[ https://issues.apache.org/jira/browse/HBASE-27964?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17740578#comment-17740578 ]
Zheng Wang commented on HBASE-27964: ------------------------------------ Here is a test. I write data to a cluster which has two regionServers, and limit the compaction throughput, the node9(10.0.0.9) apply the patch and disable delayed selection, we can see the store_file_count and store_file_size grows similar, but the compaction_queue_length has big difference, the node23(10.0.0.23) obviously incorrect. !image-2023-07-06-19-51-21-354.png|width=503,height=192! !image-2023-07-06-20-00-21-587.png|width=502,height=190! !image-2023-07-06-20-00-41-933.png|width=501,height=192! > Adds a switch for compaction's delay selection feature > ------------------------------------------------------ > > Key: HBASE-27964 > URL: https://issues.apache.org/jira/browse/HBASE-27964 > Project: HBase > Issue Type: Improvement > Components: Compaction > Reporter: Zheng Wang > Assignee: Zheng Wang > Priority: Major > Attachments: image-2023-07-06-19-51-21-354.png, > image-2023-07-06-20-00-21-587.png, image-2023-07-06-20-00-41-933.png > > > When the compact pressure is high, delayed selection can cause the metric of > compact queue length to continuously increase incorrectly. We should provide > an option to disable this feature if the user values metric accuracy more. > See HBASE-26987 for more detail. -- This message was sent by Atlassian Jira (v8.20.10#820010)