← All concepts
Messaging & flow
Back-pressure & queues
When producers outpace consumers, buffer and slow down instead of falling over.
Spikes (a viral upload, a flash sale) arrive faster than workers can process. A queue absorbs the burst so the cost is delay, not failure, and signals upstream to slow down when it grows too long.
See it in action
Step through the scenarios on the diagram.
Laying out diagram…
1/3A viral moment brings 10× the usual uploads; the API just enqueues jobs and replies 'accepted'.
- API
- — Accepts uploads quickly
- Job queue
- — Absorbs bursts
- Workers
- — Process at their own pace
- Dead-letter queue
- — Messages that keep failing
- Autoscaler
- — Scales on queue depth
How it works
- Queue between request path and heavy work (transcoding, emails, fan-out).
- Autoscale consumers on queue depth.
- Dead-letter queues for messages that keep failing.
- Reject or shed low-priority load when queues are full.
Trade-offs
- Users see delays during spikes; queues need monitoring.
Seen in
- YouTubeGlobal video upload, transcode, and streaming
- Web crawlerFetch billions of pages politely and without repeats
- WebSockets at scaleMillions of live connections with Go on Kubernetes
- BookMyShowMovie and event tickets with live seat selection
- MakeMyTripFlights and hotels aggregated from many suppliers
- Stock brokerageLive quotes, order placement and portfolios
- Live sports scoresReal-time scores, commentary and alerts for millions