hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ted Yu <yuzhih...@gmail.com>
Subject Re: batch version of checkAndPut
Date Mon, 29 Apr 2013 20:31:57 GMT
Hari:
Do you mind logging a JIRA to capture the requirement ?

Thanks

On Mon, Apr 29, 2013 at 1:30 PM, Hari Mankude <hmankude@talena-inc.com>wrote:

> Yes, client side will partition the puts on region boundaries. However,
> instead of doing batch put, it would be required to do something equivalent
> of batch checkAndPut().
>
> thanks
> hari
>
> On Mon, Apr 29, 2013 at 10:57 AM, Ted Yu <yuzhihong@gmail.com> wrote:
>
> > Looks like this would require client side to partition the batch Puts
> > according to region boundaries.
> >
> > Is my understanding correct ?
> >
> > On Mon, Apr 29, 2013 at 10:54 AM, Hari Mankude <hmankude@talena-inc.com
> > >wrote:
> >
> > > Sure, the use case is updating 100s of rows [each row has to be updated
> > > atomically] by multiple threads. Batching is setup to do aggregate data
> > on
> > > a regionserver granularity. The objective is to reduce the rpc overhead
> > > similar to batch get or batch put. The objective is NOT to do cross-row
> > > locking or expect any type of multi-row atomicity.
> > >
> > > hari
> > >
> > > On Mon, Apr 29, 2013 at 9:13 AM, Ted Yu <yuzhihong@gmail.com> wrote:
> > >
> > > > checkAndPut() checks specific row. Can you outline your requirement
> in
> > a
> > > > bit more detail ?
> > > >
> > > > Meaning, if you supply List of Puts to the batch API. Are the Puts
> > > > targeting the same row ? If not, how would the Puts outside the
> region
> > > > where the checked row belongs be handled ?
> > > >
> > > > Cheers
> > > >
> > > > On Mon, Apr 29, 2013 at 8:30 AM, Hari Mankude <
> hmankude@talena-inc.com
> > > > >wrote:
> > > >
> > > > > Is there a batch version of checkAndPut() in the trunk? If so, can
> > you
> > > > > point me to the appropriate jira?
> > > > >
> > > > > thanks
> > > > > hari
> > > > >
> > > >
> > >
> >
>

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