[ https://issues.apache.org/jira/browse/HBASE-19346?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16266530#comment-16266530 ]
Chance Li commented on HBASE-19346: ----------------------------------- so, the action of remove the #waitingAckQueue in #FanOutOneBlockAsyncDFSOutput.failed() have been moved to #flush0 and #flushBuffer. To keep the #AsyncFSWal.syncFail()'s semantic as you mentioned before: roll have done, #AsyncFSWal.syncFail() will never receive the old callback. I think it's not make sure on the principle, such as Context Switch is very slow. pls correct me if I'm wrong. > Use EventLoopGroup to create AsyncFSOutput > ------------------------------------------ > > Key: HBASE-19346 > URL: https://issues.apache.org/jira/browse/HBASE-19346 > Project: HBase > Issue Type: Sub-task > Reporter: Duo Zhang > Assignee: Duo Zhang > Fix For: 2.0.0-beta-1 > > Attachments: HBASE-19346-v1.patch, HBASE-19346.patch > > > So that we can use different event loop to manage the connections to > different datanodes. And since EventLoop itself is also an EventLoopGroup, we > could still use the event loop to create AsyncFSOutput so the logic of > AsyncFSWAL will not be broken. > Will open a new issue to modify AsyncFSWAL and finally we can use multiple > event loop. -- This message was sent by Atlassian JIRA (v6.4.14#64029)