats node restart as fresh cluster
+     * Handles single-node node loss. Tears down all state immediately and starts
+     * a TCP probe loop. Only once the port accepts a raw TCP connection do we
+     * attempt the full Hazelcast handshake — eliminating the race between
+     * destroyConnection() re-adding failedConnections and our attempts to clear them.
      */
     ClusterService.prototype.handleSingleNodeClusterReset = function () {
-        var _this = this;
-        this.logger.info('ClusterService', '🧹 SINGLE-NODE RESET: Clearing all credentials and state...');
-        // Clear all stored credentials - node restart means fresh cluster
-        this.client.getConnectionManager().clearAllCredentials();
-        // Reset client UUIDs - will be assigned fresh by server
+        this.logger.info('ClusterService', 'SINGLE-NODE: Node lost. Tearing down state and starting TCP probe loop.');
+        // Full state teardown
         this.uuid = null;
         this.ownerUuid = null;
-        // Log state before reset
-        this.logCurrentState();
-        // Force cleanup of all dead connections
+        this.members = [];
+        this.client.getConnectionManager().clearAllCredentials();
+        this.client.getConnectionManager().clearFailedConnections();
         this.client.getConnectionManager().forceCleanupDeadConnections();
-        // Clear partition information
         this.client.getPartitionService().clearPartitionTable();
-        // IMPORTANT: Unblock all down addresses before attempting reconnection.
-        // We must clear BOTH blocklists:
-        //   1. this.downAddresses  — ClusterService-level block (checked by tryConnectingToAddresses)
-        //   2. failedConnections   — ClientConnectionManager-level block (checked by getOrConnect)
-        // We delay by 200ms to let any in-flight async destroyConnection() calls finish,
-        // because destroyConnection() re-adds the address to failedConnections after we clear it.
-        // Clearing both right before connectToCluster() ensures neither blocklist interferes.
-        this.logger.info('ClusterService', '🔓 SINGLE-NODE RESET: Scheduling reconnection after brief stabilization delay...');
-        setTimeout(function () {
-            // Clear both blocklists right before attempting reconnection
-            _this.downAddresses.clear();
-            _this.client.getConnectionManager().clearFailedConnections();
-            _this.logger.info('ClusterService', '🔄 SINGLE-NODE RESET: Attempting direct reconnection...');
-            _this.connectToCluster()
-                .then(function () {
-                _this.logger.info('ClusterService', '✅ Single-node cluster reset completed successfully');
-                _this.logCurrentState();
-            })
-                .catch(function (error) {
-                _this.logger.error('ClusterService', 'Single-node cluster reset failed', error);
-                _this.logCurrentState();
-            })
-                .finally(function () {
+        this.downAddresses.clear();
+        // Start probing — reconnect only when the port actually accepts TCP
+        var address = this.knownAddresses[0];
+        this.startSingleNodeProbeLoop(address);
+    };
+    /**
+     * Probes the single-node address with a raw TCP socket every probeIntervalMs.
+     * When the port responds (connect event), destroys the probe socket and
+     * calls connectToCluster() to perform the full Hazelcast authentication.
+     * If connectToCluster() fails (server still warming up), retries the probe.
+     * Stops if the client is shut down.
+     */
+    ClusterService.prototype.startSingleNodeProbeLoop = function (address) {
+        var _this = this;
+        var probeIntervalMs = 3000;
+        var host = address.host;
+        var port = address.port;
+        this.logger.info('ClusterService', "SINGLE-NODE PROBE: Waiting for " + host + ":" + port + " to accept connections (probing every " + probeIntervalMs + "ms)...");
+        var attempt = function () {
+            if (!_this.client.getLifecycleService().isRunning()) {
+                _this.logger.info('ClusterService', 'SINGLE-NODE PROBE: Client is shutting down, stopping probe loop.');
                 _this.failoverInProgress = false;
+                return;
+            }
+            var socket = net.createConnection({ host: host, port: port });
+            socket.once('connect', function () {
+                socket.destroy();
+                _this.logger.info('ClusterService', "SINGLE-NODE PROBE: " + host + ":" + port + " is accepting connections. Reinitialising Hazelcast connection...");
+                // Clear both blocklists right before connecting — destroyConnection()
+                // may have re-added the address to failedConnections since the teardown above.
+                _this.downAddresses.clear();
+                _this.client.getConnectionManager().clearFailedConnections();
+                _this.connectToCluster()
+                    .then(function () {
+                    _this.logger.info('ClusterService', 'SINGLE-NODE: Reconnected successfully.');
+                    _this.logCurrentState();
+                    _this.failoverInProgress = false;
+                })
+                    .catch(function (err) {
+                    _this.logger.warn('ClusterService', "SINGLE-NODE: connectToCluster failed after probe success (" + err.message + "). Retrying probe...");
+                    // Server accepted TCP but Hazelcast not fully ready yet — probe again
+                    _this.downAddresses.clear();
+                    _this.client.getConnectionManager().clearFailedConnections();
+                    setTimeout(attempt, probeIntervalMs);
+                });
             });
-        }, 200);
+            socket.once('error', function () {
+                socket.destroy();
+                setTimeout(attempt, probeIntervalMs);
+            });
+            // Guard against hung sockets that produce neither connect nor error
+            socket.setTimeout(2000, function () {
+                socket.destroy();
+                setTimeout(attempt, probeIntervalMs);
+            });
+        };
+        attempt();
     };
     /**
      * Handles multi-node failover - preserves existing logic

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@celerispay/hazelcast-client",
-  "version": "3.12.7-5",
+  "version": "3.12.7-6",
   "description": "Hazelcast - open source In-Memory Data Grid - client for NodeJS with critical connection failover fixes",
   "main": "./lib/index.js",
   "scripts": {