pvary commented on PR #13382: URL: https://github.com/apache/iceberg/pull/13382#issuecomment-3018282040
> I replaced all Caffeine `maximumSize` caches with `LRUCache` in this PR. The only one left is `DynamicWriteResultAggregator`, which uses `expireAfterWrite` semantics: > > ```java > this.specs = > Caffeine.newBuilder().expireAfterWrite(CACHE_EXPIRATION_DURATION).softValues().build(); > this.outputFileFactories = > Caffeine.newBuilder().expireAfterWrite(CACHE_EXPIRATION_DURATION).softValues().build(); > ``` > > Should we replace that with `LRUCache` as well? Before we move forward, let's take a step back and consider the size of the cache, and the cost of a cache miss. How many items do we need in the cache for the optimal operation: 1. Is this something which is static and we can configure? 2. Is this something which should be configured by the user and relatively stable for a single job? 3. Is this something which is dependent on the workload, and increases as the number of tables accessed by the job increases? This one looks like number 3 to me. Also the cache access is only several times during a checkpoint, so the performance is less important. So for me this seems like better handled by caffeine, but I could be convinced. Also please revisit the pervious decisions on the cache sizes, and reconsider if appropriate. Thanks, Peter -- 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: issues-unsubscr...@iceberg.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@iceberg.apache.org For additional commands, e-mail: issues-h...@iceberg.apache.org
