Athena/Presto find key with the max value in JSON object - json

Question

I have a column in Athena (type string) with json like this:
{
"key1": 1.1,
"key2":2.2,
"key3": 3.3
}
How do I write a query which will return me the JSON key with the highest value (in this example it is key3) for each row and the associated value (3.3).
Note: I don't know what are the key names in advance (and there can be several)

Answer 1

You can cast your json as MAP(VARCHAR, INTEGER) and process it. For example (this uses map_entries function to turn map into array of rows, reduce array function and relies on default row naming convention) :
WITH dataset AS (
SELECT *
FROM (VALUES
(JSON '{
"key1": 1.1,
"key2":2.2,
"key3": 3.3
}'),
(JSON '{
"key0": 1.1,
"key1":4.4,
"key2": 3.3
}')) AS t (json))
SELECT row.field0 as key, row.field1 as value
FROM
(SELECT reduce(
map_entries(CAST(json as MAP(VARCHAR, INTEGER))),
ROW (null, null),
(agg, curr) -> IF (agg.field1 > curr.field1, agg, curr),
s -> s) as row
FROM dataset)
Output:
key
value
key3
3
key1
4

Answer 2

So I have found a way but it seems very convoluted, would appreciate if anyone else has a better solution. Assuming there is a column called Id, and the json is stored in a separate column:
with d as (
select id,
CAST(json_extract(json_col, '$') AS MAP(VARCHAR, VARCHAR)) as s
from TABLE_NAME
),
d2 as (
select *,
element_at(s, key) AS value
from d
cross join unnest(map_keys(s)) AS sx(key)
),
d3 as (
select id, key, value,
rank() over (partition by id order by value desc) as order
from d2
order by id, order
)
select id, key, value from d3 where order = 1
Basically first cast the JSON object into a map, then unnest the map keys and cross join and in a separate column store the value, then compute the rank partitioned by the value, then only choose those rows with rank = 1