[ https://issues.apache.org/jira/browse/HBASE-21559?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16711397#comment-16711397 ]
Zheng Hu commented on HBASE-21559: ---------------------------------- Currently, the snapshotManager grab the object lock in many method. This is a very rough way of locking. I think we should change the locking way of SnapshotManager , not just synchronized the big SnapshotManager object, but use a more concrete lock (in case of dead lock). Anyway , Let me fix this dead lock firstly. So upload a patch.v1. > The RestoreSnapshotFromClientTestBase related UT are flaky > ---------------------------------------------------------- > > Key: HBASE-21559 > URL: https://issues.apache.org/jira/browse/HBASE-21559 > Project: HBase > Issue Type: Bug > Reporter: Zheng Hu > Assignee: Zheng Hu > Priority: Major > Fix For: 3.0.0, 2.1.2, 2.0.4, 2.0.5 > > Attachments: HBASE-21559.v1.patch, > TEST-org.apache.hadoop.hbase.client.TestRestoreSnapshotFromClientAfterSplittingRegions.xml, > > org.apache.hadoop.hbase.client.TestRestoreSnapshotFromClientAfterSplittingRegions-output.txt, > > org.apache.hadoop.hbase.client.TestRestoreSnapshotFromClientAfterSplittingRegions.txt > > > The related UT are: > * TestRestoreSnapshotFromClientAfterSplittingRegions > * TestRestoreSnapshotFromClientWithRegionReplicas > * TestMobRestoreSnapshotFromClientAfterSplittingRegions > I guess the main problem is: a dead lock between SplitTableRegionProcedure > and SnapshotProcedure.. > Attached logs from the failed UT. -- This message was sent by Atlassian JIRA (v7.6.3#76005)