Make the tsdb track compatible with Elasticsearch 7.17

[salvatore-campagna]

Here we make sure we exclude things from the configuration which are incompatible with
Elasticsearch v7. This is required to be able to use the challenge and the dataset in an
upgrade test which indexes metrics data into Elasticsearch 7.17.17 in standard mode. In that version
some things like source.mode, synthetic source, time series aggregations and so on were not available.
Also we need to disable the downsampling and low latency challenges because:

• downsampling requires a time series index as a source index, which is not available in Elasticsearch v7
• the low latency challenge uses document IDs to fetch documents which are likely different in Elasticsearch v7

[gareth-ellis]

Why not just have the default to be true, the feature doesnt really have anything to do with serverless, plus that keeps the default behaviour of the track the same

[dnhatn]

@gareth-ellis I was able to rally this track with ES 7.17 with these params:

{
  "force_merge_max_num_segments": 1,
  "bulk_indexing_clients": 16,
  "corpus": "split16",
  "include_esql_queries": false,
  "include_source_mode": false,
  "skip_running_tsdb_aggs": true,
  "enable_downsample_challenge": false,
  "enable_low_latency_challenge": false
}

Can you take another look? Thank you!

[gareth-ellis]

This is being merged into main, and seems that tsdb downsample challenge now is incompatible with main -

Cannot run task [downsample-1m]: Request returned an error. Error type: api, Description: exception ({'error': {'root_cause': [{'type': 'exception', 'reason': 'Rollup requires setting [index.mode=time_series] for index [tsdb]'}], 'type': 'exception', 'reason': 'Rollup requires setting [index.mode=time_series] for index [tsdb]'}, 'status': 500}), HTTP Status: 500

[salvatore-campagna]

This is being merged into main, and seems that tsdb downsample challenge now is incompatible with main -

Cannot run task [downsample-1m]: Request returned an error. Error type: api, Description: exception ({'error': {'root_cause': [{'type': 'exception', 'reason': 'Rollup requires setting [index.mode=time_series] for index [tsdb]'}], 'type': 'exception', 'reason': 'Rollup requires setting [index.mode=time_series] for index [tsdb]'}, 'status': 500}), HTTP Status: 500

Yes:

1. downsampling was introduced in v8 and never backported
2. downsampling requires the source index to be a TSDB index. TSDB is an index mode introduced in v8 and never backported (as it requires TSDB).

So the downsampling challenge is incompatible with Elasticsearch v7 and the CI tries to run all of the challenges including the downsampling one. That is why I added enable_downsample_challenge.

The other challenge low_latency requires using specific document IDs...which are likely different in v7...that is the reason why I introduced enable_low_latency_challenge.

Anyway in the CI both challenges are executed because the default value for both enable_downsample_challenge and enable_downsample_challenge is true. Will try setting the default to false...

[salvatore-campagna]

@martijnvg should we merge this?

[martijnvg]

LGTM 👍

should we merge this?

Yes.

[gareth-ellis]

LGTM

[salvatore-campagna]

I will wait on Monday to merge this, just to avoid nightly benchmark issues affecting the TSDB track during the weekend.

[salvatore-campagna]

This is a sample esbench configuration which runs the TSDB challenge with Elasticsearch 7.17.27.

{
    "elasticsearch.version": "7.17.27",
    "track.name": "tsdb",
    "track.challenge": "append-no-conflicts",
    "on_error": "abort",
    "track.params": {
      "force_merge_max_num_segments": 1,
      "bulk_indexing_clients": 16,
      "corpus": "split16",
      "include_source_mode": false,
      "skip_running_tsdb_aggs": true,
      "enable_downsample_challenge": false,
      "enable_low_latency_challenge": false
    },
    "client.options": {
        "default": {
            "timeout": 3600
        }
    },
    "elasticsearch.preserve_install": true
  }