incubator-hcatalog-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Andrew Mains <andrew.ma...@kontagent.com>
Subject Re: webhcat: running callbacks when one webhcat server goes down.
Date Thu, 02 May 2013 01:43:01 GMT
> Make sure that you are accessing your servers using a common url.
>   For example, you might have a load balancer at
>   http://webhcat.mydomain.com that points to your http://server1 and
>   http://server2

This is almost certainly the problem with my testing setup--they were being
accessed with distinct urls. I'll give it another shot with things set up
properly.

Thanks very much for the quick response!

Andrew Mains


On Wed, May 1, 2013 at 6:22 PM, Chris Dean <ctdean@sokitomi.com> wrote:

> Andrew Mains <andrew.mains@kontagent.com> writes:
> > 1. Start up two webhcat instances, server1 and server2.
> > 2. Submit a query to server1 with a callback.
> > 3. Kill server1
> >
> > The callback was never received in this case.
> >
> > Is this the expected behavior?
>
> That is not the expected behavior.  If you wish to setup multiple
> webhcat servers here are a few things to check:
>
> 1. Configure the storage engine used to store the job information.  The
>    config vars are templeton.storage.class and templeton.storage.root
>
>    The defaults are fine, but if you use HDFS (the default) make sure
>    that the webhcat servers are actually writing to
>    templeton.storage.root (which is "/templeton-hadoop" by default).
>
> 2. Make sure that you are accessing your servers using a common url.
>    For example, you might have a load balancer at
>    http://webhcat.mydomain.com that points to your http://server1 and
>    http://server2
>
>    The common url must also be accessible from the JobTracker.
>
>    The common url is needed because the JobTracker notifies webhcat that
>    the job has been completed by accessing the webhcat server using this
>    url.
>
> I'm not sure if this is documented anywhere and I'm afraid the
> information logged is of little help.
>
> Cheers,
> Chris Dean
>
>

Mime
View raw message