Shutdown node cannot re-join the cluster

Hi, everyone.

Today, we ran into a strange problem.(maybe a configuration problem?)
We have a three node cluster, using unicast discovery :

so, in the yml file, we have those settings :

discovery.zen.ping.multicast.enabled: false

discovery.zen.ping.unicast.hosts: ["1xx.xx.xxx.x8[9300-9400]",
"1xx.xx.xxx.x9[9300-9400]", "1xx.xx.xxx.x0[9300-9400]"]

On startup, everything worked fine, the three nodes formed an unique
cluster.

Then, the master node was shut down, wich resulted in the election of a new
master in our now-two node cluster.

When we restarted the node, it elected itself as a master (that's what the
logs are telling us), and never rejoigned the cluster, leaving us with two
same-name cluster, (one with 2 nodes, one with a single one).

On the two nodes cluster logs, nothing is showing. The last entries

--
You received this message because you are subscribed to the Google Groups "elasticsearch" group.
To unsubscribe from this group and stop receiving emails from it, send an email to elasticsearch+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

Sorry, double post due to a misclick.

So, the last entries are the election of a new master after the former was
shut down.
Does anyone have an idea of what could cause that problem?
Did we miss something in the ES configuration?
Did it happen to someone else? If so, was it solved?

I know we could just restart the whole cluster, but it would hurt high
availability, so we're trying to avoid that as much a s possible.

Thanks for any insight.

Le jeudi 23 mai 2013 13:11:30 UTC+2, DH a écrit :

Hi, everyone.

Today, we ran into a strange problem.(maybe a configuration problem?)
We have a three node cluster, using unicast discovery :

so, in the yml file, we have those settings :

discovery.zen.ping.multicast.enabled: false

discovery.zen.ping.unicast.hosts: ["1xx.xx.xxx.x8[9300-9400]",
"1xx.xx.xxx.x9[9300-9400]", "1xx.xx.xxx.x0[9300-9400]"]

On startup, everything worked fine, the three nodes formed an unique
cluster.

Then, the master node was shut down, wich resulted in the election of a
new master in our now-two node cluster.

When we restarted the node, it elected itself as a master (that's what the
logs are telling us), and never rejoigned the cluster, leaving us with two
same-name cluster, (one with 2 nodes, one with a single one).

On the two nodes cluster logs, nothing is showing. The last entries

--
You received this message because you are subscribed to the Google Groups "elasticsearch" group.
To unsubscribe from this group and stop receiving emails from it, send an email to elasticsearch+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.