dedicated node/cluster for one of the hbase tables

[T Vinod Gupta <tv...@readypulse.com>]

hi,
in a distributed hbase setup with multiple (just a few) tables, is it
possible to assign a table to a specific node or set of nodes (region
servers)?

the scenario im trying to address is that - i have a process that accesses
only 1 table and that process has very high requirements for throughput and
SLAs to meet. So if i want to isolate it from the other tables that may
have heavy transactions (or bursts of transactions) happening, then whats
the best approach?

one way of accomplishing is to have a separate hbase setup for this
application. but then it means managing different installations and
servicing them. also, the application uses other tables to load data into
this key table that is supporting api request traffic. so the app will have
to talk to 2 different zoo keepers. not the ideal solution it seems.

thanks

Re: dedicated node/cluster for one of the hbase tables

[Vandana Ayyalasomayajula <av...@yahoo-inc.com>]

Hi Vinod, 

You can take a look at the region server groups feature (https://issues.apache.org/jira/browse/HBASE-6721). 
Its still under development/review though. 

Thanks
Vandana

On Feb 4, 2013, at 2:38 PM, T Vinod Gupta wrote:

> hi,
> in a distributed hbase setup with multiple (just a few) tables, is it
> possible to assign a table to a specific node or set of nodes (region
> servers)?
> 
> the scenario im trying to address is that - i have a process that accesses
> only 1 table and that process has very high requirements for throughput and
> SLAs to meet. So if i want to isolate it from the other tables that may
> have heavy transactions (or bursts of transactions) happening, then whats
> the best approach?
> 
> one way of accomplishing is to have a separate hbase setup for this
> application. but then it means managing different installations and
> servicing them. also, the application uses other tables to load data into
> this key table that is supporting api request traffic. so the app will have
> to talk to 2 different zoo keepers. not the ideal solution it seems.
> 
> thanks

Re: dedicated node/cluster for one of the hbase tables

[Jean-Marc Spaggiari <je...@spaggiari.org>]

Hi Vinod,

There is no "standard" way to do that since it's not a standard way to
use HBase. So you will need to implement your own load balancer class.
Take a look at the org.apache.hadoop.hbase.master.DefaultLoadBalancer
class. Just do you own class implementing LoadBalancer. Take most of
the code from the default class, and add some lines to handle your
specific use case?

JM

2013/2/4, T Vinod Gupta <tv...@readypulse.com>:
> is there a standard way to do so? are there any examples?
>
> thanks
>
>
> On Mon, Feb 4, 2013 at 3:00 PM, Jean-Marc Spaggiari
> <jean-marc@spaggiari.org
>> wrote:
>
>> Hi Vinod,
>>
>> Another way can be to extend the existing load balancer and to add the
>> logic to spread your table only on the servers you want?
>>
>> JM
>>
>> 2013/2/4, T Vinod Gupta <tv...@readypulse.com>:
>> > hi,
>> > in a distributed hbase setup with multiple (just a few) tables, is it
>> > possible to assign a table to a specific node or set of nodes (region
>> > servers)?
>> >
>> > the scenario im trying to address is that - i have a process that
>> accesses
>> > only 1 table and that process has very high requirements for throughput
>> and
>> > SLAs to meet. So if i want to isolate it from the other tables that may
>> > have heavy transactions (or bursts of transactions) happening, then
>> > whats
>> > the best approach?
>> >
>> > one way of accomplishing is to have a separate hbase setup for this
>> > application. but then it means managing different installations and
>> > servicing them. also, the application uses other tables to load data
>> > into
>> > this key table that is supporting api request traffic. so the app will
>> have
>> > to talk to 2 different zoo keepers. not the ideal solution it seems.
>> >
>> > thanks
>> >
>>
>

Re: dedicated node/cluster for one of the hbase tables

[T Vinod Gupta <tv...@readypulse.com>]

is there a standard way to do so? are there any examples?

thanks


On Mon, Feb 4, 2013 at 3:00 PM, Jean-Marc Spaggiari <jean-marc@spaggiari.org
> wrote:

> Hi Vinod,
>
> Another way can be to extend the existing load balancer and to add the
> logic to spread your table only on the servers you want?
>
> JM
>
> 2013/2/4, T Vinod Gupta <tv...@readypulse.com>:
> > hi,
> > in a distributed hbase setup with multiple (just a few) tables, is it
> > possible to assign a table to a specific node or set of nodes (region
> > servers)?
> >
> > the scenario im trying to address is that - i have a process that
> accesses
> > only 1 table and that process has very high requirements for throughput
> and
> > SLAs to meet. So if i want to isolate it from the other tables that may
> > have heavy transactions (or bursts of transactions) happening, then whats
> > the best approach?
> >
> > one way of accomplishing is to have a separate hbase setup for this
> > application. but then it means managing different installations and
> > servicing them. also, the application uses other tables to load data into
> > this key table that is supporting api request traffic. so the app will
> have
> > to talk to 2 different zoo keepers. not the ideal solution it seems.
> >
> > thanks
> >
>

Re: dedicated node/cluster for one of the hbase tables

[Jean-Marc Spaggiari <je...@spaggiari.org>]

Hi Vinod,

Another way can be to extend the existing load balancer and to add the
logic to spread your table only on the servers you want?

JM

2013/2/4, T Vinod Gupta <tv...@readypulse.com>:
> hi,
> in a distributed hbase setup with multiple (just a few) tables, is it
> possible to assign a table to a specific node or set of nodes (region
> servers)?
>
> the scenario im trying to address is that - i have a process that accesses
> only 1 table and that process has very high requirements for throughput and
> SLAs to meet. So if i want to isolate it from the other tables that may
> have heavy transactions (or bursts of transactions) happening, then whats
> the best approach?
>
> one way of accomplishing is to have a separate hbase setup for this
> application. but then it means managing different installations and
> servicing them. also, the application uses other tables to load data into
> this key table that is supporting api request traffic. so the app will have
> to talk to 2 different zoo keepers. not the ideal solution it seems.
>
> thanks
>