Uptime monitoring service inspired by UptimeRobot and Pingdom: probes report check results, public status pages are cached and served from the edge, and outage alerts go out through a background queue.
See why reading the database for every status page view does not scale.
What was missing: A status page should not query the database on every view.
Why that's risky: Status pages get most traffic exactly when something is down, so the page that tells you about an outage is the first thing to fail.
What gets added: One server that reads a site's status from the database on every view and stores each probe's result.
Trade-offs: Always current, but fragile under load.
Separate viewing from reporting and serve status from memory.
What was missing: Viewing and reporting share one path.
Why that's risky: A cached status can show 'up' for a few seconds after a site goes down.
What gets added: A status service reading from a cache keyed by site, a result service for probe reports, and a status database as source of truth.
Trade-offs: A few seconds of lag is acceptable; a status page must stay up more than it must be instant.
Serve public status pages from the edge so an outage never overloads us.
What was missing: A big outage sends everyone to the status page at once.
Why that's risky: At the edge a status is no fresher than its TTL.
What gets added: A CDN serves public status pages from the edge for a few seconds at a time.
Trade-offs: A short TTL (5 to 10 seconds) is invisible to viewers and still cuts origin load by orders of magnitude.
Send outage alerts in the background and watch the pipeline.
What was missing: Outage alerts are not sent and a faulty probe can flood results.
Why that's risky: A single failed probe is often the probe's own network, not the site; alerting on it pages people for nothing, and people learn to ignore alerts.
What gets added: An alert queue with workers that confirm a failure from a second region before notifying the site's contacts, a rate limiter on probe pushes and monitoring.
Trade-offs: Requiring agreement from more regions (for example 2 of 3) cuts false alarms but delays the alert and can hide an outage that affects only one region.
You're in the middle of an interview session. Leaving now will end your current attempt.
Explore concept overviews, real-system examples, key tradeoffs, and interview talking points for each roadmap section.
You've conquered this phase. These are the skills you now own: