Objective.
Kill one of two instances of an application while customers use it, and count the errors they see.
Scenario.
One instance was killed while a client called the application continuously through the edge.
Expected behaviour.
- Serve every request from the remaining instance.
- Replace the killed instance.
Measured recovery.
- Recovery time RTO
- 0 s
- How long recovery took.
- Application health
- Pass
Customer impact.
- Application availability
- 0 errors in 806 requests
What happened.
- One instance killed.
- Requests served by the remaining instance.
What we learned.
- With two instances, customers saw no error.
What we changed.
- No change needed.
Scope of this drill.
An application with two instances.