Hi Demai,
I thought making the client communicate with Central Server
frequently may not be efficient. In case of replication, I can do that
periodically in bulk.
On 7 April 2014 22:00, Demai Ni <nid...@gmail.com> wrote:
> I looks to me that. You like to have the small clusters located close to
> the client, and then use the smaller clusters as Masters. So there will be
> multi-Masters and one-Slave cluster setup. And the one-Slave cluster is the
> centralized and large HBase server.
>
> Well, it works. But I don't get the two points:
> 1) what's saved here? to get the replication works from the smaller
> clusters to the centralized large cluster will consume the cpu/storage and
> network resource. So why not get the client to talk directly to the
> centralized cluster. The added-on layer of the smaller clusters will only
> reduce the performance. One factor is that if the network doesn't allow
> the client to talk directly to centralized cluster, the replication won't
> work well either due to the lag.
> 2) I still don't get why allow client to delete on smaller cluster and
> doesn't allow such transactions replayed on the centralized cluster.
> Assume you have a good business reason to disallow client to delete
> existing data, an application layer of permission control may be better.
> That is don't allow delete on smaller clusters neither.
>
> just my 2 cents.
>
> Demai
>
>
>
>
>
> On Sun, Apr 6, 2014 at 11:09 PM, Manthosh Kumar T <manth...@gmail.com
> >wrote:
>
> > My use case is that I need to replicate between two geographically
> distant
> > clusters. In the simplest form, I have several geographically distant
> > clients that needs to add data to a geographically distant centralized
> > HBase server. So, I thought I'll maintain a small cluster at each client
> > and make that replicate to the central server. But when I delete in the
> > small cluster, I don't want that to be replicated in the centralized
> > server. Hence, I think the coproc route is fine. Please correct me if I'm
> > wrong. Or is there a better solution for my use case?
> >
> >
> > On 5 April 2014 05:16, Demai Ni <nid...@gmail.com> wrote:
> >
> > > agree about the suggestion above. just like to chime in a bit more.
> > >
> > > One question, how do you like to treat the 'put' on the existing row?
> > well,
> > > it is a delete + addition to some degree.
> > >
> > > If go to the coproc route, maybe better not to use replication at all.
> > > Basically, you can have two tables: "source" and "backup", and when
> ever
> > a
> > > 'put' is on 'source', the coproc will replay it to 'backup'. And the
> > table
> > > 'backup' can be on the same cluster or another cluster.
> > >
> > > Not sure about your use case, maybe the existing 'version' feature (
> > > http://hbase.apache.org/book/schema.versions.html) can be used with a
> > > large
> > > max-version. If the same row/cell won't rewritten a lot, then the cost
> > will
> > > be similar as the replication or coproc.
> > >
> > > Demai
> > >
> > >
> > > On Fri, Apr 4, 2014 at 6:45 AM, Jean-Marc Spaggiari <
> > > jean-m...@spaggiari.org
> > > > wrote:
> > >
> > > > If you add a coproc on the destination cluster and ask it to reject
> all
> > > the
> > > > deletes, that might do the trick, but you might end up with some
> issues
> > > at
> > > > the end if you want to do some maintenance in the target cluster...
> > > >
> > > >
> > > > 2014-04-04 4:26 GMT-04:00 冯宏华 <fenghong...@xiaomi.com>:
> > > >
> > > > > Can't figure out solution to achieve this behavior using existing
> > means
> > > > in
> > > > > HBase immediately.
> > > > >
> > > > > But it seems not that hard to implement it by changing some code, a
> > > rough
> > > > > thought is to filter out delete entries when pushing entries to the
> > > > > according replication peer and this behavior can be made
> > configurable.
> > > > > ________________________________________
> > > > > 发件人: Manthosh Kumar T [manth...@gmail.com]
> > > > > 发送时间: 2014年4月4日 16:11
> > > > > 收件人: user@hbase.apache.org
> > > > > 主题: Re: 答复: HBase Replication - Addition alone
> > > > >
> > > > > Is it possible by any other meansin HBase?
> > > > >
> > > > >
> > > > > On 4 April 2014 13:37, 冯宏华 <fenghong...@xiaomi.com> wrote:
> > > > >
> > > > > > No
> > > > > > ________________________________________
> > > > > > 发件人: Manthosh Kumar T [manth...@gmail.com]
> > > > > > 发送时间: 2014年4月4日 16:00
> > > > > > 收件人: user@hbase.apache.org
> > > > > > 主题: HBase Replication - Addition alone
> > > > > >
> > > > > > Hi All,
> > > > > > In a master-slave replication, is it possible to
> replicate
> > > > only
> > > > > > the addition of rows?. If I delete in the master it shouldn't be
> > > > deleted
> > > > > in
> > > > > > the slave.
> > > > > >
> > > > > > --
> > > > > > Cheers,
> > > > > > Manthosh Kumar. T
> > > > > >
> > > > >
> > > > >
> > > > >
> > > > > --
> > > > > Cheers,
> > > > > Manthosh Kumar. T
> > > > >
> > > >
> > >
> >
> >
> >
> > --
> > Cheers,
> > Manthosh Kumar. T
> >
>
--
Cheers,
Manthosh Kumar. T
