Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Threaded View
HBase >> mail # dev >> HEADSUP: Working on new 0.96.0RC


Copy link to this message
-
Re: HEADSUP: Working on new 0.96.0RC
Not the move call, the HMaster#unassign() call.  It reads the region from
meta.  It is used in MoveRandomRegionOfTableAction.

For this call, we don't want to check AM region states map because it is
used by hbck in case AM region states is stale.
On Wed, Oct 9, 2013 at 4:05 PM, Enis Söztutar <[EMAIL PROTECTED]> wrote:

> On Wed, Oct 9, 2013 at 2:49 PM, Jimmy Xiang <[EMAIL PROTECTED]> wrote:
>
> > I prefer to have 9696 in. It's not just about merging.  I am also trying
> > to make sure splitting is good.  Currently, if a region is splitting, the
> > two daughters are wrote to meta at first. CM could move them around
> before
> > master knows about these two new regions. So they could be
> double-assigned
> > for a short while.  It could be a cause why ITBLL still shows data loss
> > somewhere.
> >
>
> Is this really the case? If client learns the daughter regions from meta,
> before master learns about the split, even if they call
> HMaster.moveRegion(), they would get UnknownRegionException, no?
>
>  void move(final byte[] encodedRegionName,
>
>       final byte[] destServerName) throws HBaseIOException {
>
> RegionState regionState = assignmentManager.getRegionStates().
>
>       getRegionState(Bytes.toString(encodedRegionName));
>
>     if (regionState == null) {
>
>       throw
> newUnknownRegionException(Bytes.toStringBinary(encodedRegionName));
>     }
>
>
>
> >
> > I think we should make sure ITBLL runs well with no data loss before we
> > release 0.96.0.  Data loss is a big concern to me.
> >
> >
> > On Wed, Oct 9, 2013 at 2:33 PM, Devaraj Das <[EMAIL PROTECTED]>
> wrote:
> >
> >> I am not sure I agree with this though. The reason being - HBASE-9696
> was
> >> raised on Saturday and we have cut an RC after that. So why not another
> >> one
> >> now? For the 0.96.0 version, can we not say that "merge" should be used
> >> with caution. Also, it is not guaranteed that we will not face any new
> IT
> >> issues after 9696 goes in, right?
> >> Let's cut 0.96.0 now and fix remaining issues in 0.96.1. Thoughts?
> >>
> >>
> >> On Wed, Oct 9, 2013 at 2:17 PM, Elliott Clark <[EMAIL PROTECTED]>
> wrote:
> >>
> >> > At this point I think that we should have real clean IT test runs
> before
> >> > cutting another release.  And we can't really get that until the
> master
> >> > always comes back up (The issue stack was working on yesterday) and
> >> until
> >> > merging is stable.  I would like to see those two things fixed before
> >> 0.96
> >> >
> >> >
> >> > On Wed, Oct 9, 2013 at 1:38 PM, Devaraj Das <[EMAIL PROTECTED]>
> >> wrote:
> >> >
> >> > > I'd say we cut an RC now (without any more fixes).
> >> > >
> >> > >
> >> > > On Wed, Oct 9, 2013 at 12:45 PM, Jimmy Xiang <[EMAIL PROTECTED]>
> >> > wrote:
> >> > >
> >> > > > It's testing now. :)
> >> > > >
> >> > > >
> >> > > > On Wed, Oct 9, 2013 at 12:42 PM, Sergey Shelukhin <
> >> > > [EMAIL PROTECTED]
> >> > > > >wrote:
> >> > > >
> >> > > > > 9696 looks a little bit scary... did you guys test it on your
> rig?
> >> > > > >
> >> > > > >
> >> > > > > On Wed, Oct 9, 2013 at 11:54 AM, Stack <[EMAIL PROTECTED]>
> wrote:
> >> > > > >
> >> > > > > > On Wed, Oct 9, 2013 at 10:45 AM, Enis Söztutar <
> [EMAIL PROTECTED]
> >> >
> >> > > > wrote:
> >> > > > > >
> >> > > > > > > I just committed
> >> > > https://issues.apache.org/jira/browse/HBASE-9730for
> >> > > > > > > this.
> >> > > > > > > Time for another RC, what do you think?
> >> > > > > > >
> >> > > > > > > You know what they say, sixth time is the charm.
> >> > > > > >
> >> > > > > >
> >> > > > > >
> >> > > > > > I can cut one no problem.  Just say.
> >> > > > > >
> >> > > > > > Does your test rig pass?  Ours hasn't yet because of
> HBASE-9563;
> >> > > master
> >> > > > > is
> >> > > > > > killed and won't come back though restarted and tests fail.
> >> > > > > >
> >> > > > > > Do we want HBASE-9696 in there?  It is currently under test.
> >> > > > > >
> >> > > > > > And HBASE-9724 Failed region split is not handled correctly by