hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Enis Söztutar <e...@apache.org>
Subject Re: HBCK options to disable master maintenance threads
Date Mon, 20 Jun 2016 21:33:37 GMT
On Mon, Jun 20, 2016 at 1:49 PM, Stephen Jiang <syuanjiangdev@gmail.com>
wrote:

> Enis, what I suggested was that even no repair is suggested, we still
> should disable master maint tasks in online check for more deterministic
> result.
>

I see, makes sense as long as we are finished with HBASE-16008.


>
> Thanks
> Stephen
>
> On Mon, Jun 20, 2016 at 1:44 PM, Enis Söztutar <enis.soz@gmail.com> wrote:
>
> > check out the corresponding shouldXXX commands:
> >
> >   public boolean shouldDisableBalancer() {
> >
> >     return fixAny || disableBalancer;
> >
> >   }
> > If fixAny which is true if any of the -fix is run, we disable the master
> > chores.
> >
> > For -fixHdfsOverlaps and -fixHdfsHoles, I've mentioned this in the jira I
> > think, but we should deprecate those, and do -fixOverlaps and -fixHoles
> > separately. These two new commands will look at BOTH hdfs and meta to
> > decide on what to do.
> >
> > Enis
> >
> > On Mon, Jun 20, 2016 at 12:30 PM, Stephen Jiang <syuanjiangdev@gmail.com
> >
> > wrote:
> >
> > >       } else if (cmd.equals("-disableBalancer")) {
> > >
> > >         setDisableBalancer();
> > >
> > >       }  else if (cmd.equals("-disableSplitAndMerge")) {
> > >
> > >         setDisableSplitAndMerge();
> > >
> > > In HBCK, we will either use the options to disable master maintenance
> > work
> > > (see above) or the master maintenance are disabled during repair.
> > >
> > > I think we should always disable master maintenance work during online
> > > HBCK, because balancer moving regions around during online check; or
> > > split/merge regions during online check would have unexpected side
> > effect.
> > >
> > > How do you think?
> > >
> > > Thanks
> > > Stephen
> > >
> > > Also, I think we have too many options.  We really should reduce
> options
> > in
> > > hbck so that it is more user friendly (eg. currently implementation of
> > > -fixHdfsOverlaps would almost 100% create hole, it does not make sense
> to
> > > run it alone, it should always run with -fixHdfsHoles option; and very
> > > likely with -fixMeta option)
> > >
> >
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message