Elasticsearch is not spreading indexing load across all hosts in a cluster

Summary

Detected in the course of https://gitlab.com/gitlab-com/infrastructure/issues/2157#note_34814771

Steps to reproduce

Start indexing against a multi-node elasticsearch cluster

What is the current bug behavior?

PUT requests only go to the first node in elasticsearch_urls.

What is the expected correct behavior?

PUT requests should go to all nodes.

Possible fixes

While addressing https://gitlab.com/gitlab-org/gitlab-ee/issues/2681 we discussed the possibility that we might need to add reload_connections: true to the configuration. However, since we're explicitly specifying several nodes, I don't think this is the answer here.

/cc @victorcete