NODEJS-681: ControlConnection Concurrent Read and Write on .host and .connection

lib/control-connection.js

@@ -87,6 +87,7 @@ class ControlConnection extends events.EventEmitter {
     this._addressTranslator = this.options.policies.addressResolution;
     this._reconnectionPolicy = this.options.policies.reconnection;
     this._reconnectionSchedule = this._reconnectionPolicy.newSchedule();
+    this._refreshInProgress = false;
     this._isShuttingDown = false;
 
     // Reference to the encoder of the last valid connection
@@ -216,7 +217,6 @@ class ControlConnection extends events.EventEmitter {
 
   _setHealthListeners(host, connection) {
     const self = this;
-    let wasRefreshCalled = 0;
 
     function removeListeners() {
       host.removeListener('down', downOrIgnoredHandler);
@@ -225,11 +225,6 @@ class ControlConnection extends events.EventEmitter {
     }
 
     function startReconnecting(hostDown) {
-      if (wasRefreshCalled++ !== 0) {
-        // Prevent multiple calls to reconnect
-        return;
-      }
-
       removeListeners();
 
       if (self._isShuttingDown) {
@@ -446,13 +441,31 @@ class ControlConnection extends events.EventEmitter {
   }
 
   /**
-   * Acquires a new connection and refreshes topology and keyspace metadata.
+   * Acquires a new connection and refreshes topology and keyspace metadata, with protection against concurrent refreshes.
    * <p>When it fails obtaining a connection and there aren't any more hosts, it schedules reconnection.</p>
    * <p>When it fails obtaining the metadata, it marks connection and/or host unusable and retries using the same
    * iterator from query plan / host list</p>
    * @param {Iterator<Host>} [hostIterator]
    */
   async _refresh(hostIterator) {
+    if (this._refreshInProgress) {
+      return;
+    }
+    this._refreshInProgress = true;
+
+    try {
+      return await this._unsafeDoRefresh(hostIterator);
+    } finally {
+      this._refreshInProgress = false;
+    }
+  }
+
+  /**
+   * The actual implementation of the refresh logic, without protection against concurrent executions.
+   * <p>Should only be used via _refresh.</p>
+   * @param {Iterator<Host>} [hostIterator]
+   */
+  async _unsafeDoRefresh(hostIterator) {
     if (this._isShuttingDown) {
       this.log('info', 'The ControlConnection will not be refreshed as the Client is being shutdown');
       return;
@@ -499,7 +512,7 @@ class ControlConnection extends events.EventEmitter {
       }
 
       // Retry the whole thing with the same query plan
-      return await this._refresh(hostIterator);
+      return await this._unsafeDoRefresh(hostIterator);
     }
 
     this._reconnectionSchedule = this._reconnectionPolicy.newSchedule();

lib/promise-utils.js

@@ -149,7 +149,7 @@ function times(count, limit, fn) {
  * @returns {undefined}
  */
 function toBackground(promise) {
-  promise.catch(() => {});
+  promise?.catch(() => {});
 }
 
 /**

test/integration/short/control-connection-tests.js

@@ -153,6 +153,25 @@ describe('ControlConnection', function () {
       assert.strictEqual(cc.hosts.length, 1);
     });
 
+    it('should not break when refreshing concurrently', async () => {
+      const cc = newInstance();
+      cc.options.policies.loadBalancing = new policies.loadBalancing.RoundRobinPolicy();
+      disposeAfter(cc);
+
+      await cc.init();
+      await new Promise(r => cc.options.policies.loadBalancing.init(null, cc.hosts, r));
+
+      const refreshPromises = [];
+      // randomly emit cc._refresh 100 times
+      for (let i = 0; i < 100; i++) {
+        refreshPromises.push(cc._refresh());
+        await helper.delayAsync(~~(Math.random() * 100));
+      }
+      await Promise.all(refreshPromises);
+      assert.ok(cc.host);
+      assert.ok(cc.connection);
+    });
+
     it('should reconnect when host used goes down', async () => {
       const options = clientOptions.extend(
         utils.extend({ pooling: helper.getPoolingOptions(1, 1, 500) }, helper.baseOptions));