To parse JSON consider using the specialized json functions:
SELECT
toInt32(column_values[1]) AS I2,
toInt32(column_values[2]) AS I3,
column_values[3] AS I8
FROM
(
SELECT
arrayJoin(JSONExtract(json, 'DataSets', 'Array(Array(Tuple(Int32, String)))')) AS row,
arraySort(x -> (x.1), row) AS row_with_sorted_columns,
arrayMap(x -> (x.2), row_with_sorted_columns) AS column_values
FROM
(
SELECT '{"AgentID":"10.1.8.1", "Header":{"Version":9,"Count":2}, "DataSets":[\n [{"I":3,"V":"151"},{"I":8,"V":"109.195.122.130"},{"I":2,"V":"231"}],\n [{"I":2,"V":"341"},{"I":3,"V":"221"},{"I":8,"V":"109.195.122.233"}]]}' AS json
)
)
/*
┌─I2──┬─I3──┬─I8──────────────┐
│ 231 │ 151 │ 109.195.122.130 │
│ 341 │ 221 │ 109.195.122.233 │
└─────┴─────┴─────────────────┘
*/
(To get more about JSON parsing see How to extract json from json in clickhouse?)
The implementation above relies on the fixed structure of Datasets-array. As I understood in the real world this structure has an arbitrary schema (https://www.iana.org/assignments/ipfix/ipfix.xhtml), such as:
{
"AgentID":"192.168.21.15",
"Header":{},
"DataSets":[
[
{"I":8, "V":"192.16.28.217"},
{"I":12, "V":"180.10.210.240"},
{"I":5, "V":2},
{"I":4, "V":6},
{"I":7, "V":443},
{"I":6, "V":"0x10"}
]
]
}
Thus it appears the question about a table with arbitrary columns count. ClickHouse doesn't support this feature - see how possible to present the table in this case https://stackoverflow.com/search?q=%5Bclickhouse%5D+pivot.