You are viewing a plain text version of this content. The canonical link for it is here.

Posted to dev@hbase.apache.org by "stack (JIRA)" <ji...@apache.org> on 2018/02/20 15:11:00 UTC

[jira] [Created] (HBASE-20022) [AMv2] Unabortable Procedures

stack created HBASE-20022:
-----------------------------

             Summary: [AMv2] Unabortable Procedures
                 Key: HBASE-20022
                 URL: https://issues.apache.org/jira/browse/HBASE-20022
             Project: HBase
          Issue Type: Bug
          Components: amv2
            Reporter: stack


HBASE-18018 added abort to all Procedures. Makes sense. What to do though for those Procedures that are not abortable? For example, HBASE-18016 made it so truncate table could not be aborted (hard to undo deletes of table data or at least, it'd be a bunch of work making a Procedure that supported this...).

The addition of abort to all Procedures added a bit of headache around region merge and split; these Procedures have Points of No Return where once we pass out a step, we cannot go back -- we must press on. Again, we could rewrite the Procedures to support rollback from all points but work has not been done.

So, what is an operator to do if they know their truncate or merge will never succeed? Perhaps this a hypothetical and we want the AMv2 to keep trying; operator can address the issue that is preventing completion? Opening issue that recognizes this hole in our being able to have Procedures ignore abort.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)