[SPARK-3478] [PySpark] Profile the Python tasks by davies · Pull Request #2351 · apache/spark

davies · 2014-09-11T01:09:29Z

This patch add profiling support for PySpark, it will show the profiling results
before the driver exits, here is one example:

============================================================
Profile of RDD<id=3>
============================================================
5146507 function calls (5146487 primitive calls) in 71.094 seconds
Ordered by: internal time, cumulative time
ncalls tottime percall cumtime percall filename:lineno(function)
5144576 68.331 0.000 68.331 0.000 statcounter.py:44(merge)
20 2.735 0.137 71.071 3.554 statcounter.py:33(__init__)
20 0.017 0.001 0.017 0.001 {cPickle.dumps}
1024 0.003 0.000 0.003 0.000 t.py:16(<lambda>)
20 0.001 0.000 0.001 0.000 {reduce}
21 0.001 0.000 0.001 0.000 {cPickle.loads}
20 0.001 0.000 0.001 0.000 copy_reg.py:95(_slotnames)
41 0.001 0.000 0.001 0.000 serializers.py:461(read_int)
40 0.001 0.000 0.002 0.000 serializers.py:179(_batched)
62 0.000 0.000 0.000 0.000 {method 'read' of 'file' objects}
20 0.000 0.000 71.072 3.554 rdd.py:863(<lambda>)
20 0.000 0.000 0.001 0.000 serializers.py:198(load_stream)
40/20 0.000 0.000 71.072 3.554 rdd.py:2093(pipeline_func)
41 0.000 0.000 0.002 0.000 serializers.py:130(load_stream)
40 0.000 0.000 71.072 1.777 rdd.py:304(func)
20 0.000 0.000 71.094 3.555 worker.py:82(process)

Also, use can show profile result manually by sc.show_profiles() or dump it into disk
by sc.dump_profiles(path), such as

>>>sc._conf.set("spark.python.profile", "true")
>>>rdd=sc.parallelize(range(100)).map(str)
>>>rdd.count()
100>>>sc.show_profiles()
============================================================ProfileofRDD<id=1>============================================================284functioncalls (276primitivecalls) in0.001secondsOrderedby: internaltime, cumulativetimencallstottimepercallcumtimepercallfilename:lineno(function)
40.0000.0000.0000.000serializers.py:198(load_stream)
40.0000.0000.0000.000 {reduce}
12/40.0000.0000.0010.000rdd.py:2092(pipeline_func)
40.0000.0000.0000.000 {cPickle.loads}
40.0000.0000.0000.000 {cPickle.dumps}
1040.0000.0000.0000.000rdd.py:852(<genexpr>)
80.0000.0000.0000.000serializers.py:461(read_int)
120.0000.0000.0000.000rdd.py:303(func)

The profiling is disabled by default, can be enabled by "spark.python.profile=true".

Also, users can dump the results into disks automatically for future analysis, by "spark.python.profile.dump=path_to_dump"

SparkQA · 2014-09-11T01:16:18Z

QA tests have started for PR 2351 at commit 4b20494.

This patch merges cleanly.

SparkQA · 2014-09-11T01:52:42Z

QA tests have started for PR 2351 at commit 4b20494.

This patch merges cleanly.

SparkQA · 2014-09-11T02:15:44Z

QA tests have finished for PR 2351 at commit 4b20494.

This patch fails unit tests.
This patch merges cleanly.
This patch adds the following public classes (experimental):
- class StatsParam(AccumulatorParam):

SparkQA · 2014-09-11T02:45:50Z

QA tests have finished for PR 2351 at commit 4b20494.

This patch fails unit tests.
This patch merges cleanly.
This patch adds the following public classes (experimental):
- class StatsParam(AccumulatorParam):

SparkQA · 2014-09-11T03:52:48Z

QA tests have started for PR 2351 at commit 0a5b6eb.

This patch merges cleanly.

SparkQA · 2014-09-11T04:49:04Z

QA tests have finished for PR 2351 at commit 0a5b6eb.

This patch fails unit tests.
This patch merges cleanly.
This patch adds the following public classes (experimental):
- class StatsParam(AccumulatorParam):

SparkQA · 2014-09-11T05:24:27Z

QA tests have started for PR 2351 at commit 0a5b6eb.

This patch merges cleanly.

SparkQA · 2014-09-11T06:25:13Z

QA tests have finished for PR 2351 at commit 0a5b6eb.

This patch passes unit tests.
This patch merges cleanly.
This patch adds the following public classes (experimental):
- class StatsParam(AccumulatorParam):

JoshRosen · 2014-09-13T01:04:57Z

Do you think it would be clearer to name this ProfilingStatsParam or PStatsParam?

SparkQA · 2014-09-13T03:19:18Z

QA tests have started for PR 2351 at commit 4f8309d.

This patch merges cleanly.

SparkQA · 2014-09-13T04:25:08Z

QA tests have finished for PR 2351 at commit 4f8309d.

This patch passes unit tests.
This patch merges cleanly.
This patch adds the following public classes (experimental):
- class PStatsParam(AccumulatorParam):

davies · 2014-09-13T05:12:15Z

@JoshRosen I had addressed your comment, also added docs for configs and tests.

I realized that the profile result also can be showed interactively, by rdd.show_profile(), I had updated the PR description for this.

SparkQA · 2014-09-13T05:14:35Z

QA tests have started for PR 2351 at commit 15d6f18.

This patch merges cleanly.

SparkQA · 2014-09-13T05:19:14Z

QA tests have started for PR 2351 at commit c23865c.

This patch merges cleanly.

SparkQA · 2014-09-13T05:20:18Z

QA tests have finished for PR 2351 at commit c23865c.

This patch fails unit tests.
This patch merges cleanly.
This patch adds the following public classes (experimental):
- class TaskCompletionListenerException(errorMessages: Seq[String]) extends Exception
- class PStatsParam(AccumulatorParam):

SparkQA · 2014-09-13T06:25:33Z

QA tests have finished for PR 2351 at commit 15d6f18.

This patch passes unit tests.
This patch merges cleanly.
This patch adds the following public classes (experimental):
- class TaskCompletionListenerException(errorMessages: Seq[String]) extends Exception
- class PStatsParam(AccumulatorParam):

Conflicts: docs/configuration.md

SparkQA · 2014-09-14T04:29:12Z

QA tests have started for PR 2351 at commit 09d02c3.

This patch merges cleanly.

SparkQA · 2014-09-14T05:35:04Z

QA tests have finished for PR 2351 at commit 09d02c3.

This patch passes unit tests.
This patch merges cleanly.
This patch adds the following public classes (experimental):
- class PStatsParam(AccumulatorParam):

Conflicts: python/pyspark/worker.py

SparkQA · 2014-09-17T17:19:19Z

QA tests have started for PR 2351 at commit 116d52a.

This patch merges cleanly.

SparkQA · 2014-09-17T18:25:43Z

QA tests have finished for PR 2351 at commit 116d52a.

This patch passes unit tests.
This patch merges cleanly.
This patch adds the following public classes (experimental):
- class PStatsParam(AccumulatorParam):

Conflicts: python/pyspark/worker.py

SparkQA · 2014-09-23T22:19:27Z

QA tests have started for PR 2351 at commit fb9565b.

This patch merges cleanly.

JoshRosen · 2014-09-24T23:16:20Z

(I killed the test here so that I could re-run it with the newer commits).

SparkQA · 2014-09-24T23:18:05Z

QA tests have started for PR 2351 at commit 2b0daf2.

This patch merges cleanly.

SparkQA · 2014-09-24T23:19:24Z

QA tests have started for PR 2351 at commit 2b0daf2.

This patch merges cleanly.

SparkQA · 2014-09-25T00:24:52Z

QA tests have finished for PR 2351 at commit 2b0daf2.

This patch passes unit tests.
This patch merges cleanly.
This patch adds the following public classes (experimental):
- class PStatsParam(AccumulatorParam):

SparkQA · 2014-09-25T00:27:55Z

QA tests have finished for PR 2351 at commit 2b0daf2.

This patch fails unit tests.
This patch merges cleanly.
This patch adds the following public classes (experimental):
- class PStatsParam(AccumulatorParam):

SparkQA · 2014-09-25T00:27:58Z

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/SparkPullRequestBuilder/20769/

shaneknapp · 2014-09-25T00:28:59Z

jenkins, retest this please

SparkQA · 2014-09-25T00:34:29Z

QA tests have started for PR 2351 at commit 2b0daf2.

This patch merges cleanly.

SparkQA · 2014-09-25T01:38:52Z

QA tests have finished for PR 2351 at commit 2b0daf2.

This patch passes unit tests.
This patch merges cleanly.
This patch adds the following public classes (experimental):
- class PStatsParam(AccumulatorParam):

AmplabJenkins · 2014-09-25T01:38:56Z

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/SparkPullRequestBuilder/20771/

JoshRosen · 2014-09-25T18:10:12Z

This looks good to me. Thanks!

JoshRosen · 2014-09-25T18:39:41Z

I noticed that we don't have any automated tests for show_profiles(), so I tested it manually and found a problem when running this file through spark-submit:

frompysparkimportSparkContext, SparkConfconf=SparkConf()
conf.set("spark.python.profile", "true")
sc=SparkContext(appName="test", conf=conf)
count=sc.parallelize(range(10000)).count()
sc.show_profiles()

This results in:

Traceback (most recent call last):
File "/Users/joshrosen/Documents/spark/test.py", line 6, in <module>
sc.show_profiles()
File "/Users/joshrosen/Documents/Spark/python/pyspark/context.py", line 811, in show_profiles
for i, (id, acc, showed) in self._profile_stats:
ValueError: too many values to unpack
Error in atexit._run_exitfuncs:
Traceback (most recent call last):
File "/Users/joshrosen/anaconda/lib/python2.7/atexit.py", line 24, in _run_exitfuncs
func(*targs, **kargs)
File "/Users/joshrosen/Documents/Spark/python/pyspark/context.py", line 811, in show_profiles
for i, (id, acc, showed) in self._profile_stats:
ValueError: too many values to unpack
Error in sys.exitfunc:
Traceback (most recent call last):
File "/Users/joshrosen/anaconda/lib/python2.7/atexit.py", line 24, in _run_exitfuncs
func(*targs, **kargs)
File "/Users/joshrosen/Documents/Spark/python/pyspark/context.py", line 811, in show_profiles
for i, (id, acc, showed) in self._profile_stats:
ValueError: too many values to unpack

Can we add a test for this, too?

davies · 2014-09-25T21:48:41Z

@JoshRosen sorry for this mistake, fixed.

AmplabJenkins · 2014-09-25T22:02:19Z

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/SparkPullRequestBuilder/20822/

SparkQA · 2014-09-25T22:06:56Z

QA tests have started for PR 2351 at commit 7ef2aa0.

This patch merges cleanly.

SparkQA · 2014-09-25T23:04:27Z

QA tests have finished for PR 2351 at commit 7ef2aa0.

This patch fails unit tests.
This patch merges cleanly.
This patch adds the following public classes (experimental):
- class PStatsParam(AccumulatorParam):

JoshRosen · 2014-09-25T23:40:20Z

Is this still true? It looks like we now use a showed flag to detect whether they've been printed instead of clearing the profiles array.

I think it's true.

It looks like we clear _profile_stats when we perform manual dump_profiles() calls, but not when we call show_profiles(), so it seems like this is half-true (unless I've overlooked something).

if showed is true, it will not be displayed again, but will be dumped.

Ah, right. If it's been manually dumped, the it won't be dumped again when exiting. If it's been manually dumped or displayed, then it won't be displayed when exiting.
This makes sense; sorry for the confusion.

davies · 2014-09-26T16:50:42Z

Thanks for review this, your comments made it much better.

JoshRosen · 2014-09-26T21:47:00Z

Whoops, looks like this failed unit tests and caused a build-break. I'm going to revert it to un-break the build while we investigate.

This patch add profiling support for PySpark, it will show the profiling results before the driver exits, here is one example: ``` ============================================================ Profile of RDD<id=3> ============================================================ 5146507 function calls (5146487 primitive calls) in 71.094 seconds Ordered by: internal time, cumulative time ncalls tottime percall cumtime percall filename:lineno(function) 5144576 68.331 0.000 68.331 0.000 statcounter.py:44(merge) 20 2.735 0.137 71.071 3.554 statcounter.py:33(__init__) 20 0.017 0.001 0.017 0.001 {cPickle.dumps} 1024 0.003 0.000 0.003 0.000 t.py:16(<lambda>) 20 0.001 0.000 0.001 0.000 {reduce} 21 0.001 0.000 0.001 0.000 {cPickle.loads} 20 0.001 0.000 0.001 0.000 copy_reg.py:95(_slotnames) 41 0.001 0.000 0.001 0.000 serializers.py:461(read_int) 40 0.001 0.000 0.002 0.000 serializers.py:179(_batched) 62 0.000 0.000 0.000 0.000 {method 'read' of 'file' objects} 20 0.000 0.000 71.072 3.554 rdd.py:863(<lambda>) 20 0.000 0.000 0.001 0.000 serializers.py:198(load_stream) 40/20 0.000 0.000 71.072 3.554 rdd.py:2093(pipeline_func) 41 0.000 0.000 0.002 0.000 serializers.py:130(load_stream) 40 0.000 0.000 71.072 1.777 rdd.py:304(func) 20 0.000 0.000 71.094 3.555 worker.py:82(process) ``` Also, use can show profile result manually by `sc.show_profiles()` or dump it into disk by `sc.dump_profiles(path)`, such as ```python >>> sc._conf.set("spark.python.profile", "true") >>> rdd = sc.parallelize(range(100)).map(str) >>> rdd.count() 100 >>> sc.show_profiles() ============================================================ Profile of RDD<id=1> ============================================================ 284 function calls (276 primitive calls) in 0.001 seconds Ordered by: internal time, cumulative time ncalls tottime percall cumtime percall filename:lineno(function) 4 0.000 0.000 0.000 0.000 serializers.py:198(load_stream) 4 0.000 0.000 0.000 0.000 {reduce} 12/4 0.000 0.000 0.001 0.000 rdd.py:2092(pipeline_func) 4 0.000 0.000 0.000 0.000 {cPickle.loads} 4 0.000 0.000 0.000 0.000 {cPickle.dumps} 104 0.000 0.000 0.000 0.000 rdd.py:852(<genexpr>) 8 0.000 0.000 0.000 0.000 serializers.py:461(read_int) 12 0.000 0.000 0.000 0.000 rdd.py:303(func) ``` The profiling is disabled by default, can be enabled by "spark.python.profile=true". Also, users can dump the results into disks automatically for future analysis, by "spark.python.profile.dump=path_to_dump" This is bugfix of #2351 cc JoshRosen Author: Davies Liu <davies.liu@gmail.com> Closes#2556 from davies/profiler and squashes the following commits: e68df5a [Davies Liu] Merge branch 'master' of github.com:apache/spark into profiler 858e74c [Davies Liu] compatitable with python 2.6 7ef2aa0 [Davies Liu] bugfix, add tests for show_profiles and dump_profiles() 2b0daf2 [Davies Liu] fix docs 7a56c24 [Davies Liu] bugfix cba9463 [Davies Liu] move show_profiles and dump_profiles to SparkContext fb9565b [Davies Liu] Merge branch 'master' of github.com:apache/spark into profiler 116d52a [Davies Liu] Merge branch 'master' of github.com:apache/spark into profiler 09d02c3 [Davies Liu] Merge branch 'master' into profiler c23865c [Davies Liu] Merge branch 'master' into profiler 15d6f18 [Davies Liu] add docs for two configs dadee1a [Davies Liu] add docs string and clear profiles after show or dump 4f8309d [Davies Liu] address comment, add tests 0a5b6eb [Davies Liu] fix Python UDF 4b20494 [Davies Liu] add profile for python

add profile for python
4b20494

fix Python UDF
0a5b6eb

JoshRosen reviewed Sep 13, 2014
View reviewed changes

address comment, add tests
4f8309d

davies added 2 commits September 12, 2014 21:51

add docs string and clear profiles after show or dump
dadee1a

add docs for two configs
15d6f18

Merge branch 'master' into profiler
c23865c

Merge branch 'master' into profiler
09d02c3
Conflicts: docs/configuration.md

Merge branch 'master' of github.com:apache/spark into profiler
116d52a
Conflicts: python/pyspark/worker.py

Merge branch 'master' of github.com:apache/spark into profiler
fb9565b
Conflicts: python/pyspark/worker.py

bugfix, add tests for show_profiles and dump_profiles()
7ef2aa0

JoshRosen reviewed Sep 25, 2014
View reviewed changes

asfgit closed this in 1aa549bSep 26, 2014

This was referenced Sep 26, 2014
[SPARK-3597][Mesos] Implement killTask. #2453
Closed
[SPARK-3701][MLLIB] update python linalg api and small fixes #2548
Closed

davies mentioned this pull request Sep 27, 2014
[SPARK-3478] [PySpark] Profile the Python tasks #2556
Closed

Uh oh!

Conversation

davies commented Sep 11, 2014

Uh oh!

SparkQA commented Sep 11, 2014

Uh oh!

SparkQA commented Sep 11, 2014

Uh oh!

SparkQA commented Sep 11, 2014

Uh oh!

SparkQA commented Sep 11, 2014

Uh oh!

SparkQA commented Sep 11, 2014

Uh oh!

SparkQA commented Sep 11, 2014

Uh oh!

SparkQA commented Sep 11, 2014

Uh oh!

SparkQA commented Sep 11, 2014

Uh oh!

JoshRosenSep 13, 2014

Choose a reason for hiding this comment

Uh oh!

SparkQA commented Sep 13, 2014

Uh oh!

SparkQA commented Sep 13, 2014

Uh oh!

davies commented Sep 13, 2014

Uh oh!

SparkQA commented Sep 13, 2014

Uh oh!

SparkQA commented Sep 13, 2014

Uh oh!

SparkQA commented Sep 13, 2014

Uh oh!

SparkQA commented Sep 13, 2014

Uh oh!

SparkQA commented Sep 14, 2014

Uh oh!

SparkQA commented Sep 14, 2014

Uh oh!

SparkQA commented Sep 17, 2014

Uh oh!

SparkQA commented Sep 17, 2014

Uh oh!

SparkQA commented Sep 23, 2014

Uh oh!

JoshRosen commented Sep 24, 2014

Uh oh!

SparkQA commented Sep 24, 2014

Uh oh!

SparkQA commented Sep 24, 2014

Uh oh!

SparkQA commented Sep 25, 2014

Uh oh!

SparkQA commented Sep 25, 2014

Uh oh!

SparkQA commented Sep 25, 2014

Uh oh!

shaneknapp commented Sep 25, 2014

Uh oh!

SparkQA commented Sep 25, 2014

Uh oh!

SparkQA commented Sep 25, 2014

Uh oh!

AmplabJenkins commented Sep 25, 2014

Uh oh!

JoshRosen commented Sep 25, 2014

Uh oh!

JoshRosen commented Sep 25, 2014

Uh oh!

davies commented Sep 25, 2014

Uh oh!

AmplabJenkins commented Sep 25, 2014

Uh oh!

SparkQA commented Sep 25, 2014

Uh oh!

SparkQA commented Sep 25, 2014

Uh oh!

JoshRosenSep 25, 2014