Environment:
- 2 VMs, each with 2 ip addresses (interfaces enp0s3 & enp0s8) called
node1 and node3
- ES master on node1, data node on node3
- CentOS 7
For some reason, elasticsearch uses the ip attached to enp0s3 as it's
publish address. Due to the way my test environment is set up, this will
not work. ES should use the ip address of enp0s8. However, we've been
trying to debug this over irc and nothing seems to work in Ambari except
setting network.publish_host manually in elasticsearch.yml on node1 &
node3 and restarting 'manually'. Unfortunately, this setting will be
gone from both hosts when I restart via Ambari.
We've tried the following in Ambari:
- network.host: 0.0.0.0 => Wrong ip used (enp0s3).
- network.host: "_lo:ipv4_","_enp0s8:ipv4_" => This blows up Java
- network.host: "_lo:ipv4","_enp0s8:ipv4" or ["_lo:ipv4","_enp0s8:ipv4"]
=> Unknown host exception in Java.
Any idea how I can add network.publish_host to Ambari so that it doesn't
get removed when I restart the services via Ambari? Or what setting I
need to use for network.host?
