All terms
Glossary
Chaos engineering
Controlled disruption in a running system, to find out whether the safeguards actually hold.
Instead of hoping failover works, you trigger it: shut down an instance, delay a connection, block a dependency.
It only makes sense once monitoring and recovery exist. Before that you create outages without insight.
How you notice it
- There is failover that has never been triggered.
- The last outage left open questions.
- Monitoring and recovery are in place.
Frequently asked
Does this have to happen in production?
No, and at the start it should not. The first run in staging with the database switched off surfaces plenty: missing timeouts, restart loops, unreadable errors. Production comes when that round stops producing news.