The Elasticsearch client's Sniffer works in a background thread, we need to make sure this thread won't fail without the failure being at very least logged.

Should we have some form of revival policies? We should investigate to see if there are recoverable situations which make it worth to simply restart it.