Module 1 · Compute and execution · Lesson 4
Queues, schedules and background jobs
How to buffer bursts, fan out events, keep one customer from starving another, and schedule work for days or weeks, with at-least-once queues, per-tenant concurrency, Little's law sizing and the right scheduler for each wait.
By the end of this lesson you can
- Explain at-least-once delivery, leases (visibility timeouts), acknowledgement, retries and idempotent publishing
- Choose between a queue, a durable workflow and a scheduler for a given piece of agent work
- Size consumer concurrency with Little's law and set leases and retry caps from processing times
- Apply per-tenant concurrency keys so one tenant's backlog can't starve another
- Implement a queue consumer in Python with idempotent handling, backoff and a poison-message cap