2014-07-25 01:53:47 -04:00
|
|
|
#
|
|
|
|
# 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.
|
|
|
|
#
|
|
|
|
|
|
|
|
import os
|
|
|
|
import platform
|
|
|
|
import shutil
|
|
|
|
import warnings
|
|
|
|
import gc
|
2014-08-26 19:57:40 -04:00
|
|
|
import itertools
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
import operator
|
2014-08-26 19:57:40 -04:00
|
|
|
import random
|
2018-03-08 06:38:34 -05:00
|
|
|
import sys
|
2014-07-25 01:53:47 -04:00
|
|
|
|
2014-08-26 19:57:40 -04:00
|
|
|
import pyspark.heapq3 as heapq
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
from pyspark.serializers import BatchedSerializer, PickleSerializer, FlattenedValuesSerializer, \
|
|
|
|
CompressedSerializer, AutoBatchedSerializer
|
2018-05-30 06:11:33 -04:00
|
|
|
from pyspark.util import fail_on_stopiteration
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
|
2014-07-25 01:53:47 -04:00
|
|
|
|
|
|
|
try:
|
|
|
|
import psutil
|
|
|
|
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
process = None
|
|
|
|
|
2014-07-25 01:53:47 -04:00
|
|
|
def get_used_memory():
|
|
|
|
""" Return the used memory in MB """
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
global process
|
|
|
|
if process is None or process._pid != os.getpid():
|
|
|
|
process = psutil.Process(os.getpid())
|
2014-07-25 01:53:47 -04:00
|
|
|
if hasattr(process, "memory_info"):
|
|
|
|
info = process.memory_info()
|
|
|
|
else:
|
|
|
|
info = process.get_memory_info()
|
|
|
|
return info.rss >> 20
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
|
2014-07-25 01:53:47 -04:00
|
|
|
except ImportError:
|
|
|
|
|
|
|
|
def get_used_memory():
|
|
|
|
""" Return the used memory in MB """
|
|
|
|
if platform.system() == 'Linux':
|
|
|
|
for line in open('/proc/self/status'):
|
|
|
|
if line.startswith('VmRSS:'):
|
|
|
|
return int(line.split()[1]) >> 10
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
|
2014-07-25 01:53:47 -04:00
|
|
|
else:
|
|
|
|
warnings.warn("Please install psutil to have better "
|
2014-08-06 15:58:24 -04:00
|
|
|
"support with spilling")
|
2014-07-25 01:53:47 -04:00
|
|
|
if platform.system() == "Darwin":
|
|
|
|
import resource
|
|
|
|
rss = resource.getrusage(resource.RUSAGE_SELF).ru_maxrss
|
|
|
|
return rss >> 20
|
|
|
|
# TODO: support windows
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
|
2014-07-25 01:53:47 -04:00
|
|
|
return 0
|
|
|
|
|
|
|
|
|
2014-08-26 19:57:40 -04:00
|
|
|
def _get_local_dirs(sub):
|
|
|
|
""" Get all the directories """
|
|
|
|
path = os.environ.get("SPARK_LOCAL_DIRS", "/tmp")
|
|
|
|
dirs = path.split(",")
|
|
|
|
if len(dirs) > 1:
|
|
|
|
# different order in different processes and instances
|
|
|
|
rnd = random.Random(os.getpid() + id(dirs))
|
|
|
|
random.shuffle(dirs, rnd.random)
|
|
|
|
return [os.path.join(d, "python", str(os.getpid()), sub) for d in dirs]
|
|
|
|
|
|
|
|
|
2014-09-14 01:31:21 -04:00
|
|
|
# global stats
|
2015-04-16 19:20:57 -04:00
|
|
|
MemoryBytesSpilled = 0
|
|
|
|
DiskBytesSpilled = 0
|
2014-09-14 01:31:21 -04:00
|
|
|
|
|
|
|
|
2014-07-25 01:53:47 -04:00
|
|
|
class Aggregator(object):
|
|
|
|
|
|
|
|
"""
|
|
|
|
Aggregator has tree functions to merge values into combiner.
|
|
|
|
|
|
|
|
createCombiner: (value) -> combiner
|
|
|
|
mergeValue: (combine, value) -> combiner
|
|
|
|
mergeCombiners: (combiner, combiner) -> combiner
|
|
|
|
"""
|
|
|
|
|
|
|
|
def __init__(self, createCombiner, mergeValue, mergeCombiners):
|
2018-05-30 06:11:33 -04:00
|
|
|
self.createCombiner = fail_on_stopiteration(createCombiner)
|
|
|
|
self.mergeValue = fail_on_stopiteration(mergeValue)
|
|
|
|
self.mergeCombiners = fail_on_stopiteration(mergeCombiners)
|
2014-07-25 01:53:47 -04:00
|
|
|
|
|
|
|
|
|
|
|
class SimpleAggregator(Aggregator):
|
|
|
|
|
|
|
|
"""
|
|
|
|
SimpleAggregator is useful for the cases that combiners have
|
|
|
|
same type with values
|
|
|
|
"""
|
|
|
|
|
|
|
|
def __init__(self, combiner):
|
|
|
|
Aggregator.__init__(self, lambda x: x, combiner, combiner)
|
|
|
|
|
|
|
|
|
|
|
|
class Merger(object):
|
|
|
|
|
|
|
|
"""
|
|
|
|
Merge shuffled data together by aggregator
|
|
|
|
"""
|
|
|
|
|
|
|
|
def __init__(self, aggregator):
|
|
|
|
self.agg = aggregator
|
|
|
|
|
|
|
|
def mergeValues(self, iterator):
|
|
|
|
""" Combine the items by creator and combiner """
|
|
|
|
raise NotImplementedError
|
|
|
|
|
|
|
|
def mergeCombiners(self, iterator):
|
|
|
|
""" Merge the combined items by mergeCombiner """
|
|
|
|
raise NotImplementedError
|
|
|
|
|
2015-04-16 19:20:57 -04:00
|
|
|
def items(self):
|
2014-07-25 01:53:47 -04:00
|
|
|
""" Return the merged items ad iterator """
|
|
|
|
raise NotImplementedError
|
|
|
|
|
|
|
|
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
def _compressed_serializer(self, serializer=None):
|
|
|
|
# always use PickleSerializer to simplify implementation
|
|
|
|
ser = PickleSerializer()
|
|
|
|
return AutoBatchedSerializer(CompressedSerializer(ser))
|
|
|
|
|
|
|
|
|
2014-07-25 01:53:47 -04:00
|
|
|
class ExternalMerger(Merger):
|
|
|
|
|
|
|
|
"""
|
|
|
|
External merger will dump the aggregated data into disks when
|
|
|
|
memory usage goes above the limit, then merge them together.
|
|
|
|
|
|
|
|
This class works as follows:
|
|
|
|
|
2014-08-06 15:58:24 -04:00
|
|
|
- It repeatedly combine the items and save them in one dict in
|
2014-07-25 01:53:47 -04:00
|
|
|
memory.
|
|
|
|
|
|
|
|
- When the used memory goes above memory limit, it will split
|
|
|
|
the combined data into partitions by hash code, dump them
|
|
|
|
into disk, one file per partition.
|
|
|
|
|
|
|
|
- Then it goes through the rest of the iterator, combine items
|
|
|
|
into different dict by hash. Until the used memory goes over
|
|
|
|
memory limit, it dump all the dicts into disks, one file per
|
|
|
|
dict. Repeat this again until combine all the items.
|
|
|
|
|
|
|
|
- Before return any items, it will load each partition and
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
combine them separately. Yield them before loading next
|
2014-07-25 01:53:47 -04:00
|
|
|
partition.
|
|
|
|
|
|
|
|
- During loading a partition, if the memory goes over limit,
|
|
|
|
it will partition the loaded data and dump them into disks
|
|
|
|
and load them partition by partition again.
|
|
|
|
|
|
|
|
`data` and `pdata` are used to hold the merged items in memory.
|
|
|
|
At first, all the data are merged into `data`. Once the used
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
memory goes over limit, the items in `data` are dumped into
|
2014-07-25 01:53:47 -04:00
|
|
|
disks, `data` will be cleared, all rest of items will be merged
|
|
|
|
into `pdata` and then dumped into disks. Before returning, all
|
|
|
|
the items in `pdata` will be dumped into disks.
|
|
|
|
|
|
|
|
Finally, if any items were spilled into disks, each partition
|
|
|
|
will be merged into `data` and be yielded, then cleared.
|
|
|
|
|
|
|
|
>>> agg = SimpleAggregator(lambda x, y: x + y)
|
|
|
|
>>> merger = ExternalMerger(agg, 10)
|
|
|
|
>>> N = 10000
|
2015-04-16 19:20:57 -04:00
|
|
|
>>> merger.mergeValues(zip(range(N), range(N)))
|
2014-07-25 01:53:47 -04:00
|
|
|
>>> assert merger.spills > 0
|
2015-04-16 19:20:57 -04:00
|
|
|
>>> sum(v for k,v in merger.items())
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
49995000
|
2014-07-25 01:53:47 -04:00
|
|
|
|
|
|
|
>>> merger = ExternalMerger(agg, 10)
|
2015-04-16 19:20:57 -04:00
|
|
|
>>> merger.mergeCombiners(zip(range(N), range(N)))
|
2014-07-25 01:53:47 -04:00
|
|
|
>>> assert merger.spills > 0
|
2015-04-16 19:20:57 -04:00
|
|
|
>>> sum(v for k,v in merger.items())
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
49995000
|
2014-07-25 01:53:47 -04:00
|
|
|
"""
|
|
|
|
|
|
|
|
# the max total partitions created recursively
|
|
|
|
MAX_TOTAL_PARTITIONS = 4096
|
|
|
|
|
|
|
|
def __init__(self, aggregator, memory_limit=512, serializer=None,
|
2014-08-06 15:58:24 -04:00
|
|
|
localdirs=None, scale=1, partitions=59, batch=1000):
|
2014-07-25 01:53:47 -04:00
|
|
|
Merger.__init__(self, aggregator)
|
|
|
|
self.memory_limit = memory_limit
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
self.serializer = _compressed_serializer(serializer)
|
2014-08-26 19:57:40 -04:00
|
|
|
self.localdirs = localdirs or _get_local_dirs(str(id(self)))
|
2014-07-25 01:53:47 -04:00
|
|
|
# number of partitions when spill data into disks
|
|
|
|
self.partitions = partitions
|
|
|
|
# check the memory after # of items merged
|
|
|
|
self.batch = batch
|
|
|
|
# scale is used to scale down the hash of key for recursive hash map
|
|
|
|
self.scale = scale
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
# un-partitioned merged data
|
2014-07-25 01:53:47 -04:00
|
|
|
self.data = {}
|
|
|
|
# partitioned merged data, list of dicts
|
|
|
|
self.pdata = []
|
|
|
|
# number of chunks dumped into disks
|
|
|
|
self.spills = 0
|
|
|
|
# randomize the hash of key, id(o) is the address of o (aligned by 8)
|
|
|
|
self._seed = id(self) + 7
|
|
|
|
|
|
|
|
def _get_spill_dir(self, n):
|
|
|
|
""" Choose one directory for spill by number n """
|
|
|
|
return os.path.join(self.localdirs[n % len(self.localdirs)], str(n))
|
|
|
|
|
|
|
|
def _next_limit(self):
|
|
|
|
"""
|
|
|
|
Return the next memory limit. If the memory is not released
|
|
|
|
after spilling, it will dump the data only when the used memory
|
|
|
|
starts to increase.
|
|
|
|
"""
|
|
|
|
return max(self.memory_limit, get_used_memory() * 1.05)
|
|
|
|
|
|
|
|
def mergeValues(self, iterator):
|
|
|
|
""" Combine the items by creator and combiner """
|
|
|
|
# speedup attribute lookup
|
|
|
|
creator, comb = self.agg.createCombiner, self.agg.mergeValue
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
c, data, pdata, hfun, batch = 0, self.data, self.pdata, self._partition, self.batch
|
|
|
|
limit = self.memory_limit
|
2014-07-25 01:53:47 -04:00
|
|
|
|
|
|
|
for k, v in iterator:
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
d = pdata[hfun(k)] if pdata else data
|
2014-07-25 01:53:47 -04:00
|
|
|
d[k] = comb(d[k], v) if k in d else creator(v)
|
|
|
|
|
|
|
|
c += 1
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
if c >= batch:
|
|
|
|
if get_used_memory() >= limit:
|
|
|
|
self._spill()
|
|
|
|
limit = self._next_limit()
|
|
|
|
batch /= 2
|
|
|
|
c = 0
|
|
|
|
else:
|
|
|
|
batch *= 1.5
|
|
|
|
|
|
|
|
if get_used_memory() >= limit:
|
|
|
|
self._spill()
|
2014-07-25 01:53:47 -04:00
|
|
|
|
|
|
|
def _partition(self, key):
|
|
|
|
""" Return the partition for key """
|
|
|
|
return hash((key, self._seed)) % self.partitions
|
|
|
|
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
def _object_size(self, obj):
|
|
|
|
""" How much of memory for this obj, assume that all the objects
|
|
|
|
consume similar bytes of memory
|
|
|
|
"""
|
|
|
|
return 1
|
2014-07-25 01:53:47 -04:00
|
|
|
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
def mergeCombiners(self, iterator, limit=None):
|
2014-07-25 01:53:47 -04:00
|
|
|
""" Merge (K,V) pair by mergeCombiner """
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
if limit is None:
|
|
|
|
limit = self.memory_limit
|
2014-07-25 01:53:47 -04:00
|
|
|
# speedup attribute lookup
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
comb, hfun, objsize = self.agg.mergeCombiners, self._partition, self._object_size
|
|
|
|
c, data, pdata, batch = 0, self.data, self.pdata, self.batch
|
2014-07-25 01:53:47 -04:00
|
|
|
for k, v in iterator:
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
d = pdata[hfun(k)] if pdata else data
|
2014-07-25 01:53:47 -04:00
|
|
|
d[k] = comb(d[k], v) if k in d else v
|
|
|
|
if not limit:
|
|
|
|
continue
|
|
|
|
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
c += objsize(v)
|
|
|
|
if c > batch:
|
|
|
|
if get_used_memory() > limit:
|
|
|
|
self._spill()
|
|
|
|
limit = self._next_limit()
|
|
|
|
batch /= 2
|
|
|
|
c = 0
|
|
|
|
else:
|
|
|
|
batch *= 1.5
|
|
|
|
|
|
|
|
if limit and get_used_memory() >= limit:
|
|
|
|
self._spill()
|
2014-07-25 01:53:47 -04:00
|
|
|
|
|
|
|
def _spill(self):
|
|
|
|
"""
|
|
|
|
dump already partitioned data into disks.
|
|
|
|
|
|
|
|
It will dump the data in batch for better performance.
|
|
|
|
"""
|
2014-09-14 01:31:21 -04:00
|
|
|
global MemoryBytesSpilled, DiskBytesSpilled
|
2014-07-25 01:53:47 -04:00
|
|
|
path = self._get_spill_dir(self.spills)
|
|
|
|
if not os.path.exists(path):
|
|
|
|
os.makedirs(path)
|
|
|
|
|
2014-09-14 01:31:21 -04:00
|
|
|
used_memory = get_used_memory()
|
2014-07-25 01:53:47 -04:00
|
|
|
if not self.pdata:
|
|
|
|
# The data has not been partitioned, it will iterator the
|
|
|
|
# dataset once, write them into different files, has no
|
|
|
|
# additional memory. It only called when the memory goes
|
|
|
|
# above limit at the first time.
|
|
|
|
|
|
|
|
# open all the files for writing
|
2015-04-16 19:20:57 -04:00
|
|
|
streams = [open(os.path.join(path, str(i)), 'wb')
|
2014-07-25 01:53:47 -04:00
|
|
|
for i in range(self.partitions)]
|
|
|
|
|
2015-04-16 19:20:57 -04:00
|
|
|
for k, v in self.data.items():
|
2014-07-25 01:53:47 -04:00
|
|
|
h = self._partition(k)
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
# put one item in batch, make it compatible with load_stream
|
2014-07-25 01:53:47 -04:00
|
|
|
# it will increase the memory if dump them in batch
|
|
|
|
self.serializer.dump_stream([(k, v)], streams[h])
|
|
|
|
|
|
|
|
for s in streams:
|
2014-09-14 01:31:21 -04:00
|
|
|
DiskBytesSpilled += s.tell()
|
2014-07-25 01:53:47 -04:00
|
|
|
s.close()
|
|
|
|
|
|
|
|
self.data.clear()
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
self.pdata.extend([{} for i in range(self.partitions)])
|
2014-07-25 01:53:47 -04:00
|
|
|
|
|
|
|
else:
|
|
|
|
for i in range(self.partitions):
|
|
|
|
p = os.path.join(path, str(i))
|
2015-04-16 19:20:57 -04:00
|
|
|
with open(p, "wb") as f:
|
2014-07-25 01:53:47 -04:00
|
|
|
# dump items in batch
|
2015-04-16 19:20:57 -04:00
|
|
|
self.serializer.dump_stream(iter(self.pdata[i].items()), f)
|
2014-07-25 01:53:47 -04:00
|
|
|
self.pdata[i].clear()
|
2014-09-14 01:31:21 -04:00
|
|
|
DiskBytesSpilled += os.path.getsize(p)
|
2014-07-25 01:53:47 -04:00
|
|
|
|
|
|
|
self.spills += 1
|
2014-08-06 15:58:24 -04:00
|
|
|
gc.collect() # release the memory as much as possible
|
2015-05-26 11:35:39 -04:00
|
|
|
MemoryBytesSpilled += max(used_memory - get_used_memory(), 0) << 20
|
2014-07-25 01:53:47 -04:00
|
|
|
|
2015-04-16 19:20:57 -04:00
|
|
|
def items(self):
|
2014-07-25 01:53:47 -04:00
|
|
|
""" Return all merged items as iterator """
|
|
|
|
if not self.pdata and not self.spills:
|
2015-04-16 19:20:57 -04:00
|
|
|
return iter(self.data.items())
|
2014-07-25 01:53:47 -04:00
|
|
|
return self._external_items()
|
|
|
|
|
|
|
|
def _external_items(self):
|
|
|
|
""" Return all partitioned items as iterator """
|
|
|
|
assert not self.data
|
|
|
|
if any(self.pdata):
|
|
|
|
self._spill()
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
# disable partitioning and spilling when merge combiners from disk
|
|
|
|
self.pdata = []
|
2014-07-25 01:53:47 -04:00
|
|
|
|
|
|
|
try:
|
|
|
|
for i in range(self.partitions):
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
for v in self._merged_items(i):
|
2014-07-25 01:53:47 -04:00
|
|
|
yield v
|
|
|
|
self.data.clear()
|
|
|
|
|
|
|
|
# remove the merged partition
|
|
|
|
for j in range(self.spills):
|
|
|
|
path = self._get_spill_dir(j)
|
|
|
|
os.remove(os.path.join(path, str(i)))
|
|
|
|
finally:
|
|
|
|
self._cleanup()
|
|
|
|
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
def _merged_items(self, index):
|
|
|
|
self.data = {}
|
|
|
|
limit = self._next_limit()
|
|
|
|
for j in range(self.spills):
|
|
|
|
path = self._get_spill_dir(j)
|
|
|
|
p = os.path.join(path, str(index))
|
|
|
|
# do not check memory during merging
|
2015-04-16 19:20:57 -04:00
|
|
|
with open(p, "rb") as f:
|
|
|
|
self.mergeCombiners(self.serializer.load_stream(f), 0)
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
|
|
|
|
# limit the total partitions
|
|
|
|
if (self.scale * self.partitions < self.MAX_TOTAL_PARTITIONS
|
|
|
|
and j < self.spills - 1
|
|
|
|
and get_used_memory() > limit):
|
|
|
|
self.data.clear() # will read from disk again
|
|
|
|
gc.collect() # release the memory as much as possible
|
|
|
|
return self._recursive_merged_items(index)
|
2014-07-25 01:53:47 -04:00
|
|
|
|
2015-04-16 19:20:57 -04:00
|
|
|
return self.data.items()
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
|
|
|
|
def _recursive_merged_items(self, index):
|
2014-07-25 01:53:47 -04:00
|
|
|
"""
|
|
|
|
merge the partitioned items and return the as iterator
|
|
|
|
|
|
|
|
If one partition can not be fit in memory, then them will be
|
|
|
|
partitioned and merged recursively.
|
|
|
|
"""
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
subdirs = [os.path.join(d, "parts", str(index)) for d in self.localdirs]
|
|
|
|
m = ExternalMerger(self.agg, self.memory_limit, self.serializer, subdirs,
|
|
|
|
self.scale * self.partitions, self.partitions, self.batch)
|
|
|
|
m.pdata = [{} for _ in range(self.partitions)]
|
|
|
|
limit = self._next_limit()
|
|
|
|
|
|
|
|
for j in range(self.spills):
|
|
|
|
path = self._get_spill_dir(j)
|
|
|
|
p = os.path.join(path, str(index))
|
2015-04-16 19:20:57 -04:00
|
|
|
with open(p, 'rb') as f:
|
|
|
|
m.mergeCombiners(self.serializer.load_stream(f), 0)
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
|
|
|
|
if get_used_memory() > limit:
|
|
|
|
m._spill()
|
|
|
|
limit = self._next_limit()
|
2014-07-25 01:53:47 -04:00
|
|
|
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
return m._external_items()
|
2014-07-25 01:53:47 -04:00
|
|
|
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
def _cleanup(self):
|
|
|
|
""" Clean up all the files in disks """
|
|
|
|
for d in self.localdirs:
|
|
|
|
shutil.rmtree(d, True)
|
2014-07-25 01:53:47 -04:00
|
|
|
|
|
|
|
|
2014-08-26 19:57:40 -04:00
|
|
|
class ExternalSorter(object):
|
|
|
|
"""
|
|
|
|
ExtenalSorter will divide the elements into chunks, sort them in
|
|
|
|
memory and dump them into disks, finally merge them back.
|
|
|
|
|
|
|
|
The spilling will only happen when the used memory goes above
|
|
|
|
the limit.
|
|
|
|
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
|
2014-08-26 19:57:40 -04:00
|
|
|
>>> sorter = ExternalSorter(1) # 1M
|
|
|
|
>>> import random
|
2015-04-16 19:20:57 -04:00
|
|
|
>>> l = list(range(1024))
|
2014-08-26 19:57:40 -04:00
|
|
|
>>> random.shuffle(l)
|
|
|
|
>>> sorted(l) == list(sorter.sorted(l))
|
|
|
|
True
|
|
|
|
>>> sorted(l) == list(sorter.sorted(l, key=lambda x: -x, reverse=True))
|
|
|
|
True
|
|
|
|
"""
|
|
|
|
def __init__(self, memory_limit, serializer=None):
|
|
|
|
self.memory_limit = memory_limit
|
|
|
|
self.local_dirs = _get_local_dirs("sort")
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
self.serializer = _compressed_serializer(serializer)
|
2014-08-26 19:57:40 -04:00
|
|
|
|
|
|
|
def _get_path(self, n):
|
|
|
|
""" Choose one directory for spill by number n """
|
|
|
|
d = self.local_dirs[n % len(self.local_dirs)]
|
|
|
|
if not os.path.exists(d):
|
|
|
|
os.makedirs(d)
|
|
|
|
return os.path.join(d, str(n))
|
|
|
|
|
2014-11-19 18:45:37 -05:00
|
|
|
def _next_limit(self):
|
|
|
|
"""
|
|
|
|
Return the next memory limit. If the memory is not released
|
|
|
|
after spilling, it will dump the data only when the used memory
|
|
|
|
starts to increase.
|
|
|
|
"""
|
|
|
|
return max(self.memory_limit, get_used_memory() * 1.05)
|
|
|
|
|
2014-08-26 19:57:40 -04:00
|
|
|
def sorted(self, iterator, key=None, reverse=False):
|
|
|
|
"""
|
|
|
|
Sort the elements in iterator, do external sort when the memory
|
|
|
|
goes above the limit.
|
|
|
|
"""
|
2014-09-14 01:31:21 -04:00
|
|
|
global MemoryBytesSpilled, DiskBytesSpilled
|
2015-06-18 16:45:58 -04:00
|
|
|
batch, limit = 100, self._next_limit()
|
2014-08-26 19:57:40 -04:00
|
|
|
chunks, current_chunk = [], []
|
|
|
|
iterator = iter(iterator)
|
|
|
|
while True:
|
|
|
|
# pick elements in batch
|
|
|
|
chunk = list(itertools.islice(iterator, batch))
|
|
|
|
current_chunk.extend(chunk)
|
|
|
|
if len(chunk) < batch:
|
|
|
|
break
|
|
|
|
|
2014-09-14 01:31:21 -04:00
|
|
|
used_memory = get_used_memory()
|
2015-04-21 20:49:55 -04:00
|
|
|
if used_memory > limit:
|
2014-08-26 19:57:40 -04:00
|
|
|
# sort them inplace will save memory
|
|
|
|
current_chunk.sort(key=key, reverse=reverse)
|
|
|
|
path = self._get_path(len(chunks))
|
2015-04-16 19:20:57 -04:00
|
|
|
with open(path, 'wb') as f:
|
2014-08-26 19:57:40 -04:00
|
|
|
self.serializer.dump_stream(current_chunk, f)
|
2015-04-16 19:20:57 -04:00
|
|
|
|
|
|
|
def load(f):
|
|
|
|
for v in self.serializer.load_stream(f):
|
|
|
|
yield v
|
|
|
|
# close the file explicit once we consume all the items
|
|
|
|
# to avoid ResourceWarning in Python3
|
|
|
|
f.close()
|
|
|
|
chunks.append(load(open(path, 'rb')))
|
2014-08-26 19:57:40 -04:00
|
|
|
current_chunk = []
|
2015-05-26 11:35:39 -04:00
|
|
|
MemoryBytesSpilled += max(used_memory - get_used_memory(), 0) << 20
|
2014-09-14 01:31:21 -04:00
|
|
|
DiskBytesSpilled += os.path.getsize(path)
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
os.unlink(path) # data will be deleted after close
|
2014-08-26 19:57:40 -04:00
|
|
|
|
|
|
|
elif not chunks:
|
2015-04-21 20:49:55 -04:00
|
|
|
batch = min(int(batch * 1.5), 10000)
|
2014-08-26 19:57:40 -04:00
|
|
|
|
|
|
|
current_chunk.sort(key=key, reverse=reverse)
|
|
|
|
if not chunks:
|
|
|
|
return current_chunk
|
|
|
|
|
|
|
|
if current_chunk:
|
|
|
|
chunks.append(iter(current_chunk))
|
|
|
|
|
|
|
|
return heapq.merge(chunks, key=key, reverse=reverse)
|
|
|
|
|
|
|
|
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
class ExternalList(object):
|
|
|
|
"""
|
|
|
|
ExternalList can have many items which cannot be hold in memory in
|
|
|
|
the same time.
|
|
|
|
|
2015-04-16 19:20:57 -04:00
|
|
|
>>> l = ExternalList(list(range(100)))
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
>>> len(l)
|
|
|
|
100
|
|
|
|
>>> l.append(10)
|
|
|
|
>>> len(l)
|
|
|
|
101
|
|
|
|
>>> for i in range(20240):
|
|
|
|
... l.append(i)
|
|
|
|
>>> len(l)
|
|
|
|
20341
|
|
|
|
>>> import pickle
|
|
|
|
>>> l2 = pickle.loads(pickle.dumps(l))
|
|
|
|
>>> len(l2)
|
|
|
|
20341
|
|
|
|
>>> list(l2)[100]
|
|
|
|
10
|
|
|
|
"""
|
|
|
|
LIMIT = 10240
|
|
|
|
|
|
|
|
def __init__(self, values):
|
|
|
|
self.values = values
|
|
|
|
self.count = len(values)
|
|
|
|
self._file = None
|
|
|
|
self._ser = None
|
|
|
|
|
|
|
|
def __getstate__(self):
|
|
|
|
if self._file is not None:
|
|
|
|
self._file.flush()
|
2015-04-16 19:20:57 -04:00
|
|
|
with os.fdopen(os.dup(self._file.fileno()), "rb") as f:
|
|
|
|
f.seek(0)
|
|
|
|
serialized = f.read()
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
else:
|
2015-04-16 19:20:57 -04:00
|
|
|
serialized = b''
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
return self.values, self.count, serialized
|
|
|
|
|
|
|
|
def __setstate__(self, item):
|
|
|
|
self.values, self.count, serialized = item
|
|
|
|
if serialized:
|
|
|
|
self._open_file()
|
|
|
|
self._file.write(serialized)
|
|
|
|
else:
|
|
|
|
self._file = None
|
|
|
|
self._ser = None
|
|
|
|
|
|
|
|
def __iter__(self):
|
|
|
|
if self._file is not None:
|
|
|
|
self._file.flush()
|
|
|
|
# read all items from disks first
|
2015-04-16 19:20:57 -04:00
|
|
|
with os.fdopen(os.dup(self._file.fileno()), 'rb') as f:
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
f.seek(0)
|
|
|
|
for v in self._ser.load_stream(f):
|
|
|
|
yield v
|
|
|
|
|
|
|
|
for v in self.values:
|
|
|
|
yield v
|
|
|
|
|
|
|
|
def __len__(self):
|
|
|
|
return self.count
|
|
|
|
|
|
|
|
def append(self, value):
|
|
|
|
self.values.append(value)
|
|
|
|
self.count += 1
|
|
|
|
# dump them into disk if the key is huge
|
|
|
|
if len(self.values) >= self.LIMIT:
|
|
|
|
self._spill()
|
|
|
|
|
|
|
|
def _open_file(self):
|
|
|
|
dirs = _get_local_dirs("objects")
|
|
|
|
d = dirs[id(self) % len(dirs)]
|
|
|
|
if not os.path.exists(d):
|
|
|
|
os.makedirs(d)
|
2015-04-16 19:20:57 -04:00
|
|
|
p = os.path.join(d, str(id(self)))
|
2015-07-30 01:30:49 -04:00
|
|
|
self._file = open(p, "w+b", 65536)
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
self._ser = BatchedSerializer(CompressedSerializer(PickleSerializer()), 1024)
|
|
|
|
os.unlink(p)
|
|
|
|
|
2015-04-16 19:20:57 -04:00
|
|
|
def __del__(self):
|
|
|
|
if self._file:
|
|
|
|
self._file.close()
|
|
|
|
self._file = None
|
|
|
|
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
def _spill(self):
|
|
|
|
""" dump the values into disk """
|
|
|
|
global MemoryBytesSpilled, DiskBytesSpilled
|
|
|
|
if self._file is None:
|
|
|
|
self._open_file()
|
|
|
|
|
|
|
|
used_memory = get_used_memory()
|
|
|
|
pos = self._file.tell()
|
|
|
|
self._ser.dump_stream(self.values, self._file)
|
|
|
|
self.values = []
|
|
|
|
gc.collect()
|
|
|
|
DiskBytesSpilled += self._file.tell() - pos
|
2015-05-26 11:35:39 -04:00
|
|
|
MemoryBytesSpilled += max(used_memory - get_used_memory(), 0) << 20
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
|
|
|
|
|
|
|
|
class ExternalListOfList(ExternalList):
|
|
|
|
"""
|
|
|
|
An external list for list.
|
|
|
|
|
|
|
|
>>> l = ExternalListOfList([[i, i] for i in range(100)])
|
|
|
|
>>> len(l)
|
|
|
|
200
|
|
|
|
>>> l.append(range(10))
|
|
|
|
>>> len(l)
|
|
|
|
210
|
|
|
|
>>> len(list(l))
|
|
|
|
210
|
|
|
|
"""
|
|
|
|
|
|
|
|
def __init__(self, values):
|
|
|
|
ExternalList.__init__(self, values)
|
|
|
|
self.count = sum(len(i) for i in values)
|
|
|
|
|
|
|
|
def append(self, value):
|
|
|
|
ExternalList.append(self, value)
|
|
|
|
# already counted 1 in ExternalList.append
|
|
|
|
self.count += len(value) - 1
|
|
|
|
|
|
|
|
def __iter__(self):
|
|
|
|
for values in ExternalList.__iter__(self):
|
|
|
|
for v in values:
|
|
|
|
yield v
|
|
|
|
|
|
|
|
|
|
|
|
class GroupByKey(object):
|
|
|
|
"""
|
|
|
|
Group a sorted iterator as [(k1, it1), (k2, it2), ...]
|
|
|
|
|
2015-04-16 19:20:57 -04:00
|
|
|
>>> k = [i // 3 for i in range(6)]
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
>>> v = [[i] for i in range(6)]
|
2015-04-16 19:20:57 -04:00
|
|
|
>>> g = GroupByKey(zip(k, v))
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
>>> [(k, list(it)) for k, it in g]
|
|
|
|
[(0, [0, 1, 2]), (1, [3, 4, 5])]
|
|
|
|
"""
|
|
|
|
|
|
|
|
def __init__(self, iterator):
|
2015-04-16 19:20:57 -04:00
|
|
|
self.iterator = iterator
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
|
|
|
|
def __iter__(self):
|
2015-04-16 19:20:57 -04:00
|
|
|
key, values = None, None
|
|
|
|
for k, v in self.iterator:
|
|
|
|
if values is not None and k == key:
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
values.append(v)
|
2015-04-16 19:20:57 -04:00
|
|
|
else:
|
|
|
|
if values is not None:
|
|
|
|
yield (key, values)
|
|
|
|
key = k
|
|
|
|
values = ExternalListOfList([v])
|
|
|
|
if values is not None:
|
|
|
|
yield (key, values)
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
|
|
|
|
|
|
|
|
class ExternalGroupBy(ExternalMerger):
|
|
|
|
|
|
|
|
"""
|
|
|
|
Group by the items by key. If any partition of them can not been
|
|
|
|
hold in memory, it will do sort based group by.
|
|
|
|
|
|
|
|
This class works as follows:
|
|
|
|
|
|
|
|
- It repeatedly group the items by key and save them in one dict in
|
|
|
|
memory.
|
|
|
|
|
|
|
|
- When the used memory goes above memory limit, it will split
|
|
|
|
the combined data into partitions by hash code, dump them
|
|
|
|
into disk, one file per partition. If the number of keys
|
|
|
|
in one partitions is smaller than 1000, it will sort them
|
|
|
|
by key before dumping into disk.
|
|
|
|
|
|
|
|
- Then it goes through the rest of the iterator, group items
|
|
|
|
by key into different dict by hash. Until the used memory goes over
|
|
|
|
memory limit, it dump all the dicts into disks, one file per
|
|
|
|
dict. Repeat this again until combine all the items. It
|
|
|
|
also will try to sort the items by key in each partition
|
|
|
|
before dumping into disks.
|
|
|
|
|
|
|
|
- It will yield the grouped items partitions by partitions.
|
|
|
|
If the data in one partitions can be hold in memory, then it
|
|
|
|
will load and combine them in memory and yield.
|
|
|
|
|
|
|
|
- If the dataset in one partition cannot be hold in memory,
|
|
|
|
it will sort them first. If all the files are already sorted,
|
|
|
|
it merge them by heap.merge(), so it will do external sort
|
|
|
|
for all the files.
|
|
|
|
|
|
|
|
- After sorting, `GroupByKey` class will put all the continuous
|
|
|
|
items with the same key as a group, yield the values as
|
|
|
|
an iterator.
|
|
|
|
"""
|
|
|
|
SORT_KEY_LIMIT = 1000
|
|
|
|
|
|
|
|
def flattened_serializer(self):
|
|
|
|
assert isinstance(self.serializer, BatchedSerializer)
|
|
|
|
ser = self.serializer
|
|
|
|
return FlattenedValuesSerializer(ser, 20)
|
|
|
|
|
|
|
|
def _object_size(self, obj):
|
|
|
|
return len(obj)
|
|
|
|
|
|
|
|
def _spill(self):
|
|
|
|
"""
|
|
|
|
dump already partitioned data into disks.
|
|
|
|
"""
|
|
|
|
global MemoryBytesSpilled, DiskBytesSpilled
|
|
|
|
path = self._get_spill_dir(self.spills)
|
|
|
|
if not os.path.exists(path):
|
|
|
|
os.makedirs(path)
|
|
|
|
|
|
|
|
used_memory = get_used_memory()
|
|
|
|
if not self.pdata:
|
|
|
|
# The data has not been partitioned, it will iterator the
|
|
|
|
# data once, write them into different files, has no
|
|
|
|
# additional memory. It only called when the memory goes
|
|
|
|
# above limit at the first time.
|
|
|
|
|
|
|
|
# open all the files for writing
|
2015-04-16 19:20:57 -04:00
|
|
|
streams = [open(os.path.join(path, str(i)), 'wb')
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
for i in range(self.partitions)]
|
|
|
|
|
|
|
|
# If the number of keys is small, then the overhead of sort is small
|
|
|
|
# sort them before dumping into disks
|
|
|
|
self._sorted = len(self.data) < self.SORT_KEY_LIMIT
|
|
|
|
if self._sorted:
|
|
|
|
self.serializer = self.flattened_serializer()
|
|
|
|
for k in sorted(self.data.keys()):
|
|
|
|
h = self._partition(k)
|
|
|
|
self.serializer.dump_stream([(k, self.data[k])], streams[h])
|
|
|
|
else:
|
2015-04-16 19:20:57 -04:00
|
|
|
for k, v in self.data.items():
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
h = self._partition(k)
|
|
|
|
self.serializer.dump_stream([(k, v)], streams[h])
|
|
|
|
|
|
|
|
for s in streams:
|
|
|
|
DiskBytesSpilled += s.tell()
|
|
|
|
s.close()
|
|
|
|
|
|
|
|
self.data.clear()
|
|
|
|
# self.pdata is cached in `mergeValues` and `mergeCombiners`
|
|
|
|
self.pdata.extend([{} for i in range(self.partitions)])
|
|
|
|
|
|
|
|
else:
|
|
|
|
for i in range(self.partitions):
|
|
|
|
p = os.path.join(path, str(i))
|
2015-04-16 19:20:57 -04:00
|
|
|
with open(p, "wb") as f:
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
# dump items in batch
|
|
|
|
if self._sorted:
|
|
|
|
# sort by key only (stable)
|
2015-04-16 19:20:57 -04:00
|
|
|
sorted_items = sorted(self.pdata[i].items(), key=operator.itemgetter(0))
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
self.serializer.dump_stream(sorted_items, f)
|
|
|
|
else:
|
2015-04-16 19:20:57 -04:00
|
|
|
self.serializer.dump_stream(self.pdata[i].items(), f)
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
self.pdata[i].clear()
|
|
|
|
DiskBytesSpilled += os.path.getsize(p)
|
|
|
|
|
|
|
|
self.spills += 1
|
|
|
|
gc.collect() # release the memory as much as possible
|
2015-05-26 11:35:39 -04:00
|
|
|
MemoryBytesSpilled += max(used_memory - get_used_memory(), 0) << 20
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
|
|
|
|
def _merged_items(self, index):
|
|
|
|
size = sum(os.path.getsize(os.path.join(self._get_spill_dir(j), str(index)))
|
|
|
|
for j in range(self.spills))
|
|
|
|
# if the memory can not hold all the partition,
|
|
|
|
# then use sort based merge. Because of compression,
|
|
|
|
# the data on disks will be much smaller than needed memory
|
2015-04-16 19:20:57 -04:00
|
|
|
if size >= self.memory_limit << 17: # * 1M / 8
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
return self._merge_sorted_items(index)
|
|
|
|
|
|
|
|
self.data = {}
|
|
|
|
for j in range(self.spills):
|
|
|
|
path = self._get_spill_dir(j)
|
|
|
|
p = os.path.join(path, str(index))
|
|
|
|
# do not check memory during merging
|
2015-04-16 19:20:57 -04:00
|
|
|
with open(p, "rb") as f:
|
|
|
|
self.mergeCombiners(self.serializer.load_stream(f), 0)
|
|
|
|
return self.data.items()
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
|
|
|
|
def _merge_sorted_items(self, index):
|
|
|
|
""" load a partition from disk, then sort and group by key """
|
|
|
|
def load_partition(j):
|
|
|
|
path = self._get_spill_dir(j)
|
|
|
|
p = os.path.join(path, str(index))
|
2015-04-16 19:20:57 -04:00
|
|
|
with open(p, 'rb', 65536) as f:
|
|
|
|
for v in self.serializer.load_stream(f):
|
|
|
|
yield v
|
[SPARK-3074] [PySpark] support groupByKey() with single huge key
This patch change groupByKey() to use external sort based approach, so it can support single huge key.
For example, it can group by a dataset including one hot key with 40 millions values (strings), using 500M memory for Python worker, finished in about 2 minutes. (it will need 6G memory in hash based approach).
During groupByKey(), it will do in-memory groupBy first. If the dataset can not fit in memory, then data will be partitioned by hash. If one partition still can not fit in memory, it will switch to sort based groupBy().
Author: Davies Liu <davies.liu@gmail.com>
Author: Davies Liu <davies@databricks.com>
Closes #1977 from davies/groupby and squashes the following commits:
af3713a [Davies Liu] make sure it's iterator
67772dd [Davies Liu] fix tests
e78c15c [Davies Liu] address comments
0b0fde8 [Davies Liu] address comments
0dcf320 [Davies Liu] address comments, rollback changes in ResultIterable
e3b8eab [Davies Liu] fix narrow dependency
2a1857a [Davies Liu] typo
d2f053b [Davies Liu] add repr for FlattedValuesSerializer
c6a2f8d [Davies Liu] address comments
9e2df24 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
2b9c261 [Davies Liu] fix typo in comments
70aadcd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
a14b4bd [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
ab5515b [Davies Liu] Merge branch 'master' into groupby
651f891 [Davies Liu] simplify GroupByKey
1578f2e [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
1f69f93 [Davies Liu] fix tests
0d3395f [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
341f1e0 [Davies Liu] add comments, refactor
47918b8 [Davies Liu] remove unused code
6540948 [Davies Liu] address comments:
17f4ec6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into groupby
4d4bc86 [Davies Liu] bugfix
8ef965e [Davies Liu] Merge branch 'master' into groupby
fbc504a [Davies Liu] Merge branch 'master' into groupby
779ed03 [Davies Liu] fix merge conflict
2c1d05b [Davies Liu] refactor, minor turning
b48cda5 [Davies Liu] Merge branch 'master' into groupby
85138e6 [Davies Liu] Merge branch 'master' into groupby
acd8e1b [Davies Liu] fix memory when groupByKey().count()
905b233 [Davies Liu] Merge branch 'sort' into groupby
1f075ed [Davies Liu] Merge branch 'master' into sort
4b07d39 [Davies Liu] compress the data while spilling
0a081c6 [Davies Liu] Merge branch 'master' into groupby
f157fe7 [Davies Liu] Merge branch 'sort' into groupby
eb53ca6 [Davies Liu] Merge branch 'master' into sort
b2dc3bf [Davies Liu] Merge branch 'sort' into groupby
644abaf [Davies Liu] add license in LICENSE
19f7873 [Davies Liu] improve tests
11ba318 [Davies Liu] typo
085aef8 [Davies Liu] Merge branch 'master' into groupby
3ee58e5 [Davies Liu] switch to sort based groupBy, based on size of data
1ea0669 [Davies Liu] choose sort based groupByKey() automatically
b40bae7 [Davies Liu] bugfix
efa23df [Davies Liu] refactor, add spark.shuffle.sort=False
250be4e [Davies Liu] flatten the combined values when dumping into disks
d05060d [Davies Liu] group the same key before shuffle, reduce the comparison during sorting
083d842 [Davies Liu] sorted based groupByKey()
55602ee [Davies Liu] use external sort in sortBy() and sortByKey()
2015-04-09 20:07:23 -04:00
|
|
|
|
|
|
|
disk_items = [load_partition(j) for j in range(self.spills)]
|
|
|
|
|
|
|
|
if self._sorted:
|
|
|
|
# all the partitions are already sorted
|
|
|
|
sorted_items = heapq.merge(disk_items, key=operator.itemgetter(0))
|
|
|
|
|
|
|
|
else:
|
|
|
|
# Flatten the combined values, so it will not consume huge
|
|
|
|
# memory during merging sort.
|
|
|
|
ser = self.flattened_serializer()
|
|
|
|
sorter = ExternalSorter(self.memory_limit, ser)
|
|
|
|
sorted_items = sorter.sorted(itertools.chain(*disk_items),
|
|
|
|
key=operator.itemgetter(0))
|
|
|
|
return ((k, vs) for k, vs in GroupByKey(sorted_items))
|
|
|
|
|
|
|
|
|
2014-07-25 01:53:47 -04:00
|
|
|
if __name__ == "__main__":
|
|
|
|
import doctest
|
2015-06-26 11:12:22 -04:00
|
|
|
(failure_count, test_count) = doctest.testmod()
|
|
|
|
if failure_count:
|
2018-03-08 06:38:34 -05:00
|
|
|
sys.exit(-1)
|