Python data source: execute_query with GA4 returns JSON string (not dict), breaking existing scripts with string indices must be integers

[netgineer]

Description

After upgrading Redash from 10.x to 25.8.0, Python data source queries that call execute_query on a GA4 data source started failing with:

Error running query: <class 'TypeError'> string indices must be integers

Previously, the same Python scripts worked without changes.

The root cause appears to be that execute_query("ga4-career", ...) now returns a JSON string instead of a dict with "columns" and "rows", which breaks existing code that expects res["rows"].

Environment

Redash version: 25.8.0 (Docker image redash/redash:25.8.0)
Data source: google_analytics4 (GA4)
Python data source (Redash built-in)

Example Python Query (simplified)

ga_query = """
{
  "dimensions": [
    { "name": "date" },
    { "name": "customUser:user_type" }
  ],
  "metrics": [
    { "name": "screenPageViews" }
  ],
  "dateRanges": [
    {
      "startDate": "{{period.start}}",
      "endDate": "{{period.end}}"
    }
  ],
  "dimensionFilter": {
    "filter": {
      "fieldName": "unifiedPagePathScreen",
      "stringFilter": {
        "matchType": "CONTAINS",
        "value": "/salaries"
      }
    }
  },
  "orderBys": [
    {
      "desc": true,
      "dimension": { "dimensionName": "date" }
    }
  ]
}
"""

res = execute_query("ga4-career", ga_query)

for r in res["rows"]:
    # some processing...
    pass

This code used to work on Redash 10.x.

Expected Behavior (old behavior)

execute_query("ga4-career", ga_query) should return a Python object like:

{
  "columns": [...],
  "rows": [
    { "date": "...", "customUser:user_type": "...", "screenPageViews": "123" },
    ...
  ]
}

Then for r in res["rows"]: iterates over rows without errors.

Actual Behavior (25.8.0)

The same code now fails with:

Error running query: <class 'TypeError'> string indices must be integers

Inspection shows that res is now a string, containing JSON:

res == '{"columns": [...], "rows": [...]}'

Trying to access res["rows"] on a string triggers TypeError: string indices must be integers.

Workaround

To restore functionality, I had to wrap execute_query results in a custom normalizer, e.g.:

def normalize_result(res):
    import json

    # Handle (data, error) tuple
    if isinstance(res, tuple) and len(res) == 2:
        data, error = res
        if error:
            raise Exception(error)
        res = data

    # If GA4 returns JSON string
    if isinstance(res, str):
        res = json.loads(res)

    # If format is {"query_result": {...}}
    if isinstance(res, dict) and "query_result" in res:
        res = res["query_result"]["data"]

    return res

raw = execute_query("ga4-career", ga_query)
res = normalize_result(raw)

for r in res["rows"]:
    ...

This suggests that the contract of execute_query for GA4 changed (or became inconsistent), while existing Python scripts rely on the older behavior where it returned a dict with "columns" and "rows".

Why This Is a Problem

Existing Python data source queries that worked before the upgrade now fail without any changes to user code.

The behavior of execute_query appears to differ between data sources and between versions, which makes it hard to write robust Python queries that work across upgrades.

It would be very helpful if:

• execute_query for GA4 returned a consistent structured object (dict with "columns" and "rows", or a documented shape), or
• The change of return type was documented and/or wrapped so that Python data source sees a consistent format.