1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
|
#
# Licensed to the Apache Software Foundation (ASF) under one or more
# contributor license agreements. See the NOTICE file distributed with
# this work for additional information regarding copyright ownership.
# The ASF licenses this file to You under the Apache License, Version 2.0
# (the "License"); you may not use this file except in compliance with
# the License. You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
from pyspark.rdd import RDD, PipelinedRDD
from pyspark.serializers import BatchedSerializer, PickleSerializer
from py4j.protocol import Py4JError
__all__ = ["SQLContext", "HiveContext", "LocalHiveContext", "TestHiveContext", "SchemaRDD", "Row"]
class SQLContext:
"""Main entry point for SparkSQL functionality.
A SQLContext can be used create L{SchemaRDD}s, register L{SchemaRDD}s as
tables, execute SQL over tables, cache tables, and read parquet files.
"""
def __init__(self, sparkContext, sqlContext=None):
"""Create a new SQLContext.
@param sparkContext: The SparkContext to wrap.
>>> srdd = sqlCtx.inferSchema(rdd)
>>> sqlCtx.inferSchema(srdd) # doctest: +IGNORE_EXCEPTION_DETAIL
Traceback (most recent call last):
...
ValueError:...
>>> bad_rdd = sc.parallelize([1,2,3])
>>> sqlCtx.inferSchema(bad_rdd) # doctest: +IGNORE_EXCEPTION_DETAIL
Traceback (most recent call last):
...
ValueError:...
>>> from datetime import datetime
>>> allTypes = sc.parallelize([{"int": 1, "string": "string", "double": 1.0, "long": 1L,
... "boolean": True, "time": datetime(2010, 1, 1, 1, 1, 1), "dict": {"a": 1},
... "list": [1, 2, 3]}])
>>> srdd = sqlCtx.inferSchema(allTypes).map(lambda x: (x.int, x.string, x.double, x.long,
... x.boolean, x.time, x.dict["a"], x.list))
>>> srdd.collect()[0]
(1, u'string', 1.0, 1, True, datetime.datetime(2010, 1, 1, 1, 1, 1), 1, [1, 2, 3])
"""
self._sc = sparkContext
self._jsc = self._sc._jsc
self._jvm = self._sc._jvm
self._pythonToJavaMap = self._jvm.PythonRDD.pythonToJavaMap
if sqlContext:
self._scala_SQLContext = sqlContext
@property
def _ssql_ctx(self):
"""Accessor for the JVM SparkSQL context.
Subclasses can override this property to provide their own
JVM Contexts.
"""
if not hasattr(self, '_scala_SQLContext'):
self._scala_SQLContext = self._jvm.SQLContext(self._jsc.sc())
return self._scala_SQLContext
def inferSchema(self, rdd):
"""Infer and apply a schema to an RDD of L{dict}s.
We peek at the first row of the RDD to determine the fields names
and types, and then use that to extract all the dictionaries. Nested
collections are supported, which include array, dict, list, set, and
tuple.
>>> srdd = sqlCtx.inferSchema(rdd)
>>> srdd.collect() == [{"field1" : 1, "field2" : "row1"}, {"field1" : 2, "field2": "row2"},
... {"field1" : 3, "field2": "row3"}]
True
>>> from array import array
>>> srdd = sqlCtx.inferSchema(nestedRdd1)
>>> srdd.collect() == [{"f1" : [1, 2], "f2" : {"row1" : 1.0}},
... {"f1" : [2, 3], "f2" : {"row2" : 2.0}}]
True
>>> srdd = sqlCtx.inferSchema(nestedRdd2)
>>> srdd.collect() == [{"f1" : [[1, 2], [2, 3]], "f2" : [1, 2]},
... {"f1" : [[2, 3], [3, 4]], "f2" : [2, 3]}]
True
"""
if (rdd.__class__ is SchemaRDD):
raise ValueError("Cannot apply schema to %s" % SchemaRDD.__name__)
elif not isinstance(rdd.first(), dict):
raise ValueError("Only RDDs with dictionaries can be converted to %s: %s" %
(SchemaRDD.__name__, rdd.first()))
jrdd = self._pythonToJavaMap(rdd._jrdd)
srdd = self._ssql_ctx.inferSchema(jrdd.rdd())
return SchemaRDD(srdd, self)
def registerRDDAsTable(self, rdd, tableName):
"""Registers the given RDD as a temporary table in the catalog.
Temporary tables exist only during the lifetime of this instance of
SQLContext.
>>> srdd = sqlCtx.inferSchema(rdd)
>>> sqlCtx.registerRDDAsTable(srdd, "table1")
"""
if (rdd.__class__ is SchemaRDD):
jschema_rdd = rdd._jschema_rdd
self._ssql_ctx.registerRDDAsTable(jschema_rdd, tableName)
else:
raise ValueError("Can only register SchemaRDD as table")
def parquetFile(self, path):
"""Loads a Parquet file, returning the result as a L{SchemaRDD}.
>>> import tempfile, shutil
>>> parquetFile = tempfile.mkdtemp()
>>> shutil.rmtree(parquetFile)
>>> srdd = sqlCtx.inferSchema(rdd)
>>> srdd.saveAsParquetFile(parquetFile)
>>> srdd2 = sqlCtx.parquetFile(parquetFile)
>>> sorted(srdd.collect()) == sorted(srdd2.collect())
True
"""
jschema_rdd = self._ssql_ctx.parquetFile(path)
return SchemaRDD(jschema_rdd, self)
def jsonFile(self, path):
"""Loads a text file storing one JSON object per line,
returning the result as a L{SchemaRDD}.
It goes through the entire dataset once to determine the schema.
>>> import tempfile, shutil
>>> jsonFile = tempfile.mkdtemp()
>>> shutil.rmtree(jsonFile)
>>> ofn = open(jsonFile, 'w')
>>> for json in jsonStrings:
... print>>ofn, json
>>> ofn.close()
>>> srdd = sqlCtx.jsonFile(jsonFile)
>>> sqlCtx.registerRDDAsTable(srdd, "table1")
>>> srdd2 = sqlCtx.sql(
... "SELECT field1 AS f1, field2 as f2, field3 as f3, field6 as f4 from table1")
>>> srdd2.collect() == [
... {"f1":1, "f2":"row1", "f3":{"field4":11, "field5": None}, "f4":None},
... {"f1":2, "f2":None, "f3":{"field4":22, "field5": [10, 11]}, "f4":[{"field7": "row2"}]},
... {"f1":None, "f2":"row3", "f3":{"field4":33, "field5": []}, "f4":None}]
True
"""
jschema_rdd = self._ssql_ctx.jsonFile(path)
return SchemaRDD(jschema_rdd, self)
def jsonRDD(self, rdd):
"""Loads an RDD storing one JSON object per string, returning the result as a L{SchemaRDD}.
It goes through the entire dataset once to determine the schema.
>>> srdd = sqlCtx.jsonRDD(json)
>>> sqlCtx.registerRDDAsTable(srdd, "table1")
>>> srdd2 = sqlCtx.sql(
... "SELECT field1 AS f1, field2 as f2, field3 as f3, field6 as f4 from table1")
>>> srdd2.collect() == [
... {"f1":1, "f2":"row1", "f3":{"field4":11, "field5": None}, "f4":None},
... {"f1":2, "f2":None, "f3":{"field4":22, "field5": [10, 11]}, "f4":[{"field7": "row2"}]},
... {"f1":None, "f2":"row3", "f3":{"field4":33, "field5": []}, "f4":None}]
True
"""
def func(split, iterator):
for x in iterator:
if not isinstance(x, basestring):
x = unicode(x)
yield x.encode("utf-8")
keyed = PipelinedRDD(rdd, func)
keyed._bypass_serializer = True
jrdd = keyed._jrdd.map(self._jvm.BytesToString())
jschema_rdd = self._ssql_ctx.jsonRDD(jrdd.rdd())
return SchemaRDD(jschema_rdd, self)
def sql(self, sqlQuery):
"""Return a L{SchemaRDD} representing the result of the given query.
>>> srdd = sqlCtx.inferSchema(rdd)
>>> sqlCtx.registerRDDAsTable(srdd, "table1")
>>> srdd2 = sqlCtx.sql("SELECT field1 AS f1, field2 as f2 from table1")
>>> srdd2.collect() == [{"f1" : 1, "f2" : "row1"}, {"f1" : 2, "f2": "row2"},
... {"f1" : 3, "f2": "row3"}]
True
"""
return SchemaRDD(self._ssql_ctx.sql(sqlQuery), self)
def table(self, tableName):
"""Returns the specified table as a L{SchemaRDD}.
>>> srdd = sqlCtx.inferSchema(rdd)
>>> sqlCtx.registerRDDAsTable(srdd, "table1")
>>> srdd2 = sqlCtx.table("table1")
>>> sorted(srdd.collect()) == sorted(srdd2.collect())
True
"""
return SchemaRDD(self._ssql_ctx.table(tableName), self)
def cacheTable(self, tableName):
"""Caches the specified table in-memory."""
self._ssql_ctx.cacheTable(tableName)
def uncacheTable(self, tableName):
"""Removes the specified table from the in-memory cache."""
self._ssql_ctx.uncacheTable(tableName)
class HiveContext(SQLContext):
"""A variant of Spark SQL that integrates with data stored in Hive.
Configuration for Hive is read from hive-site.xml on the classpath.
It supports running both SQL and HiveQL commands.
"""
@property
def _ssql_ctx(self):
try:
if not hasattr(self, '_scala_HiveContext'):
self._scala_HiveContext = self._get_hive_ctx()
return self._scala_HiveContext
except Py4JError as e:
raise Exception("You must build Spark with Hive. Export 'SPARK_HIVE=true' and run "
"sbt/sbt assembly", e)
def _get_hive_ctx(self):
return self._jvm.HiveContext(self._jsc.sc())
def hiveql(self, hqlQuery):
"""
Runs a query expressed in HiveQL, returning the result as a L{SchemaRDD}.
"""
return SchemaRDD(self._ssql_ctx.hiveql(hqlQuery), self)
def hql(self, hqlQuery):
"""
Runs a query expressed in HiveQL, returning the result as a L{SchemaRDD}.
"""
return self.hiveql(hqlQuery)
class LocalHiveContext(HiveContext):
"""Starts up an instance of hive where metadata is stored locally.
An in-process metadata data is created with data stored in ./metadata.
Warehouse data is stored in in ./warehouse.
>>> import os
>>> hiveCtx = LocalHiveContext(sc)
>>> try:
... supress = hiveCtx.hql("DROP TABLE src")
... except Exception:
... pass
>>> kv1 = os.path.join(os.environ["SPARK_HOME"], 'examples/src/main/resources/kv1.txt')
>>> supress = hiveCtx.hql("CREATE TABLE IF NOT EXISTS src (key INT, value STRING)")
>>> supress = hiveCtx.hql("LOAD DATA LOCAL INPATH '%s' INTO TABLE src" % kv1)
>>> results = hiveCtx.hql("FROM src SELECT value").map(lambda r: int(r.value.split('_')[1]))
>>> num = results.count()
>>> reduce_sum = results.reduce(lambda x, y: x + y)
>>> num
500
>>> reduce_sum
130091
"""
def _get_hive_ctx(self):
return self._jvm.LocalHiveContext(self._jsc.sc())
class TestHiveContext(HiveContext):
def _get_hive_ctx(self):
return self._jvm.TestHiveContext(self._jsc.sc())
# TODO: Investigate if it is more efficient to use a namedtuple. One problem is that named tuples
# are custom classes that must be generated per Schema.
class Row(dict):
"""A row in L{SchemaRDD}.
An extended L{dict} that takes a L{dict} in its constructor, and
exposes those items as fields.
>>> r = Row({"hello" : "world", "foo" : "bar"})
>>> r.hello
'world'
>>> r.foo
'bar'
"""
def __init__(self, d):
d.update(self.__dict__)
self.__dict__ = d
dict.__init__(self, d)
class SchemaRDD(RDD):
"""An RDD of L{Row} objects that has an associated schema.
The underlying JVM object is a SchemaRDD, not a PythonRDD, so we can
utilize the relational query api exposed by SparkSQL.
For normal L{pyspark.rdd.RDD} operations (map, count, etc.) the
L{SchemaRDD} is not operated on directly, as it's underlying
implementation is an RDD composed of Java objects. Instead it is
converted to a PythonRDD in the JVM, on which Python operations can
be done.
"""
def __init__(self, jschema_rdd, sql_ctx):
self.sql_ctx = sql_ctx
self._sc = sql_ctx._sc
self._jschema_rdd = jschema_rdd
self.is_cached = False
self.is_checkpointed = False
self.ctx = self.sql_ctx._sc
self._jrdd_deserializer = self.ctx.serializer
@property
def _jrdd(self):
"""Lazy evaluation of PythonRDD object.
Only done when a user calls methods defined by the
L{pyspark.rdd.RDD} super class (map, filter, etc.).
"""
if not hasattr(self, '_lazy_jrdd'):
self._lazy_jrdd = self._toPython()._jrdd
return self._lazy_jrdd
@property
def _id(self):
return self._jrdd.id()
def saveAsParquetFile(self, path):
"""Save the contents as a Parquet file, preserving the schema.
Files that are written out using this method can be read back in as
a SchemaRDD using the L{SQLContext.parquetFile} method.
>>> import tempfile, shutil
>>> parquetFile = tempfile.mkdtemp()
>>> shutil.rmtree(parquetFile)
>>> srdd = sqlCtx.inferSchema(rdd)
>>> srdd.saveAsParquetFile(parquetFile)
>>> srdd2 = sqlCtx.parquetFile(parquetFile)
>>> sorted(srdd2.collect()) == sorted(srdd.collect())
True
"""
self._jschema_rdd.saveAsParquetFile(path)
def registerAsTable(self, name):
"""Registers this RDD as a temporary table using the given name.
The lifetime of this temporary table is tied to the L{SQLContext}
that was used to create this SchemaRDD.
>>> srdd = sqlCtx.inferSchema(rdd)
>>> srdd.registerAsTable("test")
>>> srdd2 = sqlCtx.sql("select * from test")
>>> sorted(srdd.collect()) == sorted(srdd2.collect())
True
"""
self._jschema_rdd.registerAsTable(name)
def insertInto(self, tableName, overwrite=False):
"""Inserts the contents of this SchemaRDD into the specified table.
Optionally overwriting any existing data.
"""
self._jschema_rdd.insertInto(tableName, overwrite)
def saveAsTable(self, tableName):
"""Creates a new table with the contents of this SchemaRDD."""
self._jschema_rdd.saveAsTable(tableName)
def schemaString(self):
"""Returns the output schema in the tree format."""
return self._jschema_rdd.schemaString()
def printSchema(self):
"""Prints out the schema in the tree format."""
print self.schemaString()
def count(self):
"""Return the number of elements in this RDD.
Unlike the base RDD implementation of count, this implementation
leverages the query optimizer to compute the count on the SchemaRDD,
which supports features such as filter pushdown.
>>> srdd = sqlCtx.inferSchema(rdd)
>>> srdd.count()
3L
>>> srdd.count() == srdd.map(lambda x: x).count()
True
"""
return self._jschema_rdd.count()
def _toPython(self):
# We have to import the Row class explicitly, so that the reference Pickler has is
# pyspark.sql.Row instead of __main__.Row
from pyspark.sql import Row
jrdd = self._jschema_rdd.javaToPython()
# TODO: This is inefficient, we should construct the Python Row object
# in Java land in the javaToPython function. May require a custom
# pickle serializer in Pyrolite
return RDD(jrdd, self._sc, BatchedSerializer(
PickleSerializer())).map(lambda d: Row(d))
# We override the default cache/persist/checkpoint behavior as we want to cache the underlying
# SchemaRDD object in the JVM, not the PythonRDD checkpointed by the super class
def cache(self):
self.is_cached = True
self._jschema_rdd.cache()
return self
def persist(self, storageLevel):
self.is_cached = True
javaStorageLevel = self.ctx._getJavaStorageLevel(storageLevel)
self._jschema_rdd.persist(javaStorageLevel)
return self
def unpersist(self):
self.is_cached = False
self._jschema_rdd.unpersist()
return self
def checkpoint(self):
self.is_checkpointed = True
self._jschema_rdd.checkpoint()
def isCheckpointed(self):
return self._jschema_rdd.isCheckpointed()
def getCheckpointFile(self):
checkpointFile = self._jschema_rdd.getCheckpointFile()
if checkpointFile.isDefined():
return checkpointFile.get()
else:
return None
def coalesce(self, numPartitions, shuffle=False):
rdd = self._jschema_rdd.coalesce(numPartitions, shuffle)
return SchemaRDD(rdd, self.sql_ctx)
def distinct(self):
rdd = self._jschema_rdd.distinct()
return SchemaRDD(rdd, self.sql_ctx)
def intersection(self, other):
if (other.__class__ is SchemaRDD):
rdd = self._jschema_rdd.intersection(other._jschema_rdd)
return SchemaRDD(rdd, self.sql_ctx)
else:
raise ValueError("Can only intersect with another SchemaRDD")
def repartition(self, numPartitions):
rdd = self._jschema_rdd.repartition(numPartitions)
return SchemaRDD(rdd, self.sql_ctx)
def subtract(self, other, numPartitions=None):
if (other.__class__ is SchemaRDD):
if numPartitions is None:
rdd = self._jschema_rdd.subtract(other._jschema_rdd)
else:
rdd = self._jschema_rdd.subtract(other._jschema_rdd, numPartitions)
return SchemaRDD(rdd, self.sql_ctx)
else:
raise ValueError("Can only subtract another SchemaRDD")
def _test():
import doctest
from array import array
from pyspark.context import SparkContext
globs = globals().copy()
# The small batch size here ensures that we see multiple batches,
# even in these small test examples:
sc = SparkContext('local[4]', 'PythonTest', batchSize=2)
globs['sc'] = sc
globs['sqlCtx'] = SQLContext(sc)
globs['rdd'] = sc.parallelize(
[{"field1": 1, "field2": "row1"},
{"field1": 2, "field2": "row2"},
{"field1": 3, "field2": "row3"}]
)
jsonStrings = [
'{"field1": 1, "field2": "row1", "field3":{"field4":11}}',
'{"field1" : 2, "field3":{"field4":22, "field5": [10, 11]}, "field6":[{"field7": "row2"}]}',
'{"field1" : null, "field2": "row3", "field3":{"field4":33, "field5": []}}'
]
globs['jsonStrings'] = jsonStrings
globs['json'] = sc.parallelize(jsonStrings)
globs['nestedRdd1'] = sc.parallelize([
{"f1": array('i', [1, 2]), "f2": {"row1": 1.0}},
{"f1": array('i', [2, 3]), "f2": {"row2": 2.0}}])
globs['nestedRdd2'] = sc.parallelize([
{"f1": [[1, 2], [2, 3]], "f2": [1, 2]},
{"f1": [[2, 3], [3, 4]], "f2": [2, 3]}])
(failure_count, test_count) = doctest.testmod(globs=globs, optionflags=doctest.ELLIPSIS)
globs['sc'].stop()
if failure_count:
exit(-1)
if __name__ == "__main__":
_test()
|