hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jonathan Gray (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HBASE-2409) Daughters deployed to parents regionserver node, always
Date Wed, 02 Jun 2010 01:15:38 GMT

    [ https://issues.apache.org/jira/browse/HBASE-2409?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12874356#action_12874356

Jonathan Gray commented on HBASE-2409:

IMO this "issue" still exists where splits almost always both go back to same RS.  To change
that behavior requires explicit work to be done.  The "tendencies" that region transitions
follow today will no longer hold with the master rewrite stuff going on as the piggybacking
behavior is going away.  We need to define what we want to happen.

I think this still might be an invalid issue or won't fix because we may not care about that
behavior anymore.  Or we may actually want it.

The primary reason this was problematic before was during an import into a new table.  With
the new createTable api that allows you to specify a bunch of regions up front, it doesn't
really matter as much if splits go to the same server.

If we could do anything then and we aren't concerned about achieving immediate load balancing,
it might be preferable for splits to stay on the same server.  This allows for a split to
be totally contained on the RS without any interaction with the Master so we can have data
unavailability really short, no need to reassignment.  This also lets client requests for
both regions to still go to the right server, reducing META fetches.  Normal load balancing
will balance overall cluster load over time.

> Daughters deployed to parents regionserver node, always
> -------------------------------------------------------
>                 Key: HBASE-2409
>                 URL: https://issues.apache.org/jira/browse/HBASE-2409
>             Project: HBase
>          Issue Type: Bug
>            Reporter: stack
>             Fix For: 0.21.0
> Ryan apparently said this 6 months ago, that on split, the regions are assigned back
to the parent hosting regionserver almost always.  Then, to keep up some kinda balance, the
load balancer kicks and closes the just opened daughters -- which have just been opened a
moment ago -- to deploy them elsewhere in the name of keeping good balance.
> This issue is to confirm the above behavior indeed happens and then to take action to
make it so at least one of the daughters is held up so it doesn't go back to the current heartbeating
host, the parent hosting server.

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

View raw message