New nodes do not consistently find existing master


(Jamshid) #1

I know I need to try a more recent version than 0.16.2, but just
wondering if other people notice this problem. I've seen it across
environments (eg, Ubuntu 10.04, OpenJDK 64 1.6.0_20).

I start multiple elasticsearch processes on different machines, but
even if I start one first and wait for it to fully initialize, the
other processes don't always see it as the Master. They choose
themselves as Master. This seems to happen every couple of restarts.
Usually killing the process and restarting allows it to join.

I don't know if I need to mess with any of the parameters in
http://www.elasticsearch.org/guide/reference/modules/discovery/zen.html.

Thanks,
Jamshid


(Shay Banon) #2

Are you using the default discovery, using multicast? It might be that you
need to increase the ping_timeout which controls how long a new node will
wait for a ping response when it starts up. The setting
is: discovery.zen.ping.timeout and defaults to 3s.

On Mon, Aug 1, 2011 at 2:34 AM, Jamshid jamshid69@gmail.com wrote:

I know I need to try a more recent version than 0.16.2, but just
wondering if other people notice this problem. I've seen it across
environments (eg, Ubuntu 10.04, OpenJDK 64 1.6.0_20).

I start multiple elasticsearch processes on different machines, but
even if I start one first and wait for it to fully initialize, the
other processes don't always see it as the Master. They choose
themselves as Master. This seems to happen every couple of restarts.
Usually killing the process and restarting allows it to join.

I don't know if I need to mess with any of the parameters in
http://www.elasticsearch.org/guide/reference/modules/discovery/zen.html.

Thanks,
Jamshid


(system) #3