@dudousxd/nestjs-durable 0.43.0 → 0.43.1

package/dist/index.js

@@ -1169,6 +1169,15 @@ var RetentionPoller = class {
   onModuleDestroy() {
     if (this.timer) clearInterval(this.timer);
   }
+  /**
+  * Prune one round.
+  *
+  * Never rejects, for the same reason {@link TimerPoller.poll} does not: it is awaited inside
+  * `onApplicationBootstrap` (so a rejection propagates into the host's `bootstrap()` and kills the
+  * process) and fired as `void` on the interval (so a rejection is an unhandled one, also fatal).
+  * Pruning history is the least urgent thing this package does — a `pruneTerminalRuns` that fails
+  * on a locked table must cost this sweep and nothing more.
+  */
   async sweep() {
     if (this.sweeping) return;
     const retention = this.options.retention;
@@ -1185,6 +1194,8 @@ var RetentionPoller = class {
           if (deleted < batchSize) break;
         }
       }
+    } catch (error) {
+      console.error("[nestjs-durable] retention sweep failed; retrying next interval:", error);
     } finally {
       this.sweeping = false;
     }
@@ -1579,7 +1590,11 @@ var TimerPoller = class {
   }
   async onApplicationBootstrap() {
     if (!isDrivingOperator(this.options)) return;
-    this.unsubscribeEnqueued = this.engine.onEnqueued((runId) => void this.engine.runOne(runId));
+    this.unsubscribeEnqueued = this.engine.onEnqueued((runId) => {
+      void this.engine.runOne(runId).catch((error) => {
+        console.error(`[nestjs-durable] could not run enqueued run ${runId}:`, error);
+      });
+    });
     await this.poll();
     const intervalMs = this.options.timerPollMs ?? 1e3;
     if (intervalMs > 0) {
@@ -1591,22 +1606,45 @@ var TimerPoller = class {
     if (this.timer) clearInterval(this.timer);
     this.unsubscribeEnqueued?.();
   }
+  /**
+  * One sweep of the engine's recovery work.
+  *
+  * **This method never rejects, and that is a hard requirement rather than defensiveness.** It has
+  * exactly two callers and a rejection out of either one kills the process: on boot it is awaited
+  * inside `onApplicationBootstrap`, so the rejection propagates through `NestApplication.init()`
+  * into the host's `bootstrap()`; on the interval it is a fire-and-forget `void`, so it surfaces as
+  * an unhandled rejection. Both are fatal, and both re-occur on the next boot with the same state,
+  * so a single unrecoverable run crash-loops every worker in the deployment — taking down every
+  * other workflow with it. A poll that goes wrong must cost this tick, not the pod.
+  *
+  * Each sweep is isolated separately because they are independent: a store that fails
+  * `recoverIncomplete` says nothing about whether durable timers should still fire, and skipping
+  * the rest of the tick would silently widen one subsystem's outage into all of them.
+  */
   async poll() {
     if (this.polling) return;
     this.polling = true;
     try {
-      await this.engine.runPending();
-      await this.engine.recoverIncomplete();
-      await this.engine.resumeDueTimers();
-      await this.engine.sweepTimeouts();
+      await this.sweep("runPending", () => this.engine.runPending());
+      await this.sweep("recoverIncomplete", () => this.engine.recoverIncomplete());
+      await this.sweep("resumeDueTimers", () => this.engine.resumeDueTimers());
+      await this.sweep("sweepTimeouts", () => this.engine.sweepTimeouts());
       const schedules = this.options.schedules;
       if (schedules && schedules.length > 0) {
-        await runSchedules(this.engine, schedules, Date.now());
+        await this.sweep("schedules", () => runSchedules(this.engine, schedules, Date.now()));
       }
     } finally {
       this.polling = false;
     }
   }
+  /** Run one sweep, absorbing and reporting its failure — see {@link poll}. */
+  async sweep(name, run) {
+    try {
+      await run();
+    } catch (error) {
+      console.error(`[nestjs-durable] timer poll: ${name} failed; continuing:`, error);
+    }
+  }
 };
 TimerPoller = _ts_decorate9([
   Injectable9(),