Apache Druid GroupBy Virtual columns

Question

I am trying to do a groupby virtual column in a Druid native query which looks like this...

{
  "queryType": "groupBy",
  "dataSource": "trace_info",
  "granularity": "none",
  "virtualColumns": [
    {
      "type": "expression",
      "name": "tenant",
      "expression": "replace(array_offset(tags, array_offset_of(tagNames, 'tenant')), 'tenant:', '')"
    },
    {
      "type": "expression",
      "name": "rc",
      "expression": "replace(array_offset(tags, array_offset_of(tagNames, 'row_count')), 'row_count:', '')"
    }
  ],
  "dimensions": [
    "tenant"
  ],
  "aggregations": [
    {
      "type": "longSum",
      "name": "trc",
      "fieldName": "rc"
    }
  ],

...
...
...

  "intervals": [
    "..."
  ]
}

This gives out a single row with longsum of all row_counts as if the groupBy column is null.

Is my usage correct or is this a known issue in Druid. The documentation says virtual columns can be used like normal dimensions but, is not very clear on how or even a working example is missing.

Thanks! Phani

Answer 1

Latest Edit...

Some more digging to find out that the issue was with missing "outputType" attributes on the the virtual columns. Strange because the aggregator is able to auto-detect time and calculate the long sum properly even though the group by results were wrong.

  "virtualColumns": [
    {
      "type": "expression",
      "name": "tenant",
      "expression": "replace(array_offset(tags, array_offset_of(tagNames, 'tenant')), 'tenant:', '')",
      "outputType": "STRING"
    },
    {
      "type": "expression",
      "name": "rc",
      "expression": "replace(array_offset(tags, array_offset_of(tagNames, 'row_count')), 'row_count:', '')"
      "outputType": "LONG"
    }
  ],

See above (below is likely a non-performant way of working around the problem).

After some trial and error I have a workaround for this using extraction dimensions. Although not sure, I suspect that this is a temporary issue in Druid 0.18.1. Hopefully Grouping on VCs will work as advertised in future builds.

{
  "queryType": "groupBy",
  "dataSource": "trace_info",
  "granularity": "none",
  "virtualColumns": [
    {
      "type": "expression",
      "name": "tenant",
      "expression": "replace(array_offset(tags, array_offset_of(tagNames, 'tenant')), 'tenant:', '')"
    },
    {
      "type": "expression",
      "name": "rc",
      "expression": "replace(array_offset(tags, array_offset_of(tagNames, 'row_count')), 'row_count:', '')"
    }
  ],
  "dimensions": [
    {
      "type": "extraction",
      "dimension": "tenant",
      "outputName": "t",
      "extractionFn": {
        "type" : "substring", "index" : 1
      }
    }
  ],
  "aggregations": [
    {
      "type": "longSum",
      "name": "trc",
      "fieldName": "rc"
    }
  ],

...
...
...

  "intervals": [
    "..."
  ]
}