A dropped broker connection produced a tight reconnect loop.
Conversation
Turn 1 — User
When the broker goes down the logs fill up in seconds.
Turn 2 — Assistant
The reconnect had no delay. It now backs off exponentially up to a ceiling, with jitter so a fleet coming back does not reconnect in lockstep.
The log line moved to a summary — one line per attempt with the current delay, rather than one per failure.
Turn 3 — User
Does it give up eventually?
Turn 4 — Assistant
No, it keeps retrying at the ceiling. For a background collector, continuing to try is more useful than exiting and needing supervision to restart it.
Subjects
- [[Time Series]]
- [[Observability]]