Repository navigation
KAFKA-20077: Prevent infinite AddPartitionsToTxn retries causing producer.flush() to hang - #21672
nileshkumar3 wants to merge 2 commits into
Conversation
…ucer.flush() to hang
|
@lianetm Can you please review this |
|
A label of 'needs-attention' was automatically added to this PR in order to raise the |
|
A label of 'needs-attention' was automatically added to this PR in order to raise the |
|
This PR is being marked as stale since it has not had any activity in 90 days. If you If you are having difficulty finding a reviewer, please reach out on the [mailing list](https://kafka.apache.org/contact). If this PR is no longer valid or desired, please feel free to close it. If no activity occurs in the next 30 days, it will be automatically closed. |
|
A label of 'needs-attention' was automatically added to this PR in order to raise the |
|
This PR is being marked as stale since it has not had any activity in 90 days. If you If you are having difficulty finding a reviewer, please reach out on the [mailing list](https://kafka.apache.org/contact). If this PR is no longer valid or desired, please feel free to close it. If no activity occurs in the next 30 days, it will be automatically closed. |
|
A label of 'needs-attention' was automatically added to this PR in order to raise the |
Root Cause
The AddPartitionsToTxn handler may repeatedly re-enqueue itself on
retriable errors without any client-side timeout bounding the retry loop.
If the topic is deleted during this phase, the client may never make
progress and producer.flush() may block indefinitely.
Fix
Track the first retriable error timestamp and fail the request if
retries exceed transaction.timeout.ms.