Stuart created SPARK-21985:
------------------------------
Summary: PySpark PairDeserializer is broken for double-zipped RDDs
Key: SPARK-21985
URL: https://issues.apache.org/jira/browse/SPARK-21985
Project: Spark
Issue Type: Bug
Components: PySpark
Affects Versions: 2.2.0, 2.1.1, 2.1.0
Reporter: Stuart
PySpark fails to deserialize double-zipped RDDs. For example, the following
example used to work in Spark 2.0.2:
{{>>> a = sc.parallelize('aaa')
>>> b = sc.parallelize('bbb')
>>> c = sc.parallelize('ccc')
>>> a_bc = a.zip( b.zip(c) )
>>> a_bc.collect()
[('a', ('b', 'c')), ('a', ('b', 'c')), ('a', ('b', 'c'))]}}
But in Spark >=2.1.0, it fails:
{{>>> a_bc.collect()
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/magnetic/workspace/spark-2.2.0-bin-hadoop2.7/python/pyspark/rdd.py",
line 810, in collect
return list(_load_from_socket(port, self._jrdd_deserializer))
File
"/magnetic/workspace/spark-2.2.0-bin-hadoop2.7/python/pyspark/serializers.py",
line 329, in _load_stream_without_unbatching
if len(key_batch) != len(val_batch):
TypeError: object of type 'itertools.izip' has no len()}}
As you can see, the error seems to be caused by [a check in the
PairDeserializer
class|https://github.com/apache/spark/blob/master/python/pyspark/serializers.py#L346-L348]:
{{if len(key_batch) != len(val_batch):
raise ValueError("Can not deserialize PairRDD with different number of
items"
" in batches: (%d, %d)" % (len(key_batch), len(val_batch)))
}}
If that check is removed, then the example above works without error. Can the
check simply be removed?
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
