Transform a list of files (JSON) to a dataframe - hadoop

Question

Spark Version: '2.0.0.2.5.0.0-1245'
So, my original question changed a bit but it's still the same issue.
What I want to do is load a huge amount of JSON files and transform those to a DataFrame - also probably save them as CSV or parquet file for further processing. Each JSON file represents one row in the final DataFrame.
import os
import glob
HDFS_MOUNT = # ...
DATA_SET_BASE = # ...
schema = StructType([
StructField("documentId", StringType(), True),
StructField("group", StringType(), True),
StructField("text", StringType(), True)
])
# Get the file paths
file_paths = glob.glob(os.path.join(HDFS_MOUNT, DATA_SET_BASE, '**/*.json'))
file_paths = [f.replace(HDFS_MOUNT + '/', '') for f in file_paths]
print('Found {:d} files'.format(len(file_paths))) # 676 files
sql = SQLContext(sc)
df = sql.read.json(file_paths, schema=schema)
print('Loaded {:d} rows'.format(df.count())) # 9660 rows (what !?)
Besides the fact that there are 9660 rows instead of 676 (number of available files) I also have the problem that the content seems to be None:
df.head(2)[0].asDict()
gives
{
'documentId': None,
'group': None,
'text': None,
}
Example Data
This is just fake data of course but it resembles the actual data.
Note: Some fields may be missing e.g. text must not always be present.
a.json
{
"documentId" : "001",
"group" : "A",
"category" : "indexed_document",
"linkIDs": ["adiojer", "asdi555", "1337"]
}
b.json
{
"documentId" : "002",
"group" : "B",
"category" : "indexed_document",
"linkIDs": ["linkId", "1000"],
"text": "This is the text of this document"
}

Answer 1

assuming that all your files has the same structure and are in the same directory:
df = sql_cntx.read.json('/hdfs/path/to/folder/*.json')
There might be a problem if any of the columns has Null values for all rows. Then spark will not be able to determine schema, so you have an option to tell spark which schema to use:
from pyspark import SparkContext, SQLContext
from pyspark.sql.types import StructType, StructField, StringType, LongType
sc = SparkContext(appName="My app")
sql_cntx = SQLContext(sc)
schema = StructType([
StructField("field1", StringType(), True),
StructField("field2", LongType(), True)
])
df = sql_cntx.read.json('/hdfs/path/to/folder/*.json', schema=schema)
UPD:
in case if file has multirows formatted json you can try this code:
sc = SparkContext(appName='Test')
sql_context = SQLContext(sc)
rdd = sc.wholeTextFiles('/tmp/test/*.json').values()
df = sql_context.read.json(rdd, schema=schema)
df.show()