@testdriverai/runner 7.11.78-canary → 7.11.79-canary

package/index.js

@@ -333,9 +333,23 @@ class PresenceRunner {
   }
 
   startHeartbeat() {
-    // Update presence every 30 seconds with lastSeen timestamp
+    // Update presence every ~30 seconds with lastSeen timestamp
     // This allows detection of zombie runners (present but not responding)
-    this.heartbeatInterval = setInterval(async () => {
+    //
+    // All runners in a fleet tend to boot around the same time (autoscaling
+    // group launches, deploys), which would otherwise make every runner's
+    // heartbeat fire in lockstep on the shared per-team "runners" presence
+    // channel. That thundering-herd burst can push the channel over Ably's
+    // per-channel publish rate limit as the fleet grows. Use a
+    // self-rescheduling setTimeout with random jitter (instead of a fixed
+    // setInterval) so heartbeats spread out over time rather than
+    // clustering at the same instant.
+    const HEARTBEAT_BASE_MS = 30000;
+    const HEARTBEAT_JITTER_MS = 5000; // +/- up to 5s
+    // Integer in the inclusive range [-HEARTBEAT_JITTER_MS, HEARTBEAT_JITTER_MS]
+    const nextDelay = () => HEARTBEAT_BASE_MS + Math.floor(Math.random() * (2 * HEARTBEAT_JITTER_MS + 1)) - HEARTBEAT_JITTER_MS;
+
+    const beat = async () => {
       if (this.shuttingDown) return;
       try {
         // Free space rides along on the heartbeat so every runner reports its
@@ -355,12 +369,26 @@ class PresenceRunner {
       } catch (err) {
         log(`Warning: Heartbeat update failed: ${err.message}`);
       }
-    }, 30000);
+      // Reschedule regardless of success/failure (so a transient publish
+      // error doesn't permanently kill the heartbeat), but only if the
+      // runner isn't shutting down. There is no `await` between this check
+      // and the `setTimeout` call, so `stopHeartbeat()` cannot race with it
+      // and clear a handle we're about to overwrite — Node's single-threaded
+      // event loop guarantees this block runs to completion uninterrupted.
+      if (!this.shuttingDown) {
+        this.heartbeatInterval = setTimeout(beat, nextDelay());
+      }
+    };
+
+    this.heartbeatInterval = setTimeout(beat, nextDelay());
   }
 
   stopHeartbeat() {
     if (this.heartbeatInterval) {
-      clearInterval(this.heartbeatInterval);
+      // heartbeatInterval now holds a setTimeout handle (self-rescheduling
+      // heartbeat with jitter — see startHeartbeat), but clearTimeout also
+      // safely clears a setInterval handle, so this stays defensive either way.
+      clearTimeout(this.heartbeatInterval);
       this.heartbeatInterval = null;
     }
   }

package/lib/ably-service.js

@@ -311,6 +311,7 @@ class AblyService extends EventEmitter {
     this._execOutputDraining = false;
 
     this._automation.on('exec.output', ({ requestId, chunk }) => {
+      if (this._closed) return; // don't enqueue once we're tearing down
       this._execOutputQueue.push({ requestId, chunk });
       this._drainExecOutputQueue();
     });
@@ -593,6 +594,15 @@ class AblyService extends EventEmitter {
 
     try {
       while (this._execOutputQueue.length > 0) {
+        // Stop draining once close() has started tearing down the channel —
+        // a publish landing after detach() can make Ably silently
+        // re-attach the channel, doubling channel-open/close lifecycle
+        // events for what should be one clean session teardown.
+        if (this._closed) {
+          this._execOutputQueue.length = 0;
+          break;
+        }
+
         // Rate limit: wait if needed
         const now = Date.now();
         const elapsed = now - this._execOutputLastTime;
@@ -600,6 +610,11 @@ class AblyService extends EventEmitter {
           await new Promise((resolve) => setTimeout(resolve, this._execOutputMinIntervalMs - elapsed));
         }
 
+        if (this._closed) {
+          this._execOutputQueue.length = 0;
+          break;
+        }
+
         // Coalesce all queued chunks for the same requestId
         const batch = {};
         while (this._execOutputQueue.length > 0) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@testdriverai/runner",
-  "version": "7.11.78-canary",
+  "version": "7.11.79-canary",
   "description": "TestDriver Runner - Ably-based remote automation agent with Node.js automation",
   "main": "index.js",
   "bin": {