WIP fixing 1.1 merge

Author: giwa

python/pyspark/streaming/dstream.py

@@ -25,6 +25,7 @@
 from pyspark.storagelevel import StorageLevel
 from pyspark.resultiterable import ResultIterable
 from pyspark.streaming.util import rddToFileName, RDDFunction
+from pyspark.rdd import portable_hash, _parse_memory
 from pyspark.traceback_utils import SCCallSiteSync
 
 from py4j.java_collections import ListConverter, MapConverter
@@ -40,6 +41,7 @@ def __init__(self, jdstream, ssc, jrdd_deserializer):
         self._jrdd_deserializer = jrdd_deserializer
         self.is_cached = False
         self.is_checkpointed = False
+        self._partitionFunc = None
 
     def context(self):
         """
@@ -161,32 +163,71 @@ def _mergeCombiners(iterator):
 
         return shuffled.mapPartitions(_mergeCombiners)
 
-    def partitionBy(self, numPartitions, partitionFunc=None):
+    def partitionBy(self, numPartitions, partitionFunc=portable_hash):
         """
         Return a copy of the DStream partitioned using the specified partitioner.
         """
         if numPartitions is None:
             numPartitions = self.ctx._defaultReducePartitions()
 
-        if partitionFunc is None:
-            partitionFunc = lambda x: 0 if x is None else hash(x)
-
         # Transferring O(n) objects to Java is too expensive.  Instead, we'll
         # form the hash buckets in Python, transferring O(numPartitions) objects
         # to Java.  Each object is a (splitNumber, [objects]) pair.
+
         outputSerializer = self.ctx._unbatched_serializer
+#
+#        def add_shuffle_key(split, iterator):
+#            buckets = defaultdict(list)
+#
+#            for (k, v) in iterator:
+#                buckets[partitionFunc(k) % numPartitions].append((k, v))
+#            for (split, items) in buckets.iteritems():
+#                yield pack_long(split)
+#                yield outputSerializer.dumps(items)
+#        keyed = PipelinedDStream(self, add_shuffle_key)
+
+        limit = (_parse_memory(self.ctx._conf.get(
+            "spark.python.worker.memory", "512m")) / 2)
 
         def add_shuffle_key(split, iterator):
+
             buckets = defaultdict(list)
+            c, batch = 0, min(10 * numPartitions, 1000)
 
-            for (k, v) in iterator:
+            for k, v in iterator:
                 buckets[partitionFunc(k) % numPartitions].append((k, v))
-            for (split, items) in buckets.iteritems():
+                c += 1
+
+                # check used memory and avg size of chunk of objects
+                if (c % 1000 == 0 and get_used_memory() > limit
+                        or c > batch):
+                    n, size = len(buckets), 0
+                    for split in buckets.keys():
+                        yield pack_long(split)
+                        d = outputSerializer.dumps(buckets[split])
+                        del buckets[split]
+                        yield d
+                        size += len(d)
+
+                    avg = (size / n) >> 20
+                    # let 1M < avg < 10M
+                    if avg < 1:
+                        batch *= 1.5
+                    elif avg > 10:
+                        batch = max(batch / 1.5, 1)
+                    c = 0
+
+            for split, items in buckets.iteritems():
                 yield pack_long(split)
                 yield outputSerializer.dumps(items)
-        keyed = PipelinedDStream(self, add_shuffle_key)
+
+        keyed = self._mapPartitionsWithIndex(add_shuffle_key)
+
+
+
+
         keyed._bypass_serializer = True
-        with SCCallSiteSync(self.context) as css:
+        with SCCallSiteSync(self.ctx) as css:
             partitioner = self.ctx._jvm.PythonPartitioner(numPartitions,
                                                           id(partitionFunc))
             jdstream = self.ctx._jvm.PythonPairwiseDStream(keyed._jdstream.dstream(),
@@ -428,6 +469,10 @@ def get_output(rdd, time):
 
 
 class PipelinedDStream(DStream):
+    """
+    Since PipelinedDStream is same to PipelindRDD, if PipliedRDD is changed,
+    this code should be changed in the same way.
+    """
     def __init__(self, prev, func, preservesPartitioning=False):
         if not isinstance(prev, PipelinedDStream) or not prev._is_pipelinable():
             # This transformation is the first in its stage:
@@ -453,19 +498,22 @@ def pipeline_func(split, iterator):
         self._jdstream_val = None
         self._jrdd_deserializer = self.ctx.serializer
         self._bypass_serializer = False
+        self._partitionFunc = prev._partitionFunc if self.preservesPartitioning else None
 
     @property
     def _jdstream(self):
         if self._jdstream_val:
             return self._jdstream_val
         if self._bypass_serializer:
-            serializer = NoOpSerializer()
-        else:
-            serializer = self.ctx.serializer
-
-        command = (self.func, self._prev_jrdd_deserializer, serializer)
-        ser = CompressedSerializer(CloudPickleSerializer())
+            self.jrdd_deserializer = NoOpSerializer()
+        command = (self.func, self._prev_jrdd_deserializer,
+                   self._jrdd_deserializer)
+        # the serialized command will be compressed by broadcast
+        ser = CloudPickleSerializer()
         pickled_command = ser.dumps(command)
+        if pickled_command > (1 << 20):  # 1M
+            broadcast = self.ctx.broadcast(pickled_command)
+            pickled_command = ser.dumps(broadcast)
         broadcast_vars = ListConverter().convert(
             [x._jbroadcast for x in self.ctx._pickled_broadcast_vars],
             self.ctx._gateway._gateway_client)