Returning OxQL timeseries data as an object requires needless transformation

[wfchandler]

The data returned by OxQL is formatted along these lines, with the timeseries data being an object:

{
  "tables": [
    {
      "name": "<METRIC_NAME>",
      "timeseries": {
        "421118574608206050": {
          "fields": { <FIELDS> },
          "points": { <POINTS> },
          "values": { <VALUES> }
        },
        "637422258989466863": {
          "fields": { <FIELDS> },
          "points": { <POINTS> },
          "values": { <VALUES> }
        },
        "771960562175568387": {
          "fields": { <FIELDS> },
          "points": { <POINTS> },
          "values": { <VALUES> }
        }
      }
    }
  ]
}

The integer keys used for each timeseries object is of no meaning to consumers and must be stripped out before accessing the actual payload.

In our own tooling we are doing this here and here, and customers consuming our metrics have also needed to do the same.

Setting timeseries as an array would allow the data to be more readily consumed:

{
  "tables": [
    {
      "name": "<METRIC_NAME>",
      "timeseries": [
        {
          "fields": { <FIELDS> },
          "points": { <POINTS> },
          "values": { <VALUES> }
        },
        {
          "fields": { <FIELDS> },
          "points": { <POINTS> },
          "values": { <VALUES> }
        },
        {
          "fields": { <FIELDS> },
          "points": { <POINTS> },
          "values": { <VALUES> }
        }
      ]
    }
  ]
}

[david-crespo]

I suspect there are situations where the data cannot be transformed into the more pleasant format. I wonder if we could gate the format behind a query param, so that maybe by default it does the nice thing but 400s or something if it can't transform, and then you can pass raw=1 or whatever to get the original format. One issue with that idea is that the OpenAPI schema has to be a oneOf with the two formats, which is kind of annoying. So maybe two endpoints would be better.

[wfchandler]

/cc @bnaecker

[bnaecker]

I don't see a reason we can't just do this. Those integer keys there are definitely implementation details, and don't do anything but confuse users. They're the hash of the fields, effectively, and so I think we can just take the values as an array before schlepping this out the front door.