fix: stabilize runner websocket across suspend/restore

Author: pnispel

.changeset/slow-lions-tickle.md

@@ -0,0 +1,5 @@
+---
+"@basicblock/trigger-cli": patch
+---
+
+Reset supervisor socket around checkpoint suspend/restore and add configurable websocket ping interval/timeout in supervisor to reduce disconnect issues during runner execution.

apps/supervisor/src/env.ts

@@ -24,6 +24,8 @@ const Env = z.object({
   TRIGGER_WORKLOAD_API_HOST_INTERNAL: z.string().default("0.0.0.0"),
   TRIGGER_WORKLOAD_API_PORT_INTERNAL: z.coerce.number().default(8020), // This is the port the workload API listens on
   TRIGGER_WORKLOAD_API_PORT_EXTERNAL: z.coerce.number().default(8020), // This is the exposed port passed to the run controller
+  TRIGGER_WORKLOAD_API_WS_PING_INTERVAL_MS: z.coerce.number().int().positive().default(25000),
+  TRIGGER_WORKLOAD_API_WS_PING_TIMEOUT_MS: z.coerce.number().int().positive().default(120000),
 
   // Runner settings
   RUNNER_HEARTBEAT_INTERVAL_SECONDS: z.coerce.number().optional(),

apps/supervisor/src/workloadServer/index.ts

@@ -398,7 +398,10 @@ export class WorkloadServer extends EventEmitter<WorkloadServerEvents> {
   }
 
   private createWebsocketServer() {
-    const io = new Server(this.httpServer.server);
+    const io = new Server(this.httpServer.server, {
+      pingInterval: env.TRIGGER_WORKLOAD_API_WS_PING_INTERVAL_MS,
+      pingTimeout: env.TRIGGER_WORKLOAD_API_WS_PING_TIMEOUT_MS,
+    });
 
     const websocketServer: Namespace<
       WorkloadClientToServerEvents,

packages/cli-v3/src/entryPoints/managed/execution.ts

@@ -865,6 +865,15 @@ export class RunExecution {
     // Short delay to give websocket time to reconnect
     await sleep(100);
 
+    // Reset the supervisor socket after checkpoint restore.
+    // Socket transports don't reliably survive process snapshot/restore boundaries.
+    try {
+      this.supervisorSocket.disconnect();
+    } catch (_) {
+      // noop
+    }
+    this.supervisorSocket.connect();
+
     // Process any env overrides
     await this.processEnvOverrides("restore");
 
@@ -1204,6 +1213,10 @@ export class RunExecution {
     }
 
     this.sendDebugLog("suspending, any day now 🚬", { suspendableSnapshot });
+
+    // Disconnect before snapshotting so the restored process does a fresh socket handshake.
+    // This avoids carrying a stale websocket transport through checkpoint/restore.
+    this.supervisorSocket.disconnect();
   }
 
   /**