[jira] [Comment Edited] (CASSANDRA-7560) 'nodetool repair -pr' leads to indefinitely hanging AntiEntropySession
[ https://issues.apache.org/jira/browse/CASSANDRA-7560?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=1407#comment-1407 ] Yuki Morishita edited comment on CASSANDRA-7560 at 7/24/14 4:18 PM: From the jstack logs, it looks like repair session on coordinator node is waiting for validations (merkle trees), but none of the logs show ValidationExecutor running. By default, repair takes snapshot before validating, so it is possible that snapshotting is taking longer on replica node. One possible 'hang' point is snapshot time out. Coordinator waits snapshot response for rpc_timeout millisec, and after that, response handler can be removed. -This is addressed in CASSANDRA-6747, and fixed for 2.1.0.- edit: actually, it is not solving the problem. we need to handle timeouts described here. You can try temporarily set rpc_timeout longer and see if that solves the problem. was (Author: yukim): From the jstack logs, it looks like repair session on coordinator node is waiting for validations (merkle trees), but none of the logs show ValidationExecutor running. By default, repair takes snapshot before validating, so it is possible that snapshotting is taking longer on replica node. One possible 'hang' point is snapshot time out. Coordinator waits snapshot response for rpc_timeout millisec, and after that, response handler can be removed. This is addressed in CASSANDRA-6747, and fixed for 2.1.0. You can try temporarily set rpc_timeout longer and see if that solves the problem. 'nodetool repair -pr' leads to indefinitely hanging AntiEntropySession -- Key: CASSANDRA-7560 URL: https://issues.apache.org/jira/browse/CASSANDRA-7560 Project: Cassandra Issue Type: Bug Components: Core Reporter: Vladimir Avram Attachments: cassandra_daemon.log, cassandra_daemon_rep1.log, cassandra_daemon_rep2.log, nodetool_command.log Running {{nodetool repair -pr}} will sometimes hang on one of the resulting AntiEntropySessions. The system logs will show the repair command starting {noformat} INFO [Thread-3079] 2014-07-15 02:22:56,514 StorageService.java (line 2569) Starting repair command #1, repairing 256 ranges for keyspace x {noformat} You can then see a few AntiEntropySessions completing with: {noformat} INFO [AntiEntropySessions:2] 2014-07-15 02:28:12,766 RepairSession.java (line 282) [repair #eefb3c30-0bc6-11e4-83f7-a378978d0c49] session completed successfully {noformat} Finally we reach an AntiEntropySession at some point that hangs just before requesting the merkle trees for the next column family in line for repair. So we first see the previous CF being finished and the whole repair sessions hangs here with no visible progress or errors on this or any of the related nodes. {noformat} INFO [AntiEntropyStage:1] 2014-07-15 02:38:20,325 RepairSession.java (line 221) [repair #8f85c1b0-0bc8-11e4-83f7-a378978d0c49] previous_cf is fully synced {noformat} Notes: * Single DC 6 node cluster with an average load of 86 GB per node. * This appears to be random; it does not always happen on the same CF or on the same session. -- This message was sent by Atlassian JIRA (v6.2#6252)
[jira] [Comment Edited] (CASSANDRA-7560) 'nodetool repair -pr' leads to indefinitely hanging AntiEntropySession
[ https://issues.apache.org/jira/browse/CASSANDRA-7560?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=1407#comment-1407 ] Yuki Morishita edited comment on CASSANDRA-7560 at 7/24/14 4:33 PM: From the jstack logs, it looks like repair session on coordinator node is waiting for validations (merkle trees), but none of the logs show ValidationExecutor running. By default, repair takes snapshot before validating, so it is possible that snapshotting is taking longer on replica node. One possible 'hang' point is snapshot time out. Coordinator waits snapshot response for rpc_timeout millisec, and after that, response handler can be removed. -This is addressed in CASSANDRA-6747, and fixed for 2.1.0.- -edit: actually, it is not solving the problem. we need to handle timeouts described here.- edit: edit: CASSANDRA-6747 handles timeout also, but the reason we put that to 2.1.0 is that we needed protocol change. It is possible that we can backport only timeout part. You can try temporarily set rpc_timeout longer and see if that solves the problem. was (Author: yukim): From the jstack logs, it looks like repair session on coordinator node is waiting for validations (merkle trees), but none of the logs show ValidationExecutor running. By default, repair takes snapshot before validating, so it is possible that snapshotting is taking longer on replica node. One possible 'hang' point is snapshot time out. Coordinator waits snapshot response for rpc_timeout millisec, and after that, response handler can be removed. -This is addressed in CASSANDRA-6747, and fixed for 2.1.0.- edit: actually, it is not solving the problem. we need to handle timeouts described here. You can try temporarily set rpc_timeout longer and see if that solves the problem. 'nodetool repair -pr' leads to indefinitely hanging AntiEntropySession -- Key: CASSANDRA-7560 URL: https://issues.apache.org/jira/browse/CASSANDRA-7560 Project: Cassandra Issue Type: Bug Components: Core Reporter: Vladimir Avram Attachments: cassandra_daemon.log, cassandra_daemon_rep1.log, cassandra_daemon_rep2.log, nodetool_command.log Running {{nodetool repair -pr}} will sometimes hang on one of the resulting AntiEntropySessions. The system logs will show the repair command starting {noformat} INFO [Thread-3079] 2014-07-15 02:22:56,514 StorageService.java (line 2569) Starting repair command #1, repairing 256 ranges for keyspace x {noformat} You can then see a few AntiEntropySessions completing with: {noformat} INFO [AntiEntropySessions:2] 2014-07-15 02:28:12,766 RepairSession.java (line 282) [repair #eefb3c30-0bc6-11e4-83f7-a378978d0c49] session completed successfully {noformat} Finally we reach an AntiEntropySession at some point that hangs just before requesting the merkle trees for the next column family in line for repair. So we first see the previous CF being finished and the whole repair sessions hangs here with no visible progress or errors on this or any of the related nodes. {noformat} INFO [AntiEntropyStage:1] 2014-07-15 02:38:20,325 RepairSession.java (line 221) [repair #8f85c1b0-0bc8-11e4-83f7-a378978d0c49] previous_cf is fully synced {noformat} Notes: * Single DC 6 node cluster with an average load of 86 GB per node. * This appears to be random; it does not always happen on the same CF or on the same session. -- This message was sent by Atlassian JIRA (v6.2#6252)
[jira] [Comment Edited] (CASSANDRA-7560) 'nodetool repair -pr' leads to indefinitely hanging AntiEntropySession
[ https://issues.apache.org/jira/browse/CASSANDRA-7560?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14067338#comment-14067338 ] Vladimir Avram edited comment on CASSANDRA-7560 at 7/19/14 1:56 AM: There is also a stalled AntiEntropySession on rep1 was (Author: vladmore): There is also a stalled AntiEntropySession on this node. 'nodetool repair -pr' leads to indefinitely hanging AntiEntropySession -- Key: CASSANDRA-7560 URL: https://issues.apache.org/jira/browse/CASSANDRA-7560 Project: Cassandra Issue Type: Bug Components: Core Reporter: Vladimir Avram Attachments: cassandra_daemon.log, cassandra_daemon_rep1.log, cassandra_daemon_rep2.log, nodetool_command.log Running {{nodetool repair -pr}} will sometimes hang on one of the resulting AntiEntropySessions. The system logs will show the repair command starting {noformat} INFO [Thread-3079] 2014-07-15 02:22:56,514 StorageService.java (line 2569) Starting repair command #1, repairing 256 ranges for keyspace x {noformat} You can then see a few AntiEntropySessions completing with: {noformat} INFO [AntiEntropySessions:2] 2014-07-15 02:28:12,766 RepairSession.java (line 282) [repair #eefb3c30-0bc6-11e4-83f7-a378978d0c49] session completed successfully {noformat} Finally we reach an AntiEntropySession at some point that hangs just before requesting the merkle trees for the next column family in line for repair. So we first see the previous CF being finished and the whole repair sessions hangs here with no visible progress or errors on this or any of the related nodes. {noformat} INFO [AntiEntropyStage:1] 2014-07-15 02:38:20,325 RepairSession.java (line 221) [repair #8f85c1b0-0bc8-11e4-83f7-a378978d0c49] previous_cf is fully synced {noformat} Notes: * Single DC 6 node cluster with an average load of 86 GB per node. * This appears to be random; it does not always happen on the same CF or on the same session. -- This message was sent by Atlassian JIRA (v6.2#6252)