What a rollback is
Deploying the previous release of one service. The platform stops the running process, starts the previous version and waits for /healthz before reporting success. Requests in flight during the switch can fail for a second or two.
Before you roll back
- Confirm which release is running now and which one ran before it (deploy history shows
versionandprevious). - Confirm the evidence: errors began within minutes of the deploy, and the error signature is in code the release changed. A deploy that happened hours before the incident is rarely the cause.
- Rollbacks are reversible: redeploying the newer release undoes them.
How
Deploy <service> at the previous version through the platform's deploy action, with a reason that links the incident. Watch the error rate for two minutes after the switch; it should return to baseline within one alert window.
After
Open a ticket for the release owner with the stack trace and the commit. Do not redeploy the bad release until it is fixed.