Quantcast

MiniAccumuloCluster connection timeout negotiation

classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

MiniAccumuloCluster connection timeout negotiation

Jeff N
Is there a means of truly changing the connection timeout? I know that the constructor for the ZooKeeperInstance has a parameter for session timeout, but I keep having my client session timeout, sometimes in as little 13 seconds. Is there any way to extend the connection timeout?
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: MiniAccumuloCluster connection timeout negotiation

Eric Newton
You can change the timeout, but zookeeper will only let you use a
connection timeout of 20x the tick rate.  By default, the tick rate for MAC
is 1s.  There is presently no way to change this value.

Also, I seem to remember that the client library sends a ping 1/3 of the
way through the timeout, and will disconnect at 2/3rds the timeout without
a response in order to attempt to reconnect to a more responsive zookeeper
before losing its session.

A connection timeout may be causing servers to fail too soon, see
ACCUMULO-1572.

Ensure your processes are not getting pushed into swap, and that you have
swappiness turned off.

-Eric


On Tue, Aug 6, 2013 at 1:07 PM, Mastergeek <[hidden email]> wrote:

> Is there a means of truly changing the connection timeout? I know that the
> constructor for the ZooKeeperInstance has a parameter for session timeout,
> but I keep having my client session timeout, sometimes in as little 13
> seconds. Is there any way to extend the connection timeout?
>
>
>
> --
> View this message in context:
> http://apache-accumulo.1065345.n5.nabble.com/MiniAccumuloCluster-connection-timeout-negotiation-tp5049.html
> Sent from the Developers mailing list archive at Nabble.com.
>
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: MiniAccumuloCluster connection timeout negotiation

Jeff N
I did some looking around and I have not found any information on turning swappiness off. I also did notice that in my log files this line is always popping up:

13/08/06 13:29:58 WARN server.Accumulo: System swappiness setting is greater than ten (60) which can cause time-sensitive operations to be delayed.  Accumulo is time sensitive because it needs to maintain distributed lock agreement.

Thanks,
Jeff
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: MiniAccumuloCluster connection timeout negotiation

Sean Busbey
Accumulo, ZK, Hadoop, and HBase will all have terrible performance if
swapping starts happening. The first step is to make sure you allocate an
amount of memory appropriate to your hardware. This is covered briefly in
the Accumulo manual:

http://accumulo.apache.org/1.5/accumulo_user_manual.html#_edit_conf_accumulo_env_sh


The second is to make sure your kernel won't start swapping stuff out
early. That's controlled via a setting called vm.swappiness. You can see
how to check vm.swappiness and turn it to 0 in the CDH guides, e.g. CDH3:

http://www.cloudera.com/content/cloudera-content/cloudera-docs/CDH3/CDH3u6/CDH3-Installation-Guide/cdh3ig_topic_9_6.html



On Tue, Aug 6, 2013 at 12:35 PM, Mastergeek <[hidden email]> wrote:

> I did some looking around and I have not found any information on turning
> swappiness off. I also did notice that in my log files this line is always
> popping up:
>
> 13/08/06 13:29:58 WARN server.Accumulo: System swappiness setting is
> greater
> than ten (60) which can cause time-sensitive operations to be delayed.
> Accumulo is time sensitive because it needs to maintain distributed lock
> agreement.
>
> Thanks,
> Jeff
>
>
>
> -----
>
>
>
> --
> View this message in context:
> http://apache-accumulo.1065345.n5.nabble.com/MiniAccumuloCluster-connection-timeout-negotiation-tp5049p5057.html
> Sent from the Developers mailing list archive at Nabble.com.
>



--
Sean
Loading...