hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ted Yu <yuzhih...@gmail.com>
Subject Re: Some questions to get clear view
Date Wed, 26 Feb 2014 18:43:46 GMT
bq. HBase guarantees ACID semantics per-row

ACID guarantees are at region level.

bq. That's why all CF have to flush when any one of them got memstore limit.

See this comment where LSN means log sequence number:

https://issues.apache.org/jira/browse/HBASE-3149?focusedCommentId=13804537&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13804537



On Wed, Feb 26, 2014 at 10:35 AM, Upendra Yadav <upendra1024@gmail.com>wrote:

> I think....
> HBase guarantees ACID semantics per-row
> That's why all CF have to flush when any one of them got memstore limit.
>
>
>
> On Tue, Feb 25, 2014 at 1:28 PM, Bharath Vissapragada <
> bharathv@cloudera.com
> > wrote:
>
> > Hi Upendra,
> >
> > Your argument is correct, especially when there is an uneven data
> > distribution across CFs in a region and this is what is discussed in
> > HBASE-3149.
> > See comments from Stack, Nicholas & Lars.
> >
> > - Bharath
> >
> >
> > On Tue, Feb 25, 2014 at 12:24 PM, Upendra Yadav <upendra1024@gmail.com
> > >wrote:
> >
> > > Thanks...
> > >
> > > but for a region, why hbase need to flush other CF when one of the CF
> got
> > > memstore limit...
> > >
> > >
> > > On Tue, Feb 25, 2014 at 2:50 AM, Ted Yu <yuzhihong@gmail.com> wrote:
> > >
> > > > Upendra:
> > > > In 0.89-fb branch, the following JIRA has been integrated:
> > > >
> > > > HBASE-3149 Make flush decisions per column family
> > > >
> > > > FYI
> > > >
> > > >
> > > > On Mon, Feb 24, 2014 at 2:11 PM, Upendra Yadav <
> upendra1024@gmail.com
> > > > >wrote:
> > > >
> > > > > Thanks for your Reply...
> > > > >
> > > > > But what is the benefits of different memstore for different CF,
> when
> > > all
> > > > > of them are going to flush on the same time?
> > > > >
> > > > >
> > > > > On Tue, Feb 25, 2014 at 1:23 AM, Jean-Marc Spaggiari <
> > > > > jean-marc@spaggiari.org> wrote:
> > > > >
> > > > > > 1. correct.
> > > > > > 2. Regions doesn't have memstores. Regions servers have
> memstores.
> > On
> > > > per
> > > > > > region per CF. all the memstores for a single regions are flush
> at
> > > the
> > > > > same
> > > > > > time when one is full, even if the others are not.
> > > > > >
> > > > > > HTH.
> > > > > >
> > > > > > JM
> > > > > >
> > > > > >
> > > > > > 2014-02-24 14:35 GMT-05:00 Upendra Yadav <upendra1024@gmail.com
> >:
> > > > > >
> > > > > > > 1. One region server can have more than one region for
same
> table
> > > > > > >
> > > > > > > 2. Which one is correct:
> > > > > > >
> > > > > > > a) Each region has one memstore( and all CF for this region
> will
> > > > reside
> > > > > > in
> > > > > > > this single
> > > > > > > memstore) and if memstore size reached its configured limit
it
> > will
> > > > > > > snapshot and flush... due to single CF all CF have to flush.
> > > > > > >
> > > > > > > b) Each region has n no. of CF and each CF has its own
> memstore.
> > > And
> > > > > when
> > > > > > > one CF's memstore get full it will snapshot and flush.
And will
> > not
> > > > > force
> > > > > > > to flush other CF.
> > > > > > >
> > > > > > > I read the current document some days before and now once
> again i
> > > got
> > > > > > that
> > > > > > > doubts...
> > > > > > >
> > > > > >
> > > > >
> > > >
> > >
> >
> >
> >
> > --
> > Bharath Vissapragada
> > <http://www.cloudera.com>
> >
>

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