I am running into kind of the same problem.
I have a cluster with 3 nodes with automated master election. Preferably I do not bring the cluster down when installing marvel-agent. Is that even possible?
Whenever I install license and marvel-agent on one of the nodes, it cannot connect to master anymore. When I remove the plugins and restart the node, everything is OK again.
The node is giving this error:
[2016-02-18 13:00:50,037][INFO ][cluster.service ] [es02-ams01-tip] detected_master {es03-ams01-tip}{H0_Dfsv9RLOGNFsS2ARshQ}{10.10.30.3}{10.10.30.3:9300}, added {{es03-ams01-tip}{H0_Dfsv9RLOGNFsS2ARshQ}{10.10.30.3}{10.10.30.3:9300},{es01-ams01-tip}{7XJd-Dk6SfiINKM2t4xQKg}{10.10.30.1}{10.10.30.1:9300},}, reason: zen-disco-receive(from master [{es03-ams01-tip}{H0_Dfsv9RLOGNFsS2ARshQ}{10.10.30.3}{10.10.30.3:9300}])
[2016-02-18 13:00:50,055][INFO ][cluster.routing.allocation.decider] [es02-ams01-tip] updating [cluster.routing.allocation.enable] from [ALL] to [NONE]
[2016-02-18 13:00:50,056][INFO ][indices.store ] [es02-ams01-tip] updating indices.store.throttle.max_bytes_per_sec from [10gb] to [150mb], note, type is [NONE]
[2016-02-18 13:00:50,085][INFO ][discovery.zen ] [es02-ams01-tip] master_left [{es03-ams01-tip}{H0_Dfsv9RLOGNFsS2ARshQ}{10.10.30.3}{10.10.30.3:9300}], reason [transport disconnected]
[2016-02-18 13:00:50,087][INFO ][discovery.zen ] [es02-ams01-tip] failed to send join request to master [{es03-ams01-tip}{H0_Dfsv9RLOGNFsS2ARshQ}{10.10.30.3}{10.10.30.3:9300}], reason [NodeDisconnectedException[[es03-ams01-tip][10.10.30.3:9300][internal:discovery/zen/join] disconnected]]
[2016-02-18 13:00:50,422][WARN ][discovery.zen ] [es02-ams01-tip] master left (reason = transport disconnected), current nodes: {{es02-ams01-tip}{fLytB-HUS5eoYctNciIuSQ}{10.10.30.2}{10.10.30.2:9300},{es01-ams01-tip}{7XJd-Dk6SfiINKM2t4xQKg}{10.10.30.1}{10.10.30.1:9300},}
[2016-02-18 13:00:50,423][INFO ][cluster.service ] [es02-ams01-tip] removed {{es03-ams01-tip}{H0_Dfsv9RLOGNFsS2ARshQ}{10.10.30.3}{10.10.30.3:9300},}, reason: zen-disco-master_failed ({es03-ams01-tip}{H0_Dfsv9RLOGNFsS2ARshQ}{10.10.30.3}{10.10.30.3:9300})
[2016-02-18 13:00:50,441][WARN ][cluster.service ] [es02-ams01-tip] failed to notify ClusterStateListener
java.lang.IllegalStateException: master not available when registering auto-generated license
And the master is throwing this error:
[2016-02-18 13:00:50,619][DEBUG][transport.netty ] [es03-ams01-tip] connected to node [{es02-ams01-tip}{fLytB-HUS5eoYctNciIuSQ}{10.10.30.2}{10.10.30.2:9300}]
[2016-02-18 13:00:50,711][DEBUG][cluster.service ] [es03-ams01-tip] processing [zen-disco-join(join from node[{es02-ams01-tip}{fLytB-HUS5eoYctNciIuSQ}{10.10.30.2}{10.10.30.2:9300}])]: execute
[2016-02-18 13:00:50,711][DEBUG][cluster.service ] [es03-ams01-tip] cluster state updated, version [791], source [zen-disco-join(join from node[{es02-ams01-tip}{fLytB-HUS5eoYctNciIuSQ}{10.10.30.2}{10.10.30.2:9300}])]
[2016-02-18 13:00:50,711][INFO ][cluster.service ] [es03-ams01-tip] added {{es02-ams01-tip}{fLytB-HUS5eoYctNciIuSQ}{10.10.30.2}{10.10.30.2:9300},}, reason: zen-disco-join(join from node[{es02-ams01-tip}{fLytB-HUS5eoYctNciIuSQ}{10.10.30.2}{10.10.30.2:9300}])
[2016-02-18 13:00:50,711][DEBUG][cluster.service ] [es03-ams01-tip] publishing cluster state version [791]
[2016-02-18 13:00:50,780][WARN ][transport.netty ] [es03-ams01-tip] exception caught on transport layer [[id: 0x10065a67, /10.10.30.2:59175 => /10.10.30.3:9300]], closing connection
java.lang.IllegalStateException: Message not fully read (request) for requestId [10], action [internal:plugin/license/cluster/register_trial_license], readerIndex [75] vs expected [75]; resetting
It seems that I have to install the plugins on all nodes and then bring the cluster down completely and then up again one-by-one. But, is there another way?