[ https://issues.apache.org/jira/browse/HBASE-24687?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17156782#comment-17156782 ]
Sean Busbey commented on HBASE-24687: ------------------------------------- the larger problem is definitely present on master. More threads probably is a good idea; I have seen deployments where it takes over an hour to get through the list of mob files to move (e.g. after a period of having either the cleaner or compaction disabled for a while). but it will need to be something we can tune down in the event HDFS gets overwhelmed. {quote} I think the problem here is not about connections, we should see the directories in the fs, not tables mob enabled, cos there are table modifications( eg. disable mob), which may prevent us from deleting some out-of-date files. {quote} That's an interesting edge case. I think currently we tell folks to manually sideline the mob files after turning off mob and compacting all the data. Doing this automatically would be an ease-of-operations improvement, maybe best to do it in a dedicated jira though? > MobFileCleanerChore uses a new Connection for each table each time it runs > -------------------------------------------------------------------------- > > Key: HBASE-24687 > URL: https://issues.apache.org/jira/browse/HBASE-24687 > Project: HBase > Issue Type: Bug > Components: mob > Affects Versions: 3.0.0-alpha-1 > Reporter: Manas > Assignee: Junhong Xu > Priority: Minor > Attachments: Screen Shot 2020-07-06 at 6.06.43 PM.png > > > Currently creating a new connection for every table under > MobFileCleanerChore.java where we should theoretically just using the > connection from HBase masterservices. > > > -- This message was sent by Atlassian Jira (v8.3.4#803005)