CODEWITHSUNDEEP
arraysjsonpyspark
Craig
Craig

Reputation: 1985

PySpark - RDD to JSON

I have a Hive query that returns data in this format:

ip, category, score
1.2.3.4, X, 5
10.10.10.10, A, 2
1.2.3.4, Y, 2
12.12.12.12, G, 10
1.2.3.4, Z, 9
10.10.10.10, X, 3

In PySpark, I get this via hive_context.sql(my_query).rdd

Each ip address can have multiple scores (hence multiple rows). I would like to get this data in a json/array format as follows:

{
    "ip": "1.2.3.4",
    "scores": [
        {
            "category": "X",
             "score": 10
        },
        {
            "category": "Y",
             "score": 2
        },
        {
            "category": "Z",
             "score": 9
        },
    ],
    "ip": "10.10.10.10",
    "scores": [
        {
            "category": "A",
             "score": 2
        },
        {
            "category": "X",
             "score": 3
        },
    ],
     "ip": "12.12.12.12",
    "scores": [
        {
            "category": "G",
             "score": 10
        },
    ],
}

Note that the RDD isn't necessarily sorted and the RDD can easily contain a couple of hundred million rows. I'm new to PySpark so any pointers on how to go about this efficiently would help.

Upvotes: 2

Views: 1571

Answers (1)

akuiper
akuiper

Reputation: 214967

groupBy ip and then transform the grouped RDD to what you needed:

rdd.groupBy(lambda r: r.ip).map(
  lambda g: {
    'ip': g[0], 
    'scores': [{'category': x['category'], 'score': x['score']} for x in g[1]]}
).collect()

# [{'ip': '1.2.3.4', 'scores': [{'category': 'X', 'score': 5}, {'category': 'Y', 'score': 2}, {'category': 'Z', 'score': 9}]}, {'ip': '12.12.12.12', 'scores': [{'category': 'G', 'score': 10}]}, {'ip': '10.10.10.10', 'scores': [{'category': 'A', 'score': 2}, {'category': 'X', 'score': 3}]}]

Upvotes: 2

Related Questions