Good question.  There is no means currently other than putting the
.logs dir back in place and restarting the cluster but we should build
one.  It shouldn't be hard... just stitching together bits of
functionality that already exists.

The awkward part is that these edits will have an edit id that is
probably behind the current state of the edit id that the live regions
are running with (Let me know if you need more detail on this).  So,
we'd need to have a mechanism that somehow ignored edit ids or --
somehow -- only ignore edits that were older than some point in time,
the time of a crash say?  If for sure no versions of data and no
deletes, then we could probably ignore edit ids (edit id is the
sequence ids in hbase internals)

Would you mind making an issue.  Sketch how you think it should work?

Thanks Matthew.
St.Ack


On Sun, Oct 10, 2010 at 11:05 AM, Matthew LeMieux <[email protected]> wrote:
> I have a /hbase/.logs directory that I moved out of the way to get the hbase 
> cluster up and running.
>
> How would I go about "replaying" or recovering that data now that I'm passed 
> the emergency?
>
> -Matthew
>
>

Reply via email to