[ https://issues.apache.org/jira/browse/HBASE-4778?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13149977#comment-13149977 ]
Nicolas Spiegelberg commented on HBASE-4778: -------------------------------------------- @Ted: In this particular case, I think we ended up finding that HDFS wasn't reading truncated HFiles properly. Overall though, the idea is to prioritize data integrity & consistency over availability. We shouldn't be silently opening regions with missing data. We should instead understand why the data is missing. If someone wants to add a 'data.loss.acceptable' flag and allow this to happen, then that's fine. > Don't ignore corrupt StoreFiles when opening a region > ----------------------------------------------------- > > Key: HBASE-4778 > URL: https://issues.apache.org/jira/browse/HBASE-4778 > Project: HBase > Issue Type: Bug > Reporter: Nicolas Spiegelberg > Assignee: Mikhail Bautin > Fix For: 0.94.0 > > Attachments: HBASE-4778.patch > > > We used to ignore StoreFiles that failed to open, which led to a situation > when only a subset of regions was opened, and HBase did not return results to > clients for the affected set of keys. This change makes sure we propagate > IOExceptions coming from an attempt to open a StoreFile all the way up to > HRegionServer.openRegion, where it will lead to a failure to open the whole > region. This way we can avoid returning corrupt data to the application. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa For more information on JIRA, see: http://www.atlassian.com/software/jira