GROUP BY in InfluxQL should produce a different response structure

[hiltontj]

Problem statement

The v1 /query API returns a different response structure when a GROUP BY is used in the query and our implementation of the /query API is not adhering to that.

Steps to reproduce

Running a v1.x InfluxDB instance, perform the following write:

curl "localhost:8086/write?db=mydb" -d "cpu,host=a usage=33"

Perform the following query:

curl "localhost:8086/query?db=mydb&q=SELECT+*+FROM+cpu+GROUP+BY+host"

This gives a response like:

{"results":[{"statement_id":0,"series":[{"name":"cpu","tags":{"host":"a"},"columns":["time","usage"],"values":[["2025-01-06T15:06:41.787035Z",33]]}]}]}

Note the tags attribute in the response.

Proposed solution

• Update the /query API handler to adhere to this response structure.
• Provide support to the InfluxQL parser to determine if the query is using a GROUP BY aggregation, and which tag is being used so we can inform the code that composes the API response (see http/v1.rs)
  • This may require upstream changes in IOx/core to expose the query syntax tree, but we are already doing some handling of that in the rewriter, so we may be able to hash out a solution here (this would need to be ported back to IOx/core at some point though)

Alternatives considered

N/A

Additional context

Slack thread.

[helenosheaa]

This may or may not be helpful but I can link you to where we are building that in our go implementation thats used in dedicated (influxqlbridge) and serverless (influxqld)

https://github.com/influxdata/idpe/blob/837a193964f591dc6a76bb8cc03f54b070c50377/influxql/ioxflight/cursor.go#L173

You can see we use the INFLUXQL_METADATA_KEY or group by key to then parse the schema to build that kind of tags, columns, values v1 response

@mhilton and I worked on this a long while back and I cant find documentation on the fact that this is the expected format but backwards compatibility. Oh I found this on the dedicated docs which sheds a bit of light

@stuartcarnie is a great resource for all things InfluxQL (he helped us out a lot when we were doing the bridge work)