Extract http stats from nodes stats into separate API

[dakrone]

Description

From the nodes stats API, we should extract the http stats into their own endpoint. This endpoint can then be non-node-centric, and present a global view of the number of connections, recent clients, etc.

[elasticsearchmachine]

Pinging @elastic/es-data-management (Team:Data Management)

[HiDAl]

@dakrone Some questions:

1. How the endpoint URL would look like? Since I'll be moving a couple of different APIs, I thought about a generic _stats/(http|ingest|...) endpoint, but not sure whether it's recommended according to the internal ES rules. Also, a _(http|ingest|..)/stats would work, but again not sure.
2. If I correctly understood the description of the ticket, the final response would be an union of the general status of the cluster, so basically, the response would look like:

{
  "_http": {
    "total": 1, // all the node called
    "successful": 1,
    "failed": 0
  },
  "cluster_name": "<cluster name?>",
  "http": {
    "current_open": 1,  // <- summing up all the nodes 
    "total_opened": 12, // <- summing up all the nodes 
    "clients": [ // an array merge of each node
      {
        "id": 896729347,
        "local_address": "127.0.0.1:9200",
        "remote_address": "127.0.0.1:62755",
        "last_uri": "/",
        "opened_time_millis": 1684139167185,
        "closed_time_millis": 1684139167185,
        "last_request_time_millis": 1684139167185,
        "request_count": 1,
        "request_size_bytes": 0
      },
      ...
    ]
  }
}

1. This has to be available only in the case the cluster is in serverless mode, right?

[dakrone]

I think those comments about the response body match what I was thinking. For the URL, the problem with /_stats/<anything> is that it's already reserved for index stats, so in this case perhaps just introducing a /_http_stats would be worthwhile. It's at least worth discussing as a team for the URL part (especially since we may have multiple of these, having a pattern could be useful).

This has to be available only in the case the cluster is in serverless mode, right?

I think this would be fine for any running case, i.e., both Serverless and Serverful.

[HiDAl]

thanks @dakrone !