Currently, reindex from remote allows to configure one single host to send http requests to. In case of failure, there are cases (e.g. 429 response code for search) where we schedule a retry for the same request against the same node. We also have a backoff policy so that bulk failures get retried too.
I wonder if it makes sense to allow reindex from remote to either use the Sniffer and send requests to multiple nodes, or allow to configure multiple nodes, or both. In case multiple nodes are available, it makes much more sense to let the RestClient do the retries on a different node, then the current custom retry code would be less needed I think.
This issue is to discuss what makes sense to do in reindex from remote and whether we need to change anything there.