SQL: Fixes issues with showing value column name prefix in legends

[ryantxu]

Fixes #35390 (partially)

This is a SQL specific and back end only alternative to #35392

http://localhost:3000/d/JYola5qzz/datasource-tests-postgres?orgId=1&editPanel=4

[torkelo]

This looks good to me

[marefr]

Works in the case where you only have two fields, but not when you have more than one value fields with label metric. There's other cases when there's not a value and metric field as can be seen in documentation:

The gdev dashboard has one such example for postgres unittest dashboard:

SELECT
  $__timeGroup("time_date_time",'5m'),
  min("value_double") as "min_value",
  max("value_double") as "max_value"
FROM test_data
WHERE $__timeFilter("time_date_time")
GROUP BY time
ORDER BY time

This gdev dashboard shows that it resolves one panel, but not the other two:

For example below query generates the following frame

SELECT
  $__timeGroup("createdAt",'$summarize'),
  max(value) as "value",
  hostname as "metric"
FROM 
  grafana_metric
WHERE
  $__timeFilter("createdAt") AND
  measurement = 'cpu' AND
  hostname IN($host)
GROUP BY time, metric
ORDER BY time

{
  "response": {
    "results": {
      "A": {
        "frames": [
          {
            "schema": {
              "name": "results",
              "meta": {
                "executedQueryString": "SELECT\n  floor(extract(epoch from \"createdAt\")/60)*60 AS \"time\",\n  max(value) as \"value\",\n  hostname as \"metric\"\nFROM \n  grafana_metric\nWHERE\n  \"createdAt\" BETWEEN '2021-06-17T07:47:59.888Z' AND '2021-06-17T07:52:59.888Z' AND\n  measurement = 'cpu' AND\n  hostname IN('10.1.100.1','10.1.100.10','server1','server2')\nGROUP BY time, metric\nORDER BY time"
              },
              "fields": [
                {
                  "name": "time",
                  "type": "time",
                  "typeInfo": {
                    "frame": "time.Time"
                  }
                },
                {
                  "name": "value",
                  "type": "number",
                  "typeInfo": {
                    "frame": "float64",
                    "nullable": true
                  },
                  "labels": {
                    "metric": "server1"
                  }
                },
                {
                  "name": "value",
                  "type": "number",
                  "typeInfo": {
                    "frame": "float64",
                    "nullable": true
                  },
                  "labels": {
                    "metric": "server2"
                  }
                }
              ]
            },
            "data": {
              "values": [
                [
                  1623916080000,
                  1623916140000,
                  1623916200000,
                  1623916260000,
                  1623916320000
                ],
                [
                  101.94567116740645,
                  101.32983803008855,
                  98.24485134174023,
                  99.11082242003671,
                  98.20762076760217
                ],
                [
                  102.17495640876385,
                  101.49139611423894,
                  98.5697798472507,
                  99.34979998130595,
                  98.5910734483771
                ]
              ]
            }
          }
        ],
        "refId": "A"
      }
    }
  }
}

We need to decide whether wide data frames for SQL is the way forward. To me it is because of unified alerting and multi dimensions plus v8 is a good opportunity to do breaking change as long as we have a good workaround story. Let us know if you need help from backend platform figuring out this.

[hugohaggmark]

This solves this issue for me #35599

[torkelo]

@marefr

SELECT
$__timeGroup("time_date_time",'5m'),
min("value_double") as "min_value",
max("value_double") as "max_value"
FROM test_data
WHERE $__timeFilter("time_date_time")
GROUP BY time
ORDER BY time

This query did not work in v7.5 for me, what format have selected time series or table? The query is missing a metric field as well so the above code should not have any effect ?

We need to figure out something out really soon as v8 breaks almost all time series queries

[torkelo]

@marefr but good catch that it does not solve all those graphs on that dashboard, a small update should fix that.

[torkelo]

@marefr pushed an update that fixes it so that it works on all panels on that test dashboard, was a wrong condition that made the fix logic only apply to data frames with 2 fields

[marefr]

@torkelo

This query did not work in v7.5 for me, what format have selected time series or table? The query is missing a metric field as well so the above code should not have any effect ?

The old logic is, if there's no metric column selected, use the names of value columns as metric :)

[torkelo]

@marefr it's working now, the query below works well for me

SELECT
$__timeGroup("time_date_time",'5m'),
min("value_double") as "min_value",
max("value_double") as "max_value"
FROM test_data
WHERE $__timeFilter("time_date_time")
GROUP BY time
ORDER BY time

[torkelo]

pushed another update that refines the logic to only be applied when the orignal query returns 3 columns (so if you have more than one value field the fixing logic wont happen).
Updated a integration test as well

[marefr]

LGTM

Works great 🎉