Today when using a new API against an old cluster or using an API without the correct plugin installed can result in errors complaining about "invalid index name" even though an index parameter isn't being used. For example using the fleet.search API on 8.0-SNAPSHOT yields:

{
  "error": {
    "root_cause": [
      {
        "type": "invalid_index_name_exception",
        "reason": "Invalid index name [_fleet], must not start with '_'.",
        "index_uuid": "_na_",
        "index": "_fleet"
      }
    ],
    "type": "invalid_index_name_exception",
    "reason": "Invalid index name [_fleet], must not start with '_'.",
    "index_uuid": "_na_",
    "index": "_fleet"
  },
  "status": 400
}

This can be confusing to users as the error message points to a problem with an index name instead of an absence of the API handler.

This issue also impacts our developers: API namespaces aren't as useful for plugin APIs as they still need to consider other API names to provide positive UX in their absence.

Proposed solution

Instead of using a wildcard for <index> path parts, instead don't match path parts with a leading underscore (except _all). This will lead to errors that more closely match the situation of an API not being available instead of anything to do with index names:

{
  "error": "no handler found for uri [/_fleet/_search] and method [POST]"
}

References:

• Modify fleet search URLs to avoid url collisions #79776