Repository navigation
Producer ownership is stale after replica handoff #705
Description
Describe the Bug
Hi! An AI coding assistant found this, but it’s reproducible on Flue 2.0.3 with both Postgres and Redis.
When a conversation runs on A, then B, then A again, the third turn logs Producer ownership is stale. twice and takes ~45 seconds to complete, even with an immediately responding fake model.
Expected Behavior
A should resume the conversation without stale ownership errors or the ~45-second delay.
Steps to Reproduce
Clone https://github.com/Aljutor/flue-replica-handoff-repro and run from the repo directory:
docker compose run --build --rm testsNo API keys needed. The README also has Redis instructions.
Is this handoff supported, or am I missing something in the setup?
Activity
FredKSchott commentedon Oct 19, 2026
Closing this for now. In Flue 3, the producer epochs behind this error are gone, but moving an instance from one process to another and back is not supported: each instance should be routed to a single owning process. We may revisit multi-process handoff later.