Ben Lau commented on HBASE-20054:

[~yuzhih...@gmail.com] I guess it doesn't hurt to forward port the test 
improvement although the bug doesn't exist in master.  For the other change 
though (queue adoption), I wonder.  If master branch decided to abort the 
server if there is a replication error instead of doing a retry, maybe that is 
something we should go with instead.  In other words, maybe we should backport 
the server abortion to branch-2, instead forward-porting the RPC retry to 
master.  I also thought about this more and I wonder if indefinitely retrying 
is a good idea anyway.  Maybe it would be better to have a limited # of retries 
and then abort.  But it might be simpler to just abort immediately like in 
master-branch.  Thoughts?

> Forward port HBASE-18282 ReplicationLogCleaner can delete WALs not yet 
> replicated in case of KeeperException
> ------------------------------------------------------------------------------------------------------------
>                 Key: HBASE-20054
>                 URL: https://issues.apache.org/jira/browse/HBASE-20054
>             Project: HBase
>          Issue Type: Bug
>            Reporter: Ted Yu
>            Assignee: Ted Yu
>            Priority: Major
>             Fix For: 3.0.0
>         Attachments: 20054.v1.txt
> Some changes from HBASE-18282 were missing in master branch:
> * The retry in ReplicationSourceManager#adoptAbandonedQueues
> *  new tests in TestLogsCleaner
> This issue is to forward port the above to master branch

This message was sent by Atlassian JIRA

Reply via email to