Actions
Feature #63915
openpropagate kafka errors to client in case of sync notifications
% Done:
0%
Source:
Tags:
kafka, notifications
Backport:
Reviewed:
Affected Versions:
Pull request ID:
Description
when broker is down it is expected to reply with an error immdeiatly.
currently, the errors are reported back to only after connection is marked idle, and being deleted.
since this timeout is only used to "garbage collect" decomissioned connections it is set to a high number, meaning that the reply that the notifications failed takes too long.
currently, the errors are reported back to only after connection is marked idle, and being deleted.
since this timeout is only used to "garbage collect" decomissioned connections it is set to a high number, meaning that the reply that the notifications failed takes too long.
- in case of persistent notifications, it will fill up memory with pending coroutines
- in case of non-persistent notifications, it will block all frontend coroutines, making the RGW unusable
in addition, any for non-persistent notifications, since this is detected in the "publish_commit()" stage, it will just be reported and there will be no feedback to the client that the notifications failed.
to solve that, the "publish_reserve()" step should be used to verify the connectivity to the broker
Updated by Yuval Lifshitz 4 months ago
- Status changed from New to Fix Under Review
- Pull request ID set to 55051
Updated by Yuval Lifshitz 25 days ago
- Tracker changed from Bug to Feature
- Subject changed from kafka does not detect and propagete broker down state to propagate kafka errors to client in case of sync notifications
- Status changed from Fix Under Review to New
- Backport deleted (
reef, quincy) - Regression deleted (
No) - Severity deleted (
3 - minor)
Actions