kirktrue commented on PR #17700: URL: https://github.com/apache/kafka/pull/17700#issuecomment-2585579158
Thanks for the review @junrao! > Also, I am wondering why the PR improves the cache eviction rate. If there are more consumer clients than allowed session slots, the eviction is always going to happen. True. From the broker's perspective, the number of partitions in its session cache entries fluctuate up and down on a continuous basis. Again, this is because fetch requests remove partitions from the cache while they're buffered on the consumer and then add them back again after they're "consumed." When a consumer attempts to create a fetch session on the broker, the broker's cache eviction logic will try to evict a "stale" entry or a candidate entry with fewer partitions than the new consumer. If an entry is evicted, its spot in the cache is snagged by the new consumer. > This PR only impacts which client will be chosen for eviction, but doesn't change the number of evictions, right? Yes, when eviction is performed, it's a one-for-one swap with the incoming consumer. However, eviction is occurring at a much frequency due to the volatility of the partition counts in the cache. If the partition counts for all consumers remained constant, the eviction logic would resemble something more like LRU. @jeffkbkim—CMIIW 😄 Thanks! -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: jira-unsubscr...@kafka.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
