z-york commented on pull request #2237: URL: https://github.com/apache/hbase/pull/2237#issuecomment-760432544
I think we can agree to disagree on the automatic/idempotent IMP, we can propose a plan for that if I have more time to devote to it in the future. For now we can keep that as a patch. I think that this PR can still go in with throwing an exception instead of deleting meta (I don't see any reason to submit a separate PR, let's keep this discussion). I think we should fail instead of allowing an automatic delete of the meta directory (or at least have an option to fail and not delete) since we are continuing to add more metadata into meta and it will become more and more costly to rebuild. I have seen many cases where operators clear out ZK nodes + restart master to unblock some assignment issues, but admittedly that is on 1.x versions of HBase, I think the recovery options might be better in 2.x. We already have offline meta repair that I believe should be able to solve these issues if an exception is thrown. btw @saintstack for the double assigned meta scenario wouldn't the IO fencing/lease on the meta WAL handle that? Or will it try to write to a unique WAL each assignment? Just curious, not blocking this PR. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
