[PROPOSAL] Define and use a standard way for returning 404 responses

[Jakob3xD]

Is your feature request related to a problem? Please describe.
Opensearch has many different user API endpoints. Some of them behave similar and some doesn't.
During the refactoring of the opensearch-go lib, I notices that 404 response bodys are different between API endpoints.

Examples for some get requests.

GET test
{
  "error": {
    "root_cause": [
      {
        "type": "index_not_found_exception",
        "reason": "no such index [test]",
        "index": "test",
        "resource.id": "test",
        "resource.type": "index_or_alias",
        "index_uuid": "_na_"
      }
    ],
    "type": "index_not_found_exception",
    "reason": "no such index [test]",
    "index": "test",
    "resource.id": "test",
    "resource.type": "index_or_alias",
    "index_uuid": "_na_"
  },
  "status": 404
}

GET _index_template/test
{
  "error": {
    "root_cause": [
      {
        "type": "resource_not_found_exception",
        "reason": "index template matching [test] not found"
      }
    ],
    "type": "resource_not_found_exception",
    "reason": "index template matching [test] not found"
  },
  "status": 404
}

GET _scripts/test
{
  "_id": "test",
  "found": false
}

GET _template/test
{}

GET _ingest/pipeline/test
{}

GET test/_explain/3
{"query":{"term":{"foo":{"value":"bar"}}}}

{
  "_index": "test",
  "_id": "3",
  "matched": false
}

Describe the solution you'd like
I would like to see a response similar to 404 responses from index or _index_templates.

{
  "error": {
    "root_cause": [
      {
        "type": "resource_not_found_exception",
        "reason": "<resource> matching [test] not found"
      }
    ],
    "type": "resource_not_found_exception",
    "reason": "<resource> matching [test] not found"
  },
  "status": 404
}

Describe alternatives you've considered
Depending on the endpoint it would also be an option to return a 200 with a body containing no results like the _nodes endpoint does. This would for example work for the _index_template and _component_template as they return an array of templates.

GET _nodes/test
{
  "_nodes": {
    "total": 0,
    "successful": 0,
    "failed": 0
  },
  "cluster_name": "fsn1-staging-opensearch1",
  "nodes": {}
}

Additional context
This is a bit in relation to opensearch-project/opensearch-plugins#215 as it is about errors but for plugins. AFAIK Opensearch has its standard way for errors but not one for when to return it.

[mgodwan]

@shwetathareja @Bukhtawar Thoughts on this?

[dblock]

I like standardization, but I wonder what the blast radius is from a change like this. OpenSearch API has a lot of inconsistencies such as this one. I would better about a big backwards incompatible change in a major release if we completed the spec in https://github.com/opensearch-project/opensearch-api-specification, then made a version of it that removes all the inconsistencies, then generated the server code from it to ensure it's always consistent.

[imvtsl]

I will attend triage meeting with core team tomorrow to see how to go about this.

[vikasvb90]

@imvtsl / @peternied / other attendees : Can you please summarize the discussion held on this issue in the core triage meeting?

[imvtsl]

@imvtsl / @peternied / other attendees : Can you please summarize the discussion held on this issue in the core triage meeting?

We didn't really conclude on it. The feasibility is uncertain due to the large number of APIs and the potential impact of changing the 404 response for all of them. Additionally, we lack documentation on the current error responses for these APIs, which further complicates the assessment.

[dblock]

I think a big chunk of work as a prerequisite of changing any responses is opensearch-project/opensearch-api-specification#445. It would let us actually assess what APIs behave in ways we like and which ones don't, and scope the change.