eldenmoon opened a new pull request, #61111: URL: https://github.com/apache/doris/pull/61111
When enable_snapshot_point_query=true in cloud mode, every point query triggers an independent RPC to MetaService to fetch the partition visible version. Under high concurrency, this creates N RPCs for N concurrent queries on the same partition, becoming a significant bottleneck. This patch introduces PointQueryVersionCache, a request-coalescing cache that reduces MetaService RPCs by: - Short TTL caching: partition versions are cached for a configurable duration (point_query_version_cache_ttl_ms, default 0 = disabled). Within the TTL window, concurrent queries reuse the cached version. - Request coalescing: when cache expires, only the first request issues the RPC; concurrent requests wait on the inflight result via CompletableFuture. Under 100 concurrent point queries with 5ms RPC latency, the cache reduces MetaService RPCs from 100+ to 1-2 (>= 20x reduction). Changes: - New: PointQueryVersionCache.java - singleton request-coalescing cache - Modified: PointQueryExecutor.updateCloudPartitionVersions() to use cache - Modified: SessionVariable.java - add point_query_version_cache_ttl_ms - New: PointQueryVersionCacheTest.java - 11 unit tests - New: PointQueryExecutorPerfTest.java - E2E performance benchmark -- 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: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
