feat(cli): auto-cancel dev runs on CLI exit via detached watchdog

ericallam · ericallam · commit 562e88ab3196 · 2026-03-07T22:03:51.000Z
When the dev CLI exits (e.g. ctrl+c via pnpm), runs that were mid-execution
previously stayed stuck in EXECUTING status for up to 5 minutes until the
heartbeat timeout fired. Now they are cancelled within seconds.

The dev CLI spawns a lightweight detached watchdog process at startup. The
watchdog monitors the CLI process ID and, when it detects the CLI has exited,
calls a new POST /engine/v1/dev/disconnect endpoint to cancel all in-flight
runs immediately (skipping PENDING_CANCEL since the worker is known to be dead).

Watchdog design:
- Fully detached (detached: true, stdio: ignore, unref()) so it survives
  even when pnpm sends SIGKILL to the process tree
- Active run IDs maintained via atomic file write (.trigger/active-runs.json)
- Single-instance guarantee via PID file (.trigger/watchdog.pid)
- Safety timeout: exits after 24 hours to prevent zombie processes
- On clean shutdown, the watchdog is killed (no disconnect needed)

Disconnect endpoint:
- Rate-limited: 5 calls/min per environment
- Capped at 500 runs per call
- Small counts (&lt;= 25): cancelled inline with pMap concurrency 10
- Large counts: delegated to the bulk action system
- Uses finalizeRun: true to skip PENDING_CANCEL and go straight to FINISHED

Run engine change:
- cancelRun() now respects finalizeRun when the run is in EXECUTING status,
  skipping the PENDING_CANCEL waiting state and going directly to FINISHED
diff --git a/.changeset/strange-moles-provide.md b/.changeset/strange-moles-provide.md
@@ -0,0 +1,6 @@
+---
+"trigger.dev": patch
+"@trigger.dev/core": patch
+---
+
+Auto-cancel in-flight dev runs when the CLI exits, using a detached watchdog process that survives pnpm SIGKILL
diff --git a/.server-changes/dev-cli-disconnect-md b/.server-changes/dev-cli-disconnect-md
@@ -0,0 +1,6 @@
+---
+area: webapp
+type: feature
+---
+
+Added `/engine/v1/dev/disconnect` endpoint to auto-cancel runs when the CLI disconnects. Maximum of 500 runs can be cancelled. Uses the bulk action system when there are more than 25 runs to cancel.
diff --git a/apps/webapp/app/routes/engine.v1.dev.disconnect.ts b/apps/webapp/app/routes/engine.v1.dev.disconnect.ts
@@ -0,0 +1,169 @@
+import { json } from "@remix-run/server-runtime";
+import { Ratelimit } from "@upstash/ratelimit";
+import { tryCatch } from "@trigger.dev/core";
+import { DevDisconnectRequestBody } from "@trigger.dev/core/v3";
+import { BulkActionId, RunId } from "@trigger.dev/core/v3/isomorphic";
+import { BulkActionNotificationType, BulkActionType } from "@trigger.dev/database";
+import { prisma } from "~/db.server";
+import { logger } from "~/services/logger.server";
+import { RateLimiter } from "~/services/rateLimiter.server";
+import { createActionApiRoute } from "~/services/routeBuilders/apiBuilder.server";
+import { CancelTaskRunService } from "~/v3/services/cancelTaskRun.server";
+import { commonWorker } from "~/v3/commonWorker.server";
+import pMap from "p-map";
+
+const CANCEL_REASON = "Dev session ended (CLI exited)";
+
+// Below this threshold, cancel runs inline with pMap.
+// Above it, create a bulk action and process asynchronously.
+const BULK_ACTION_THRESHOLD = 25;
+
+// Maximum number of runs that can be cancelled in a single disconnect call.
+const MAX_RUNS = 500;
+
+// Rate limit: 5 calls per minute per environment
+const disconnectRateLimiter = new RateLimiter({
+  keyPrefix: "dev-disconnect",
+  limiter: Ratelimit.fixedWindow(5, "1 m"),
+  logFailure: true,
+});
+
+const { action } = createActionApiRoute(
+  {
+    body: DevDisconnectRequestBody,
+    maxContentLength: 1024 * 256, // 256KB
+    method: "POST",
+  },
+  async ({ authentication, body }) => {
+    const environmentId = authentication.environment.id;
+
+    // Rate limit per environment
+    const rateLimitResult = await disconnectRateLimiter.limit(environmentId);
+    if (!rateLimitResult.success) {
+      return json(
+        { error: "Rate limit exceeded", retryAfter: Math.ceil((rateLimitResult.reset - Date.now()) / 1000) },
+        { status: 429 }
+      );
+    }
+
+    // Cap the number of runs
+    const runFriendlyIds = body.runFriendlyIds.slice(0, MAX_RUNS);
+
+    if (runFriendlyIds.length === 0) {
+      return json({ cancelled: 0 }, { status: 200 });
+    }
+
+    logger.info("Dev disconnect: cancelling runs", {
+      environmentId,
+      runCount: runFriendlyIds.length,
+      capped: body.runFriendlyIds.length > MAX_RUNS,
+    });
+
+    // For small numbers of runs, cancel inline
+    if (runFriendlyIds.length <= BULK_ACTION_THRESHOLD) {
+      const cancelled = await cancelRunsInline(runFriendlyIds, environmentId);
+      return json({ cancelled }, { status: 200 });
+    }
+
+    // For large numbers, create a bulk action to process asynchronously
+    const bulkActionId = await createBulkCancelAction(
+      runFriendlyIds,
+      authentication.environment.project.id,
+      environmentId
+    );
+
+    logger.info("Dev disconnect: created bulk action for large run set", {
+      environmentId,
+      bulkActionId,
+      runCount: runFriendlyIds.length,
+    });
+
+    return json({ cancelled: 0, bulkActionId }, { status: 200 });
+  }
+);
+
+async function cancelRunsInline(
+  runFriendlyIds: string[],
+  environmentId: string
+): Promise<number> {
+  const runIds = runFriendlyIds.map((fid) => RunId.toId(fid));
+
+  const runs = await prisma.taskRun.findMany({
+    where: {
+      id: { in: runIds },
+      runtimeEnvironmentId: environmentId,
+    },
+    select: {
+      id: true,
+      engine: true,
+      friendlyId: true,
+      status: true,
+      createdAt: true,
+      completedAt: true,
+      taskEventStore: true,
+    },
+  });
+
+  let cancelled = 0;
+  const cancelService = new CancelTaskRunService(prisma);
+
+  await pMap(
+    runs,
+    async (run) => {
+      const [error, result] = await tryCatch(
+        cancelService.call(run, { reason: CANCEL_REASON, finalizeRun: true })
+      );
+
+      if (error) {
+        logger.error("Dev disconnect: failed to cancel run", {
+          runId: run.id,
+          error,
+        });
+      } else if (result && !result.alreadyFinished) {
+        cancelled++;
+      }
+    },
+    { concurrency: 10 }
+  );
+
+  logger.info("Dev disconnect: completed inline cancellation", {
+    environmentId,
+    cancelled,
+    total: runFriendlyIds.length,
+  });
+
+  return cancelled;
+}
+
+async function createBulkCancelAction(
+  runFriendlyIds: string[],
+  projectId: string,
+  environmentId: string
+): Promise<string> {
+  const { id, friendlyId } = BulkActionId.generate();
+
+  await prisma.bulkActionGroup.create({
+    data: {
+      id,
+      friendlyId,
+      projectId,
+      environmentId,
+      name: "Dev session disconnect",
+      type: BulkActionType.CANCEL,
+      params: { runId: runFriendlyIds },
+      queryName: "bulk_action_v1",
+      totalCount: runFriendlyIds.length,
+      completionNotification: BulkActionNotificationType.NONE,
+    },
+  });
+
+  await commonWorker.enqueue({
+    id: `processBulkAction-${id}`,
+    job: "processBulkAction",
+    payload: { bulkActionId: id },
+  });
+
+  return friendlyId;
+}
+
+export { action };
diff --git a/apps/webapp/app/v3/services/cancelTaskRun.server.ts b/apps/webapp/app/v3/services/cancelTaskRun.server.ts
@@ -8,6 +8,8 @@ export type CancelTaskRunServiceOptions = {
   cancelAttempts?: boolean;
   cancelledAt?: Date;
   bulkActionId?: string;
+  /** Skip PENDING_CANCEL and finalize immediately (use when the worker is known to be dead). */
+  finalizeRun?: boolean;
 };
 
 type CancelTaskRunServiceResult = {
@@ -57,6 +59,7 @@ export class CancelTaskRunService extends BaseService {
       runId: taskRun.id,
       completedAt: options?.cancelledAt,
       reason: options?.reason,
+      finalizeRun: options?.finalizeRun,
       bulkActionId: options?.bulkActionId,
       tx: this._prisma,
     });
diff --git a/internal-packages/run-engine/src/engine/systems/runAttemptSystem.ts b/internal-packages/run-engine/src/engine/systems/runAttemptSystem.ts
@@ -1436,35 +1436,39 @@ export class RunAttemptSystem {
         });
 
         //if executing, we need to message the worker to cancel the run and put it into `PENDING_CANCEL` status
+        //unless finalizeRun is true (worker is known to be dead), in which case skip straight to FINISHED
         if (
           isExecuting(latestSnapshot.executionStatus) ||
           isPendingExecuting(latestSnapshot.executionStatus)
         ) {
-          const newSnapshot = await this.executionSnapshotSystem.createExecutionSnapshot(prisma, {
-            run,
-            snapshot: {
-              executionStatus: "PENDING_CANCEL",
-              description: "Run was cancelled",
-            },
-            previousSnapshotId: latestSnapshot.id,
-            environmentId: latestSnapshot.environmentId,
-            environmentType: latestSnapshot.environmentType,
-            projectId: latestSnapshot.projectId,
-            organizationId: latestSnapshot.organizationId,
-            workerId,
-            runnerId,
-          });
+          if (!finalizeRun) {
+            const newSnapshot = await this.executionSnapshotSystem.createExecutionSnapshot(prisma, {
+              run,
+              snapshot: {
+                executionStatus: "PENDING_CANCEL",
+                description: "Run was cancelled",
+              },
+              previousSnapshotId: latestSnapshot.id,
+              environmentId: latestSnapshot.environmentId,
+              environmentType: latestSnapshot.environmentType,
+              projectId: latestSnapshot.projectId,
+              organizationId: latestSnapshot.organizationId,
+              workerId,
+              runnerId,
+            });
 
-          //the worker needs to be notified so it can kill the run and complete the attempt
-          await sendNotificationToWorker({
-            runId,
-            snapshot: newSnapshot,
-            eventBus: this.$.eventBus,
-          });
-          return {
-            alreadyFinished: false,
-            ...executionResultFromSnapshot(newSnapshot),
-          };
+            //the worker needs to be notified so it can kill the run and complete the attempt
+            await sendNotificationToWorker({
+              runId,
+              snapshot: newSnapshot,
+              eventBus: this.$.eventBus,
+            });
+            return {
+              alreadyFinished: false,
+              ...executionResultFromSnapshot(newSnapshot),
+            };
+          }
+          // finalizeRun is true — fall through to finish the run immediately
         }
 
         //not executing, so we will actually finish the run
diff --git a/packages/cli-v3/src/apiClient.ts b/packages/cli-v3/src/apiClient.ts
@@ -7,6 +7,8 @@ import {
   DevConfigResponseBody,
   DevDequeueRequestBody,
   DevDequeueResponseBody,
+  DevDisconnectRequestBody,
+  DevDisconnectResponseBody,
   EnvironmentVariableResponseBody,
   FailDeploymentRequestBody,
   FailDeploymentResponseBody,
@@ -557,6 +559,7 @@ export class CliApiClient {
       heartbeatRun: this.devHeartbeatRun.bind(this),
       startRunAttempt: this.devStartRunAttempt.bind(this),
       completeRunAttempt: this.devCompleteRunAttempt.bind(this),
+      disconnect: this.devDisconnect.bind(this),
       setEngineURL: this.setEngineURL.bind(this),
     } as const;
   }
@@ -681,6 +684,23 @@ export class CliApiClient {
     return eventSource;
   }
 
+  private async devDisconnect(
+    body: DevDisconnectRequestBody
+  ): Promise<ApiResult<DevDisconnectResponseBody>> {
+    if (!this.accessToken) {
+      throw new Error("devDisconnect: No access token");
+    }
+
+    return wrapZodFetch(DevDisconnectResponseBody, `${this.engineURL}/engine/v1/dev/disconnect`, {
+      method: "POST",
+      headers: {
+        Authorization: `Bearer ${this.accessToken}`,
+        Accept: "application/json",
+      },
+      body: JSON.stringify(body),
+    });
+  }
+
   private async devDequeue(
     body: DevDequeueRequestBody
   ): Promise<ApiResult<DevDequeueResponseBody>> {
diff --git a/packages/cli-v3/src/dev/devSupervisor.ts b/packages/cli-v3/src/dev/devSupervisor.ts
diff --git a/packages/cli-v3/src/dev/devWatchdog.ts b/packages/cli-v3/src/dev/devWatchdog.ts
diff --git a/packages/core/src/v3/schemas/api.ts b/packages/core/src/v3/schemas/api.ts
