git.stella-ops.org/src/Scheduler/__Libraries/StellaOps.Scheduler.Queue

Scheduler Queue — Sprint 16 Coordination Notes

Queue work now has concrete contracts from StellaOps.Scheduler.Models:

• Planner inputs reference Schedule and ImpactSet samples (samples/api/scheduler/).
• Runner segment payloads should carry runId, scheduleId?, tenantId, and the impacted digest list (mirrors Run.Deltas).
• Notify fanout relies on the DeltaSummary shape already emitted by the model layer.

Action items for SCHED-QUEUE-16-401..403

1. Reference StellaOps.Scheduler.Models so adapters can serialise Run/DeltaSummary without bespoke DTOs.
2. Use the canonical serializer for queue messages to keep ordering consistent with API payloads.
3. Coverage: add fixture-driven tests that enqueue the sample payloads, then dequeue and re-serialise to verify byte-for-byte stability.
4. Expose queue depth/lease metrics with the identifiers provided by the models (Run.Id, Schedule.Id).

JetStream failover notes

• SchedulerQueueOptions.Kind = "nats" will spin up NatsSchedulerPlannerQueue / NatsSchedulerRunnerQueue instances backed by JetStream.
• SchedulerQueueHealthCheck pings both planner and runner transports; register via AddSchedulerQueueHealthCheck() to surface in /healthz.
• Planner defaults:

  scheduler:
    queue:
      kind: nats
      deadLetterEnabled: true
      nats:
        url: "nats://nats:4222"
        planner:
          stream: SCHEDULER_PLANNER
          subject: scheduler.planner
          durableConsumer: scheduler-planners
          deadLetterStream: SCHEDULER_PLANNER_DEAD
          deadLetterSubject: scheduler.planner.dead
        runner:
          stream: SCHEDULER_RUNNER
          subject: scheduler.runner
          durableConsumer: scheduler-runners
      redis:
        deadLetterStream: scheduler:planner:dead
        idempotencyKeyPrefix: scheduler:planner:idemp:
  

• Retry / dead-letter semantics mirror the Redis adapter: attempts beyond MaxDeliveryAttempts are shipped to the configured dead-letter stream with headers describing runId, scheduleId, and failure reasons. Set deadLetterEnabled: false to drop exhausted messages instead.
• Depth metrics surface through scheduler_queue_depth{transport,queue}; both transports publish lightweight counters to drive alerting dashboards.

These notes unblock the queue guild now that SCHED-MODELS-16-102 is complete.