Exactly-once claims and the reality
Why exactly-once is an application-level property, not a network guarantee.
Learning path
Reliable distributed workflows
Build workflows that survive retries, partial failures, and reordered messages.
5of 10 available now
Start here
Get started
How to make repeated operations safe in the presence of retries.
Start with IdempotencyIn order
Topics.
0120 min read→0225 min read→0325 min read→0430 min read→0520 min read→06 07Coming soon 08Coming soon 09Coming soon 10Coming soon
Idempotency
How to make repeated operations safe in the presence of retries.
Timeouts, retries, backoff, and jitter
Bound waiting time and recover from transient failures without amplifying load.
Queues and asynchronous processing
Decouple producers from consumers to smooth load and isolate failures.
Transactional outbox
Publish events reliably by writing them in the same transaction as state.
Synchronous request-response and its limits
Where synchronous calls help and where they create fragile coupling.
Coming soon
Sagas and compensating transactions
Coordinate multi-step business processes without holding locks end-to-end.
Dead-letter queues and manual workflows
Isolate unprocessable messages and give humans a path to resolve them.
Backpressure and flow control
Let a system signal overload instead of collapsing under it.
Observability for reliability
Instrument workflows so retries, duplicates, and stalls are visible.