The consumer needs to replay data the log no longer has
A new downstream model needs a year of event history to compute its first output. The broker retains a fraction of that. A team member suggests raising the retention setting and re-running.
What you would do first
Answer before revealing anything. The value of the exercise is entirely in committing to a diagnosis you can be wrong about.
- 1Establish exactly what history is required and at what grain — a year of raw events, or a year of derived state.
- 2Check what the log actually still holds, per partition, rather than what the configuration says.
- 3Look for the same events in the raw landing zone. If raw was written exactly as received, it is the replay source and the broker is not needed at all.
- 4Determine whether the source system can re-emit — a snapshot plus stream, or a historical export — and what that costs the source.
What is actually going on
The trap
The fix that looks right. Read it even if you got the answer — especially then.
Raise the retention setting and re-run the consumer from the earliest available offset. The job succeeds, the model builds, and it silently begins a year of history at whatever point the log happens to start — producing a dataset whose early periods are incomplete in a way no downstream check will attribute to this decision.