Description
The method 'resizeThreadPool' is suppose to redistributed all topic partitions to threads based on the new thread number, and they should be distributed evenly.
However, the resizing process which has a logic to add fetcher to 'fetcherThreadMap' is within the iteration of 'fetcherThreadMap' all fetchers, which could lead to skipping some fetchers and these fetchers remain their topic partition assignment.
The affected scenario: someone wants to add threads to improve fetch-thread performance(eg. replica fetch threads), but the performance improvement is not as expect.
Attachments
Issue Links
- links to