Hi Andreas,

As Jesse mentions, PQS is a relatively light-weight process; it's just proxying phoenix queries. It will contain a connection instance for each client and a cache of recently prepared statements. Eventually we hope to make query execution stateless, and allow clients to transparently hit a group of PQS instances via load-balancing -- they're just HTTP requests after all. At such time, I imagine we'll see folks running a PQS as just another "worker" node, running on all of or a subset of hosts running region server processes.

Good luck,
-n

On Sat, Dec 5, 2015 at 8:22 AM, Jesse Yates <jesse.k.yates@gmail.com> wrote:
I think he means its not a terribly expensive process - it is basically just a fancy query proxy. If you are running a cluster any larger than 3 nodes you should seriously consider running at least a second or third HMaster. When they are in standby mode they don't do very much - just watch ZK for the leader lock. When active they very very rarely impact the path of the client and you should be able to pretty easily ride over a hmaster failover event.

However, given that it is a pretty new component, I would personally be cautious putting it on nodes that affect the entire cluster health, like the servers hosting ZK or the NN.

Just my $0.02

On Sat, Dec 5, 2015 at 4:20 AM Andreas Fritzler <andreas.fritzler@gmail.com> wrote:
Thanks Thomas. Yes, that's the one.

What exactly do you mean by "heavy"? I was wondering if it would make sense to put it on the same host as the HBase master. On the other side I don't want the query server bring down the master machine ... 


On Fri, Dec 4, 2015 at 4:56 PM, Thomas Decaux <ebuildy@gmail.com> wrote:
You mean https://phoenix.apache.org/server.html?

I am not sure it's really heavy, so I would suggest to run it closest from HBase server.

2015-12-04 15:59 GMT+01:00 Andreas Fritzler <andreas.fritzler@gmail.com>:
Hi,

What is the best place to install the Phoenix Query Server within an HDP Hadoop cluster? Is it a good idea to place it on one of the master machines? Or would your rather suggest to have a dedicated host for that?

Also any suggestion/best practices on how to load balance multiple instances of the query server?

Thanks a lot in advance!

Regards,
Andreas