[Elasticsearch] Multi-server, No metric if there is a node down

[jeremy-albinet]

Hi,

My conf for elasticsearch plugin :

# Read stats from one or more Elasticsearch servers or clusters
[elasticsearch]
  # specify a list of one or more Elasticsearch servers
  servers = ["http://127.0.0.1:9200","http://127.0.0.1:9202"]

  # set local to false when you want to read the indices stats from all nodes
  # within the cluster
  local = true

Currently my node on :9200 is stopped/down, and no metric about elasticsearch are flushed but I want :9202's metrics.

I've got an error for :9200 as expected :
2015/12/26 19:57:30 Error in plugin [elasticsearch]: Get http://127.0.0.1:9200/_nodes/_local/stats: dial tcp 127.0.0.1:9200: getsockopt: connection refused

If I switch the order of servers list, I got :9202's metrics + error message for :9200 :
servers = ["http://127.0.0.1:9202","http://127.0.0.1:9200"]

I think this return breaks the process that loops over all servers : https://github.com/influxdb/telegraf/blob/master/plugins/elasticsearch/elasticsearch.go#L103

[zstyblik]

So it seems. I wanted to say that something like func escalate(old_state error, new_state error) error could be handy, but I'm not sure how this case should be handled. @sparrc?