Designing a Public Status Page and Incident Communication Service (System Design Interview)
6 views
"Design a status page like the ones SaaS companies use during outages" is a nice system design prompt because it flips the usual priorities. The system must stay up precisely when everything else is down, and its main job is communication, not raw throughput.
Clarify the requirements
Functional: operators create incidents, post updates (investigating, identified, monitoring, resolved), mark components as degraded or down, and schedule maintenance. Visitors view the current status and history. Subscribers get email, SMS, webhook, or RSS notifications. Optionally, automated health checks can flip component status.
Non-functional: very high availability, especially during the customer's own outage; fast global reads; traffic spikes of 100x when an incident goes viral; strong correctness for status (a page that says "all good" during an outage destroys trust).
Key design decision: isolate it from the main product
Say this early. The status page must not share databases, DNS provider, CDN account, or cloud region with the product it reports on. If the main stack runs in one cloud region, host the status page in another provider or at least another region. Interviewers love hearing that dependency isolation is a requirement, not an optimization.
Read path
Visitors vastly outnumber writers. Render the public page as a static snapshot (HTML plus a small JSON status document) and push it to a CDN whenever something changes. Cache TTLs can be short, around 30-60 seconds, with explicit purges on each update. Even if the origin dies, the CDN keeps serving the last known state, which you should label with a "last updated" timestamp.
Write path
Operators use an admin API backed by a small relational database: tables for components, incidents, incident_updates, maintenance_windows, and subscribers. Writes are rare, so a single primary with a replica in another region is enough. Each write publishes an event to a queue that triggers two workers: one regenerates and publishes the static snapshot, the other fans out notifications.
Notifications at scale
A big customer might have hundreds of thousands of subscribers. The fan-out worker batches recipients, calls email and SMS providers with rate limits and retries, and records delivery state per subscriber so a crash doesn't send duplicates. Use idempotency keys like incident_update_id + subscriber_id. Webhooks need exponential backoff and a dead-letter queue.
Automated checks
Probes from several regions hit customer endpoints. To avoid flapping, require failures from a quorum of regions over a short window before auto-marking a component degraded, and let humans confirm before marking a full outage.
Wrapping up
Close by naming trade-offs: static snapshots are slightly stale but survive origin failure; quorum-based probing delays detection but prevents false alarms. Practicing this kind of structured walkthrough out loud is the best preparation, and tools like Linkjob AI (https://linkjob.ai) can give real-time help during live job interviews and technical interviews when you want support keeping your design discussion organized.