_corrupt_record error when reading a JSON file into Spark

Question

I\'ve got this JSON file

{
    "a": 1, 
    "b": 2
}

which has been obtained with Python json.dump method. Now, I want to

Answer 1

You need to have one json object per row in your input file, see http://spark.apache.org/docs/latest/api/python/pyspark.sql.html#pyspark.sql.DataFrameReader.json

If your json file looks like this it will give you the expected dataframe:

{ "a": 1, "b": 2 }
{ "a": 3, "b": 4 }

....
df.show()
+---+---+
|  a|  b|
+---+---+
|  1|  2|
|  3|  4|
+---+---+

Answer 2

In Spark 2.2+ you can read json file of multiline using following command.

val dataframe = spark.read.option("multiline",true).json( " filePath ")

if there is json object per line then,

val dataframe = spark.read.json(filepath)

Answer 3

Adding to @Bernhard's great answer

# original file was written with pretty-print inside a list
with open("pretty-printed.json") as jsonfile:
    js = json.load(jsonfile)      

# write a new file with one object per line
with open("flattened.json", 'a') as outfile:
    for d in js:
        json.dump(d, outfile)
        outfile.write('\n')

Answer 4

If you want to leave your JSON file as it is (without stripping new lines characters \n), include multiLine=True keyword argument

sc = SparkContext() 
sqlc = SQLContext(sc)

df = sqlc.read.json('my_file.json', multiLine=True)

print df.show()

Answer 5

I want to share my experience in which I have a JSON column String but with Python notation, which means I have None instead of null, False instead of false and True instead of true.

When parsing this column, spark returns me a column named _corrupt_record. So what I had to do before parsing the JSON String is replacing the Python notation with the standard JSON notation:

df.withColumn("json_notation",
    F.regexp_replace(F.regexp_replace(F.regexp_replace("_corrupt_record", "None", "null"), "False", "false") ,"True", "true")

After this transformation I was then able to use for example the function F.from_json() on the json_notation column and here Pyspark was able to correctly parse the JSON object.