[ https://issues.apache.org/jira/browse/HDFS-9701?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15118375#comment-15118375 ]
Xiao Chen commented on HDFS-9701: --------------------------------- Thanks [~eddyxu] so much for the review! All your comments above are addressed in patch 4, except: bq. Can you move TestDatanodeImpl#testRemoveVolumeBeingWritten to TestDataNodeHotSwapVolumes? I put it in {{TestFsDatasetImpl}} because we need to call methods (e.g. {{createRbw}}, {{finalizeBlock}}) directly on {{FsDatasetImpl}} to control the timing of the 'holding reference but not finalized yet' scenario. Putting it into {{TestDataNodeHotSwapVolumes}} will be harder to test. I understand your concern that this is a hot swap related test, but it's also on FsDatasetImpl. :) I'll try moving it if you insist. bq. Could you verify that during removing a volume, block report can be sent in the test. I added a block report call in patch 4, though not strictly controlled it to be in the middle of removing - I'm not sure how to let it run exactly when the remove is in {{waitVolumeRemoved}} step.... please advice. > DN may deadlock when hot-swapping under load > -------------------------------------------- > > Key: HDFS-9701 > URL: https://issues.apache.org/jira/browse/HDFS-9701 > Project: Hadoop HDFS > Issue Type: Bug > Reporter: Xiao Chen > Assignee: Xiao Chen > Attachments: HDFS-9701.01.patch, HDFS-9701.02.patch, > HDFS-9701.03.patch > > > If the DN is under load (new blocks being written), a hot-swap task by {{hdfs > dfsadmin -reconfig}} may cause a dead lock. -- This message was sent by Atlassian JIRA (v6.3.4#6332)