[ https://issues.apache.org/jira/browse/HBASE-22075?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16877397#comment-16877397 ]
Sean Busbey commented on HBASE-22075: ------------------------------------- I've been having a lot of trouble trying to reproduce this failure using the IT on the CDH5 backport of the MOB feature. It's been confusing since the code that I think is responsible is essentially the same - bulk loading the updated references into the non-MOB regions. At the end of last week I started looking at the bulk load code to see if that was the difference. [~npopa] suggested it was a difference in how much failures are retried by default. It doesn't look obviously different. However, if I take the master branch and update the {{PartitionedMobCompactor}} to set {{hbase.bulkload.retries.number}} to {{MAX_INT}} on the conf it uses when calling bulk load, then the IT no longer reproduces the failure. So even if it isn't specifically the bulkload retries that are different I think we're on the right track that currently the "if it fails just bail" approach is the source of the inconsistency. I'm going to see if I can rework how we do committing the post-compaction references in the non-MOB regions to expressly handle error conditions. > Potential data loss when MOB compaction fails > --------------------------------------------- > > Key: HBASE-22075 > URL: https://issues.apache.org/jira/browse/HBASE-22075 > Project: HBase > Issue Type: Bug > Components: mob > Affects Versions: 2.1.0, 2.0.0, 2.0.1, 2.1.1, 2.0.2, 2.0.3, 2.1.2, 2.0.4, > 2.1.3 > Reporter: Vladimir Rodionov > Assignee: Vladimir Rodionov > Priority: Critical > Labels: compaction, mob > Fix For: 2.0.6, 2.2.1, 2.1.6 > > Attachments: HBASE-22075-v1.patch, HBASE-22075-v2.patch, > HBASE-22075.test-only.0.patch, HBASE-22075.test-only.1.patch, > HBASE-22075.test-only.2.patch, ReproMOBDataLoss.java > > > When MOB compaction fails during last step (bulk load of a newly created > reference file) there is a high chance of a data loss due to partially loaded > reference file, cells of which refer to (now) non-existent MOB file. The > newly created MOB file is deleted automatically in case of a MOB compaction > failure, but some cells with the references to this file might be loaded to > HBase. -- This message was sent by Atlassian JIRA (v7.6.3#76005)