TServer decommissioning

[Denis <de...@camfex.cz>]

Hi.

Is there a safer way to put a tserver off than by the killing its process?

I mean, something similar to decommissioning of HDFS data node, when
master knows which nodes are going to be retired and slowly moves the
files from them to other nodes.

Well, the killing must be safe as well, but the following log recovery
introduces few seconds or even minutes of data unavailability.

Thanks.

Re: TServer decommissioning

[Christopher <ct...@apache.org>]

Sorry, that last one should be "bin/accumulo admin stop $host".

--
Christopher L Tubbs II
http://gravatar.com/ctubbsii


On Tue, Feb 26, 2013 at 11:34 PM, Christopher <ct...@apache.org> wrote:
> If you're not ingesting to anything hosted by that tserver, doing a
> flush of the tables with tablets on that tserver (that info should be
> available on the monitor page), should avoid long log recoveries.
>
> You should be able to shut down a single tablet server with
> "bin/stop-here.sh" or "bin/accumulo admin $host" ($host, as it appears
> in the slaves file).
>
> --
> Christopher L Tubbs II
> http://gravatar.com/ctubbsii
>
>
> On Tue, Feb 26, 2013 at 9:17 PM, Denis <de...@camfex.cz> wrote:
>> Hi.
>>
>> Is there a safer way to put a tserver off than by the killing its process?
>>
>> I mean, something similar to decommissioning of HDFS data node, when
>> master knows which nodes are going to be retired and slowly moves the
>> files from them to other nodes.
>>
>> Well, the killing must be safe as well, but the following log recovery
>> introduces few seconds or even minutes of data unavailability.
>>
>> Thanks.

Re: TServer decommissioning

[Christopher <ct...@apache.org>]

If you're not ingesting to anything hosted by that tserver, doing a
flush of the tables with tablets on that tserver (that info should be
available on the monitor page), should avoid long log recoveries.

You should be able to shut down a single tablet server with
"bin/stop-here.sh" or "bin/accumulo admin $host" ($host, as it appears
in the slaves file).

--
Christopher L Tubbs II
http://gravatar.com/ctubbsii


On Tue, Feb 26, 2013 at 9:17 PM, Denis <de...@camfex.cz> wrote:
> Hi.
>
> Is there a safer way to put a tserver off than by the killing its process?
>
> I mean, something similar to decommissioning of HDFS data node, when
> master knows which nodes are going to be retired and slowly moves the
> files from them to other nodes.
>
> Well, the killing must be safe as well, but the following log recovery
> introduces few seconds or even minutes of data unavailability.
>
> Thanks.