REVISIONS / INCIDENT PATTERNS
Every outage leaves
a useful revision.
Read the cause, replay the failure, and compare the fix with the tradeoff it introduces.
Cache stampede
One expired key sends every request to the database at once.
Replay revisionThundering herd
A recovering service gets every waiting request in one breath.
Replay revisionSplit brain
Two leaders accept writes. Neither knows the other is lying.
Replay revisionHot shard
One tenant gets popular and one partition pays for it.
Replay revisionRetry storm
A small outage becomes a crowd of requests trying again.
Replay revisionPoison message
One bad payload blocks the line until the queue stops trusting it.
Replay revision