Reliable asynchronous processing

Crash a consumer after its side effect but before its ack, see the redelivery double-charge, then enable a dedup table and replay.

A client retries a timed-out payment request, and an Idempotency-Key returns the original response instead of charging again.

Crash the service between "DB commit" and "publish" to strand work, then switch to an outbox and watch the relay recover it.

Visualize a hundred clients retrying against a recovering service, with and without exponential backoff and jitter, and watch synchronized retry waves smooth out.

A malformed message fails forever and blocks its partition until bounded retries move it to a DLQ, where you replay it, possibly causing a second incident.

Silently corrupt derived state, then run a reconciler that diffs it against the source of truth and repairs the gaps.

Cancel a long job mid-flight and see which steps finished, which were rolled back, and which kept running because the worker never checked for cancellation.

Route ride-lifecycle events through Kafka-style partitions, pick the partition key, and slow one consumer group to see that the others are unaffected while ordering holds per key.

Convert a synchronous checkout call chain into events and see new failure modes (partial completion, out-of-order events) appear.

Kill the 2PC coordinator after the prepare phase and watch participants stuck holding locks.

Fail any step in a flight + hotel + car booking, watch compensations unwind backward, then make a compensation fail too.

Append events to a log, rebuild the read model by replay, and add snapshots to see replay cost drop.