[ https://issues.apache.org/jira/browse/HBASE-12424?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14196660#comment-14196660 ]
Andrew Purtell commented on HBASE-12424: ---------------------------------------- Definitely need to break out coprocessor upcall execution times during split transaction processing. See PHOENIX-1402 > Finer grained logging and metrics for split transactions > -------------------------------------------------------- > > Key: HBASE-12424 > URL: https://issues.apache.org/jira/browse/HBASE-12424 > Project: HBase > Issue Type: Improvement > Reporter: Andrew Purtell > Fix For: 2.0.0, 0.98.8, 0.99.2 > > Attachments: HowHBaseRegionSplitsareImplemented.pdf > > > A split transaction is a complex orchestration of activity between the > RegionServer, Master, ZooKeeper, and HDFS NameNode. We have some visibility > into the time taken by various phases of the split transaction in the logs. > We will see "Starting split of region $PARENT" before the transaction begins, > before the parent is offlined. Later we will see "Opening $DAUGHTER" as one > of the last steps in the transaction, this is after the parent has been > flushed, offlined, and closed. Finally ""Region split, hbase:meta updated, > and report to master ... Split took $TIME" after all steps are complete and > including the total running time of the transaction. > For debugging the cause(s) of long running split transactions it would be > useful to know the distribution of time spent in all of the phases of the > split transaction. -- This message was sent by Atlassian JIRA (v6.3.4#6332)