Query to extract ids from a deeply nested json array object in Presto

If your data has a regular structure as in the example you posted, you can use a combination of parsing the value as JSON, casting it to a structured SQL type (array/map/row) and the using array processing functions to filter, transform and extract the elements you want:

WITH data(value) AS (VALUES '{
  "results": [
    {
      "docs": [
        {
          "id": "apple1",
          "source": "dd"
        },
        {
          "id": "apple2",
          "source": "aa"
        },
        {
          "id": "apple3",
          "source": "dd"
        }
      ],
      "group": 99806
    }
  ]
}'),
parsed(value) AS (
  SELECT cast(json_parse(value) AS row(results array(row(docs array(row(id varchar, source varchar)), "group" bigint))))
  FROM data
)
SELECT 
  transform(                                        -- extract the id from the resulting docs
    filter(                                         -- filter docs with source = 'dd'
        flatten(                                    -- flatten all docs arrays into a single doc array
             transform(value.results, r -> r.docs)  -- extract the docs arrays from the result array
        ),
        doc -> doc.source = 'dd'),
    doc -> doc.id)
FROM parsed

The query above produces:

      _col0
------------------
 [apple1, apple3]
(1 row)