[Feature Request] Field level statistics of lucene index files

[rishabhmaurya]

Is your feature request related to a problem? Please describe

It seems impossible to get statistics such as disk consumption of individual lucene segment files per field. We do have a similar API which can aggregate statistics at shard, index and cluster level using index stats API (https://opensearch.org/docs/latest/api-reference/index-apis/stats/). Like -

/<index_name>/_stats/segments?level=shards&include_segment_file_sizes&pretty"

  "file_sizes" : {
    "nvm" : {
      "size_in_bytes" : 5486,
      "description" : "Norms"
    },
    "fnm" : {
      "size_in_bytes" : 127478,
      "description" : "Fields"
    },
    "kdd" : {
      "size_in_bytes" : 1119640726,
      "description" : "Others"
    },
    "tmd" : {
      "size_in_bytes" : 59232,
      "description" : "Others"
    },
    "fdm" : {
      "size_in_bytes" : 9175,
      "description" : "Others"
    },
    "kdi" : {
      "size_in_bytes" : 2926068,
      "description" : "Others"
    },
    "dvd" : {
      "size_in_bytes" : 1687766934,
      "description" : "DocValues"
    },
    "kdm" : {
      "size_in_bytes" : 10398,
      "description" : "Others"
    },
    "pos" : {
      "size_in_bytes" : 6051226,
      "description" : "Positions"
    },
    "si" : {
      "size_in_bytes" : 1176,
      "description" : "Segment Info"
    },
    "fdt" : {
      "size_in_bytes" : 9388206796,
      "description" : "Field Data"
    },
    "doc" : {
      "size_in_bytes" : 2964755700,
      "description" : "Frequencies"
    },
    "tim" : {
      "size_in_bytes" : 483838256,
      "description" : "Term Dictionary"
    },
    "dvm" : {
      "size_in_bytes" : 117015,
      "description" : "DocValues"
    },
    "tip" : {
      "size_in_bytes" : 20778232,
      "description" : "Term Index"
    },
    "fdx" : {
      "size_in_bytes" : 519052,
      "description" : "Field Index"
    },
    "nvd" : {
      "size_in_bytes" : 1534,
      "description" : "Norms"
    }
  }

Describe the solution you'd like

Introduce similar API for query param to existing index stats API to provide this information at field level which can be aggregated at shard, index and cluster level per field.
This would be useful in understanding usage statistics per field. There is no way other than writing script to read lucene indexes and compute this information.

Related component

Other

Describe alternatives you've considered

No response

Additional context

No response

[rishabhmaurya]

It would be useful to also show additional information for numeric field like min and max value per segment, shard, index etc.
One utility could be issues like opensearch-project/opensearch-benchmark#398 where it can be used to understand what range of a values a segment contains.

[bbarani]

+1. Adding a disk usage stats API to provide usage of each field would help surface the metrics for Benchmarking as well. We can add it as a telemetry devices to OSB. CC: @rishabh6788

[peternied]

[Triage - attendees 1 2 3 4]
@rishabhmaurya Thanks for filing, looking forward to seeing this developed.

[msfroh]

1. We need to decide on the possible API -- proposing something is probably a good first issue, but we would need to get consensus.
2. @lukas-vlcek called out that remote store may make this more challenging (needing to fetch files from remote store).
3. We may want to calculate when writing segments.
4. Also, when addressing this, we should do something about all the files labeled Other, since we can provide better descriptions.