hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Sergey Shelukhin <ser...@hortonworks.com>
Subject Re: Should compaction be triggered even if User periodically flushes
Date Fri, 05 Apr 2013 17:25:10 GMT
+1... logs would be interesting to see, unless there was configuration
difference it should be the same.
The only thing that comes to mind if off-peak hours starting maybe?

On Fri, Apr 5, 2013 at 9:57 AM, Jean-Daniel Cryans <jdcryans@apache.org>wrote:

> That sounds like a bug to me. Can you find why it's not being compacted?
> Was this behavior changed at some point?
>
> Thx,
>
> J-D
>
>
> On Fri, Apr 5, 2013 at 9:50 AM, ramkrishna vasudevan <
> ramkrishna.s.vasudevan@gmail.com> wrote:
>
> > Hi JM
> > Thanks for your reply.  My concern is not with respect to compaction
> > happening when a region is opening in another RS, my point was though the
> > flushes were happening and more storefiles are created
> > by user initiated flushing, the compaction were not getting triggered.
> >
> >
> > On Fri, Apr 5, 2013 at 4:50 PM, Jean-Marc Spaggiari <
> > jean-marc@spaggiari.org
> > > wrote:
> >
> > > Hi Ram,
> > >
> > > What do you think is wrong with the approach below?
> > >
> > > Doing the compaction when the region is opened in another RS will
> > > improve data locality.
> > >
> > > JM
> > >
> > > 2013/4/5 ramkrishna vasudevan <ramkrishna.s.vasudevan@gmail.com>:
> > > > Hi devs
> > > >
> > > > If user periodically flushes the memstore we keep creating Store
> files.
> > > >  Now if the number of store files  satisfies the compaction policy we
> > > dont
> > > > do compaction.
> > > >
> > > > We print the msg saying compaction_requested=true but do not include
> > into
> > > > the compactsplitthread.
> > > >
> > > > But when the RS goes down and the region gets opened in another RS as
> > > part
> > > > of postOpenDeployTasks we do compaction because the store file count
> > has
> > > > satisfied the compaction criteria.
> > > >
> > > > What could be the right approach here?
> > > >
> > > > Regards
> > > > Ram
> > >
> >
>

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