fix(CQ-ORCH-10): Make BullMQ job retention configurable via env vars

Replace hardcoded BullMQ job retention values (completed: 100 jobs / 1h,
failed: 1000 jobs / 24h) with configurable env vars to prevent memory
growth under load. Adds QUEUE_COMPLETED_RETENTION_COUNT,
QUEUE_COMPLETED_RETENTION_AGE_S, QUEUE_FAILED_RETENTION_COUNT, and
QUEUE_FAILED_RETENTION_AGE_S to orchestrator config. Defaults preserve
existing behavior.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

apps/orchestrator/src/queue/queue.service.ts

@@ -45,17 +45,35 @@ export class QueueService implements OnModuleInit, OnModuleDestroy {
       password: this.configService.get<string>("orchestrator.valkey.password"),
     };
 
+    // Read retention config
+    const completedRetentionAge = this.configService.get<number>(
+      "orchestrator.queue.completedRetentionAgeSeconds",
+      3600
+    );
+    const completedRetentionCount = this.configService.get<number>(
+      "orchestrator.queue.completedRetentionCount",
+      100
+    );
+    const failedRetentionAge = this.configService.get<number>(
+      "orchestrator.queue.failedRetentionAgeSeconds",
+      86400
+    );
+    const failedRetentionCount = this.configService.get<number>(
+      "orchestrator.queue.failedRetentionCount",
+      1000
+    );
+
     // Create queue
     this.queue = new Queue<QueuedTask>(this.queueName, {
       connection,
       defaultJobOptions: {
         removeOnComplete: {
-          age: 3600, // Keep completed jobs for 1 hour
-          count: 100, // Keep last 100 completed jobs
+          age: completedRetentionAge,
+          count: completedRetentionCount,
         },
         removeOnFail: {
-          age: 86400, // Keep failed jobs for 24 hours
-          count: 1000, // Keep last 1000 failed jobs
+          age: failedRetentionAge,
+          count: failedRetentionCount,
         },
       },
     });