[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[ https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16099686#comment-16099686 ] Paul Magnus Sørensen-Clark commented on SPARK-12261: I have a similar problem, so I have tried the solution suggested here, adding "for _ in iterator: pass". Two people suggested doing this, but in different places. Niall McCarrol said to add it in the function process in pyspark/worker.py. Shea Parkes said to add it in the function takeUpToNumLeft, which I found in pyspark/rdd.py. What is the deal with these two different locations? I don't really know the difference, so I just added it both places to be safe. This error only randomly happens every once in a while for me. So it is hard to tell if it actually helped until several days have passed with no error. I use Python and Spark version 2.1.1. > pyspark crash for large dataset > --- > > Key: SPARK-12261 > URL: https://issues.apache.org/jira/browse/SPARK-12261 > Project: Spark > Issue Type: Bug > Components: PySpark >Affects Versions: 1.5.2 > Environment: windows >Reporter: zihao > > I tried to import a local text(over 100mb) file via textFile in pyspark, when > i ran data.take(), it failed and gave error messages including: > 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times; > aborting job > Traceback (most recent call last): > File "E:/spark_python/test3.py", line 9, in > lines.take(5) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299, > in take > res = self.context.runJob(self, takeUpToNumLeft, p) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line > 916, in runJob > port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd, > partitions) > File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in > __call__ > answer, self.gateway_client, self.target_id, self.name) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line > 36, in deco > return f(*a, **kw) > File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in > get_return_value > format(target_id, ".", name), value) > py4j.protocol.Py4JJavaError: An error occurred while calling > z:org.apache.spark.api.python.PythonRDD.runJob. > : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 > in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0 > (TID 0, localhost): java.net.SocketException: Connection reset by peer: > socket write error > Then i ran the same code for a small text file, this time .take() worked fine. > How can i solve this problem? -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[ https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15928179#comment-15928179 ] Shea Parkes commented on SPARK-12261: - I simply added the following to the end: for _ in iterator: pass This will run through the rest of iterator (until the StopIteration exception like normal). Depending how you're making pyspark importable, you might need to make this change inside a zipped copy of pyspark as well (e.g. in the binary distributions available). > pyspark crash for large dataset > --- > > Key: SPARK-12261 > URL: https://issues.apache.org/jira/browse/SPARK-12261 > Project: Spark > Issue Type: Bug > Components: PySpark >Affects Versions: 1.5.2 > Environment: windows >Reporter: zihao > > I tried to import a local text(over 100mb) file via textFile in pyspark, when > i ran data.take(), it failed and gave error messages including: > 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times; > aborting job > Traceback (most recent call last): > File "E:/spark_python/test3.py", line 9, in > lines.take(5) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299, > in take > res = self.context.runJob(self, takeUpToNumLeft, p) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line > 916, in runJob > port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd, > partitions) > File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in > __call__ > answer, self.gateway_client, self.target_id, self.name) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line > 36, in deco > return f(*a, **kw) > File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in > get_return_value > format(target_id, ".", name), value) > py4j.protocol.Py4JJavaError: An error occurred while calling > z:org.apache.spark.api.python.PythonRDD.runJob. > : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 > in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0 > (TID 0, localhost): java.net.SocketException: Connection reset by peer: > socket write error > Then i ran the same code for a small text file, this time .take() worked fine. > How can i solve this problem? -- This message was sent by Atlassian JIRA (v6.3.15#6346) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[ https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15927746#comment-15927746 ] Tomas Pranckevicius commented on SPARK-12261: - I am looking as well to the solution of this pyspark crash for large dataset issue on windows. I have read several posts and spent few days on this problem. I am happy to see that there is a solution mention by Shea Parkes and I am trying to get it working by changing rdd.py, but it still does not provide the positive outcome. Could please write more details on the change that has to be done in the proposed bandaid of exhausting the iterator at the end of takeUpToNumLeft() by changing rdd.py file? def takeUpToNumLeft(): iterator = iter(iterator) taken = 0 while taken < left: yield next(iterator) taken += 1 > pyspark crash for large dataset > --- > > Key: SPARK-12261 > URL: https://issues.apache.org/jira/browse/SPARK-12261 > Project: Spark > Issue Type: Bug > Components: PySpark >Affects Versions: 1.5.2 > Environment: windows >Reporter: zihao > > I tried to import a local text(over 100mb) file via textFile in pyspark, when > i ran data.take(), it failed and gave error messages including: > 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times; > aborting job > Traceback (most recent call last): > File "E:/spark_python/test3.py", line 9, in > lines.take(5) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299, > in take > res = self.context.runJob(self, takeUpToNumLeft, p) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line > 916, in runJob > port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd, > partitions) > File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in > __call__ > answer, self.gateway_client, self.target_id, self.name) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line > 36, in deco > return f(*a, **kw) > File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in > get_return_value > format(target_id, ".", name), value) > py4j.protocol.Py4JJavaError: An error occurred while calling > z:org.apache.spark.api.python.PythonRDD.runJob. > : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 > in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0 > (TID 0, localhost): java.net.SocketException: Connection reset by peer: > socket write error > Then i ran the same code for a small text file, this time .take() worked fine. > How can i solve this problem? -- This message was sent by Atlassian JIRA (v6.3.15#6346) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[
https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15629182#comment-15629182
]
Shea Parkes commented on SPARK-12261:
-
I'm still maintaining the two-line bandaid to {{takeUpToNumLeft()}} in v2.0.x
for our purposes (and it's still working). Given that it has helped another
person now, would you accept a patch with that bandaid into the mainline code?
> pyspark crash for large dataset
> ---
>
> Key: SPARK-12261
> URL: https://issues.apache.org/jira/browse/SPARK-12261
> Project: Spark
> Issue Type: Bug
> Components: PySpark
>Affects Versions: 1.5.2
> Environment: windows
>Reporter: zihao
>
> I tried to import a local text(over 100mb) file via textFile in pyspark, when
> i ran data.take(), it failed and gave error messages including:
> 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times;
> aborting job
> Traceback (most recent call last):
> File "E:/spark_python/test3.py", line 9, in
> lines.take(5)
> File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299,
> in take
> res = self.context.runJob(self, takeUpToNumLeft, p)
> File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line
> 916, in runJob
> port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd,
> partitions)
> File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in
> __call__
> answer, self.gateway_client, self.target_id, self.name)
> File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line
> 36, in deco
> return f(*a, **kw)
> File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in
> get_return_value
> format(target_id, ".", name), value)
> py4j.protocol.Py4JJavaError: An error occurred while calling
> z:org.apache.spark.api.python.PythonRDD.runJob.
> : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0
> in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0
> (TID 0, localhost): java.net.SocketException: Connection reset by peer:
> socket write error
> Then i ran the same code for a small text file, this time .take() worked fine.
> How can i solve this problem?
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
-
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[
https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15629165#comment-15629165
]
Oleh Koval commented on SPARK-12261:
Hey guys,
Having the same issue with Spark 1.6.1 on Win7 with 16GB RAM and 80Mb dataset
(Standalone Cluster mode).
collect() succeeds, but take(1) fails with same stack as above.
Patching {{rdd.py}} to exhaust the iterator at the end of {{takeUpToNumLeft()}}
made the error go away for me.
> pyspark crash for large dataset
> ---
>
> Key: SPARK-12261
> URL: https://issues.apache.org/jira/browse/SPARK-12261
> Project: Spark
> Issue Type: Bug
> Components: PySpark
>Affects Versions: 1.5.2
> Environment: windows
>Reporter: zihao
>
> I tried to import a local text(over 100mb) file via textFile in pyspark, when
> i ran data.take(), it failed and gave error messages including:
> 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times;
> aborting job
> Traceback (most recent call last):
> File "E:/spark_python/test3.py", line 9, in
> lines.take(5)
> File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299,
> in take
> res = self.context.runJob(self, takeUpToNumLeft, p)
> File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line
> 916, in runJob
> port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd,
> partitions)
> File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in
> __call__
> answer, self.gateway_client, self.target_id, self.name)
> File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line
> 36, in deco
> return f(*a, **kw)
> File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in
> get_return_value
> format(target_id, ".", name), value)
> py4j.protocol.Py4JJavaError: An error occurred while calling
> z:org.apache.spark.api.python.PythonRDD.runJob.
> : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0
> in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0
> (TID 0, localhost): java.net.SocketException: Connection reset by peer:
> socket write error
> Then i ran the same code for a small text file, this time .take() worked fine.
> How can i solve this problem?
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
-
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[ https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15420702#comment-15420702 ] Guangyang Nie commented on SPARK-12261: --- This way exactly solve my problem. Thank you so much! > pyspark crash for large dataset > --- > > Key: SPARK-12261 > URL: https://issues.apache.org/jira/browse/SPARK-12261 > Project: Spark > Issue Type: Bug > Components: PySpark >Affects Versions: 1.5.2 > Environment: windows >Reporter: zihao > > I tried to import a local text(over 100mb) file via textFile in pyspark, when > i ran data.take(), it failed and gave error messages including: > 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times; > aborting job > Traceback (most recent call last): > File "E:/spark_python/test3.py", line 9, in > lines.take(5) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299, > in take > res = self.context.runJob(self, takeUpToNumLeft, p) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line > 916, in runJob > port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd, > partitions) > File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in > __call__ > answer, self.gateway_client, self.target_id, self.name) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line > 36, in deco > return f(*a, **kw) > File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in > get_return_value > format(target_id, ".", name), value) > py4j.protocol.Py4JJavaError: An error occurred while calling > z:org.apache.spark.api.python.PythonRDD.runJob. > : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 > in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0 > (TID 0, localhost): java.net.SocketException: Connection reset by peer: > socket write error > Then i ran the same code for a small text file, this time .take() worked fine. > How can i solve this problem? -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[
https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15394508#comment-15394508
]
Shea Parkes commented on SPARK-12261:
-
I still can't get this bug to reproduce reliably locally, but I can confirm
that my proposed bandaid of exhausting the iterator at the end of
{{takeUpToNumLeft()}} made the error go away entirely in manual testing. Would
the Spark maintainers be willing to accept such a band-aid into the main
codebase, or is this something I'd just need to maintain in our own Spark
distributions? (I already maintain a couple other modifications related to
Windows mess.)
A more root cause fix would require more Scala knowledge than I have. I'd
likely propose to define a new constant (e.g.
{{SpecialLengths.PYTHON_WORKER_EXITING}}) and put logic into Scala that would
make it immediately stop sending new data over the socket...
> pyspark crash for large dataset
> ---
>
> Key: SPARK-12261
> URL: https://issues.apache.org/jira/browse/SPARK-12261
> Project: Spark
> Issue Type: Bug
> Components: PySpark
>Affects Versions: 1.5.2
> Environment: windows
>Reporter: zihao
>
> I tried to import a local text(over 100mb) file via textFile in pyspark, when
> i ran data.take(), it failed and gave error messages including:
> 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times;
> aborting job
> Traceback (most recent call last):
> File "E:/spark_python/test3.py", line 9, in
> lines.take(5)
> File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299,
> in take
> res = self.context.runJob(self, takeUpToNumLeft, p)
> File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line
> 916, in runJob
> port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd,
> partitions)
> File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in
> __call__
> answer, self.gateway_client, self.target_id, self.name)
> File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line
> 36, in deco
> return f(*a, **kw)
> File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in
> get_return_value
> format(target_id, ".", name), value)
> py4j.protocol.Py4JJavaError: An error occurred while calling
> z:org.apache.spark.api.python.PythonRDD.runJob.
> : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0
> in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0
> (TID 0, localhost): java.net.SocketException: Connection reset by peer:
> socket write error
> Then i ran the same code for a small text file, this time .take() worked fine.
> How can i solve this problem?
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
-
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[
https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15391731#comment-15391731
]
Shea Parkes commented on SPARK-12261:
-
Also, I added extensive logging to {{worker.py}} in my environment to figure
out much of this. In my experience, {{worker.py}} never crashes, it just
{{sys.exit(-1)}} itself.
> pyspark crash for large dataset
> ---
>
> Key: SPARK-12261
> URL: https://issues.apache.org/jira/browse/SPARK-12261
> Project: Spark
> Issue Type: Bug
> Components: PySpark
>Affects Versions: 1.5.2
> Environment: windows
>Reporter: zihao
>
> I tried to import a local text(over 100mb) file via textFile in pyspark, when
> i ran data.take(), it failed and gave error messages including:
> 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times;
> aborting job
> Traceback (most recent call last):
> File "E:/spark_python/test3.py", line 9, in
> lines.take(5)
> File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299,
> in take
> res = self.context.runJob(self, takeUpToNumLeft, p)
> File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line
> 916, in runJob
> port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd,
> partitions)
> File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in
> __call__
> answer, self.gateway_client, self.target_id, self.name)
> File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line
> 36, in deco
> return f(*a, **kw)
> File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in
> get_return_value
> format(target_id, ".", name), value)
> py4j.protocol.Py4JJavaError: An error occurred while calling
> z:org.apache.spark.api.python.PythonRDD.runJob.
> : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0
> in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0
> (TID 0, localhost): java.net.SocketException: Connection reset by peer:
> socket write error
> Then i ran the same code for a small text file, this time .take() worked fine.
> How can i solve this problem?
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
-
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[
https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15391728#comment-15391728
]
Shea Parkes commented on SPARK-12261:
-
Alright, I've been spending time off and on for a week on this. I think I
better understand what's going on, but don't yet really have it nailed down.
I'm going to try and write down what I understand is going on here to get my
thoughts in order. I've been focusing on {{branch-1.6}} since that's what we
have in production.
When calling {{RDD.take()}} in {{rdd.py}}, it does not push the request down
to a ~scala implementation. Instead, it defines a closure/generator
({{takeUpToNumLeft()}}) and pushes that into a {{RDD.mapPartitions()}}.
{{takeUpToNumLeft()}} does **not** exhaust the iterator that it is given; it
yields only as many items as requested and then exits.
Next is the interplay between {{worker.py}} and {{PythonRDD.scala}}. These two
files setup bi-directional streams to each other and communicate some gnarly
state back and forth. The important part is what happens in {{worker.py}} when
the provided generator (i.e. {{takeUpToNumLeft()}} does not exhaust the stream
of data being provided by {{PythonRDD.scala}}. When this happens,
{{worker.py}} sends a {{SpecialLengths.END_OF_DATA_SECTION}}, followed by any
accumulators, and then sends a **second**
{{SpecialLengths.END_OF_DATA_SECTION}} and then kills itself.
I'm much better at Python than Scala, so I'm now trying to understand what
happens when {{PythonRDD.scala}} receives that second
{{SpecialLengths.END_OF_DATA_SECTION}}. In particular, I don't see anywhere
that {{PythonRDD.scala}} would treat the second
{{SpecialLengths.END_OF_DATA_SECTION}} any different. This means that
{{PythonRDD.scala}} would go on to try and read the accumulator information
from the stream again, but {{worker.py}} would have already exited, so it
doesn't find anything.
{{PythonRDD.scala}} actually fails when trying to **send** data as I understand
it though, which is where my understanding of Scala is a little rough. If I'm
assuming the code in {{PythonRDD.scala}} works similar to a Python generator,
then I'm assuming it's acting something like a co-routine and isn't stopping
when it gets the second {{SpecialLengths.END_OF_DATA_SECTION}}.
This still doesn't explain why this works 95% of the time for us (i.e. we only
get intermittent failures).
Also another track to chase is that Windows is treated differently in
{{PythonWorkerFactory.scala}}, so I'm occasionally trying to figure out if
{{worker.py}} being ran as a subprocess makes any difference...
So I'm going to try and spend more time with understanding what triggers more
data emission from {{PythonRDD.scala}} and why it keeps emitting after a second
{{SpecialLengths.END_OF_DATA_SECTION}}.
A simple band-aid would be to alter {{takeUpToNumLeft()}} in {{rdd.py}} to
exhaust the iterator provided, but I'm not sure that's a root cause fix. Was
it intentional to allow {{RDD.mapPartitions()}} to accept generators that did
not exhaust their streams?
> pyspark crash for large dataset
> ---
>
> Key: SPARK-12261
> URL: https://issues.apache.org/jira/browse/SPARK-12261
> Project: Spark
> Issue Type: Bug
> Components: PySpark
>Affects Versions: 1.5.2
> Environment: windows
>Reporter: zihao
>
> I tried to import a local text(over 100mb) file via textFile in pyspark, when
> i ran data.take(), it failed and gave error messages including:
> 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times;
> aborting job
> Traceback (most recent call last):
> File "E:/spark_python/test3.py", line 9, in
> lines.take(5)
> File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299,
> in take
> res = self.context.runJob(self, takeUpToNumLeft, p)
> File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line
> 916, in runJob
> port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd,
> partitions)
> File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in
> __call__
> answer, self.gateway_client, self.target_id, self.name)
> File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line
> 36, in deco
> return f(*a, **kw)
> File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in
> get_return_value
> format(target_id, ".", name), value)
> py4j.protocol.Py4JJavaError: An error occurred while calling
> z:org.apache.spark.api.python.PythonRDD.runJob.
> : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0
> in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0
> (TID 0, localhost): java.net.SocketException: Connection reset by peer:
> socket write error
> Then i
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[ https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15381167#comment-15381167 ] Sean Owen commented on SPARK-12261: --- We need logs showing the actual error. If this is local mode, the executor output is in the same log, but, this snippet doesn't show anything but 'job failed'. If there's really nothing here then it's something to do with the python process exiting or crashing. Not as sure how to get that output. > pyspark crash for large dataset > --- > > Key: SPARK-12261 > URL: https://issues.apache.org/jira/browse/SPARK-12261 > Project: Spark > Issue Type: Bug > Components: PySpark >Affects Versions: 1.5.2 > Environment: windows >Reporter: zihao > > I tried to import a local text(over 100mb) file via textFile in pyspark, when > i ran data.take(), it failed and gave error messages including: > 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times; > aborting job > Traceback (most recent call last): > File "E:/spark_python/test3.py", line 9, in > lines.take(5) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299, > in take > res = self.context.runJob(self, takeUpToNumLeft, p) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line > 916, in runJob > port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd, > partitions) > File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in > __call__ > answer, self.gateway_client, self.target_id, self.name) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line > 36, in deco > return f(*a, **kw) > File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in > get_return_value > format(target_id, ".", name), value) > py4j.protocol.Py4JJavaError: An error occurred while calling > z:org.apache.spark.api.python.PythonRDD.runJob. > : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 > in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0 > (TID 0, localhost): java.net.SocketException: Connection reset by peer: > socket write error > Then i ran the same code for a small text file, this time .take() worked fine. > How can i solve this problem? -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[ https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15380846#comment-15380846 ] Shea Parkes commented on SPARK-12261: - I believe I'm hitting the same bug. I'm also running on Windows in Standalone Cluster mode. Unfortunately, the error is non-deterministic (i.e. I've had no luck in creating an always reproducible scenario). I promise I have plenty of RAM, and I'm not working with *that* big of data. I too can try and capture better logging. As Chris hinted to though, which logs are you looking for? I can grab logs from the master, workers, and application, but it doesn't appear obvious where to grab logs from the python workers that are spawned. I'm happy to read documentation (and code) to try and chase down appropriate logs, but I haven't found any helpful directions yet. > pyspark crash for large dataset > --- > > Key: SPARK-12261 > URL: https://issues.apache.org/jira/browse/SPARK-12261 > Project: Spark > Issue Type: Bug > Components: PySpark >Affects Versions: 1.5.2 > Environment: windows >Reporter: zihao > > I tried to import a local text(over 100mb) file via textFile in pyspark, when > i ran data.take(), it failed and gave error messages including: > 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times; > aborting job > Traceback (most recent call last): > File "E:/spark_python/test3.py", line 9, in > lines.take(5) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299, > in take > res = self.context.runJob(self, takeUpToNumLeft, p) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line > 916, in runJob > port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd, > partitions) > File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in > __call__ > answer, self.gateway_client, self.target_id, self.name) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line > 36, in deco > return f(*a, **kw) > File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in > get_return_value > format(target_id, ".", name), value) > py4j.protocol.Py4JJavaError: An error occurred while calling > z:org.apache.spark.api.python.PythonRDD.runJob. > : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 > in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0 > (TID 0, localhost): java.net.SocketException: Connection reset by peer: > socket write error > Then i ran the same code for a small text file, this time .take() worked fine. > How can i solve this problem? -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[ https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15144844#comment-15144844 ] Christopher Bourez commented on SPARK-12261: Sean Owen, do you reconsider the status as a Spark issue ? > pyspark crash for large dataset > --- > > Key: SPARK-12261 > URL: https://issues.apache.org/jira/browse/SPARK-12261 > Project: Spark > Issue Type: Bug > Components: PySpark >Affects Versions: 1.5.2 > Environment: windows >Reporter: zihao > > I tried to import a local text(over 100mb) file via textFile in pyspark, when > i ran data.take(), it failed and gave error messages including: > 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times; > aborting job > Traceback (most recent call last): > File "E:/spark_python/test3.py", line 9, in > lines.take(5) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299, > in take > res = self.context.runJob(self, takeUpToNumLeft, p) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line > 916, in runJob > port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd, > partitions) > File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in > __call__ > answer, self.gateway_client, self.target_id, self.name) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line > 36, in deco > return f(*a, **kw) > File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in > get_return_value > format(target_id, ".", name), value) > py4j.protocol.Py4JJavaError: An error occurred while calling > z:org.apache.spark.api.python.PythonRDD.runJob. > : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 > in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0 > (TID 0, localhost): java.net.SocketException: Connection reset by peer: > socket write error > Then i ran the same code for a small text file, this time .take() worked fine. > How can i solve this problem? -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[ https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15144873#comment-15144873 ] Christopher Bourez commented on SPARK-12261: Here is what i see when i activate the logs : 16/02/12 18:09:22 ERROR TaskSetManager: Task 0 in stage 5.0 failed 1 times; abor ting job 16/02/12 18:09:22 INFO TaskSchedulerImpl: Removed TaskSet 5.0, whose tasks have all completed, from pool 16/02/12 18:09:22 INFO TaskSchedulerImpl: Cancelling stage 5 16/02/12 18:09:22 INFO DAGScheduler: ResultStage 5 (runJob at PythonRDD.scala:39 3) failed in 0,280 s 16/02/12 18:09:22 INFO DAGScheduler: Job 5 failed: runJob at PythonRDD.scala:393 , took 0,308529 s Traceback (most recent call last): File "", line 1, in File "C:\Documents\c.bourez\Documents\spark-1.5.2-bin-hadoop2.6\spark-1.5.2-bi n-hadoop2.6\python\pyspark\rdd.py", line 1299, in take res = self.context.runJob(self, takeUpToNumLeft, p) File "C:\Documents\c.bourez\Documents\spark-1.5.2-bin-hadoop2.6\spark-1.5.2-bi n-hadoop2.6\python\pyspark\context.py", line 916, in runJob port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd, partition s) File "C:\Documents\c.bourez\Documents\spark-1.5.2-bin-hadoop2.6\spark-1.5.2-bi n-hadoop2.6\python\lib\py4j-0.8.2.1-src.zip\py4j\java_gateway.py", line 538, in __call__ File "C:\Documents\c.bourez\Documents\spark-1.5.2-bin-hadoop2.6\spark-1.5.2-bi n-hadoop2.6\python\pyspark\sql\utils.py", line 36, in deco return f(*a, **kw) File "C:\Documents\c.bourez\Documents\spark-1.5.2-bin-hadoop2.6\spark-1.5.2-bi n-hadoop2.6\python\lib\py4j-0.8.2.1-src.zip\py4j\protocol.py", line 300, in get_ return_value py4j.protocol.Py4JJavaError: An error occurred while calling z:org.apache.spark. api.python.PythonRDD.runJob. : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 in s tage 5.0 failed 1 times, most recent failure: Lost task 0.0 in stage 5.0 (TID 5, localhost): java.net.SocketException: Connection reset by peer: socket write er ror at java.net.SocketOutputStream.socketWrite0(Native Method) at java.net.SocketOutputStream.socketWrite(SocketOutputStream.java:113) at java.net.SocketOutputStream.write(SocketOutputStream.java:159) at java.io.BufferedOutputStream.flushBuffer(BufferedOutputStream.java:82 ) at java.io.BufferedOutputStream.write(BufferedOutputStream.java:126) at java.io.DataOutputStream.write(DataOutputStream.java:107) at java.io.FilterOutputStream.write(FilterOutputStream.java:97) at org.apache.spark.api.python.PythonRDD$.writeUTF(PythonRDD.scala:622) at org.apache.spark.api.python.PythonRDD$.org$apache$spark$api$python$Py thonRDD$$write$1(PythonRDD.scala:442) at org.apache.spark.api.python.PythonRDD$$anonfun$writeIteratorToStream$ 1.apply(PythonRDD.scala:452) at org.apache.spark.api.python.PythonRDD$$anonfun$writeIteratorToStream$ 1.apply(PythonRDD.scala:452) at scala.collection.Iterator$class.foreach(Iterator.scala:727) at scala.collection.AbstractIterator.foreach(Iterator.scala:1157) at org.apache.spark.api.python.PythonRDD$.writeIteratorToStream(PythonRD D.scala:452) at org.apache.spark.api.python.PythonRunner$WriterThread$$anonfun$run$3. apply(PythonRDD.scala:280) at org.apache.spark.util.Utils$.logUncaughtExceptions(Utils.scala:1699) > pyspark crash for large dataset > --- > > Key: SPARK-12261 > URL: https://issues.apache.org/jira/browse/SPARK-12261 > Project: Spark > Issue Type: Bug > Components: PySpark >Affects Versions: 1.5.2 > Environment: windows >Reporter: zihao > > I tried to import a local text(over 100mb) file via textFile in pyspark, when > i ran data.take(), it failed and gave error messages including: > 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times; > aborting job > Traceback (most recent call last): > File "E:/spark_python/test3.py", line 9, in > lines.take(5) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299, > in take > res = self.context.runJob(self, takeUpToNumLeft, p) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line > 916, in runJob > port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd, > partitions) > File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in > __call__ > answer, self.gateway_client, self.target_id, self.name) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line > 36, in deco > return f(*a, **kw) > File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in > get_return_value > format(target_id, ".", name), value) > py4j.protocol.Py4JJavaError: An error occurred while calling >
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[ https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15144880#comment-15144880 ] Sean Owen commented on SPARK-12261: --- This is still just the driver log. > pyspark crash for large dataset > --- > > Key: SPARK-12261 > URL: https://issues.apache.org/jira/browse/SPARK-12261 > Project: Spark > Issue Type: Bug > Components: PySpark >Affects Versions: 1.5.2 > Environment: windows >Reporter: zihao > > I tried to import a local text(over 100mb) file via textFile in pyspark, when > i ran data.take(), it failed and gave error messages including: > 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times; > aborting job > Traceback (most recent call last): > File "E:/spark_python/test3.py", line 9, in > lines.take(5) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299, > in take > res = self.context.runJob(self, takeUpToNumLeft, p) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line > 916, in runJob > port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd, > partitions) > File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in > __call__ > answer, self.gateway_client, self.target_id, self.name) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line > 36, in deco > return f(*a, **kw) > File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in > get_return_value > format(target_id, ".", name), value) > py4j.protocol.Py4JJavaError: An error occurred while calling > z:org.apache.spark.api.python.PythonRDD.runJob. > : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 > in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0 > (TID 0, localhost): java.net.SocketException: Connection reset by peer: > socket write error > Then i ran the same code for a small text file, this time .take() worked fine. > How can i solve this problem? -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[ https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15144901#comment-15144901 ] Christopher Bourez commented on SPARK-12261: Sean, how can I get the executor log in local mode ? Thanks > pyspark crash for large dataset > --- > > Key: SPARK-12261 > URL: https://issues.apache.org/jira/browse/SPARK-12261 > Project: Spark > Issue Type: Bug > Components: PySpark >Affects Versions: 1.5.2 > Environment: windows >Reporter: zihao > > I tried to import a local text(over 100mb) file via textFile in pyspark, when > i ran data.take(), it failed and gave error messages including: > 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times; > aborting job > Traceback (most recent call last): > File "E:/spark_python/test3.py", line 9, in > lines.take(5) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299, > in take > res = self.context.runJob(self, takeUpToNumLeft, p) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line > 916, in runJob > port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd, > partitions) > File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in > __call__ > answer, self.gateway_client, self.target_id, self.name) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line > 36, in deco > return f(*a, **kw) > File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in > get_return_value > format(target_id, ".", name), value) > py4j.protocol.Py4JJavaError: An error occurred while calling > z:org.apache.spark.api.python.PythonRDD.runJob. > : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 > in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0 > (TID 0, localhost): java.net.SocketException: Connection reset by peer: > socket write error > Then i ran the same code for a small text file, this time .take() worked fine. > How can i solve this problem? -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[ https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15144857#comment-15144857 ] Sean Owen commented on SPARK-12261: --- The change above is definitely not correct in general, right? it's just ripping out some required functionality that happens to not affect your simple test. That does not sound like a solution at all. No, I'm still not clear on what you suggest the issue is. It sounds like you're suggesting it's a memory problem, but then, increasing memory doesn't help, and the error is not "out of memory". However, it sounds like you're attempting to run an entire cluster in 2GB of RAM. It wouldn't surprise me if indeed something runs out of memory, but, is that what you see in the logs? this info isn't given. Then you say it works on a Mac with 16GB of RAM. So, again, isn't the issue just that you have way too little resource on the Windows machine? I personally think this needs to be closed. Josh reopened, but he's not going to work on it, and I'm not clear there is really a reproduction here. (This does not fail for me.) > pyspark crash for large dataset > --- > > Key: SPARK-12261 > URL: https://issues.apache.org/jira/browse/SPARK-12261 > Project: Spark > Issue Type: Bug > Components: PySpark >Affects Versions: 1.5.2 > Environment: windows >Reporter: zihao > > I tried to import a local text(over 100mb) file via textFile in pyspark, when > i ran data.take(), it failed and gave error messages including: > 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times; > aborting job > Traceback (most recent call last): > File "E:/spark_python/test3.py", line 9, in > lines.take(5) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299, > in take > res = self.context.runJob(self, takeUpToNumLeft, p) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line > 916, in runJob > port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd, > partitions) > File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in > __call__ > answer, self.gateway_client, self.target_id, self.name) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line > 36, in deco > return f(*a, **kw) > File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in > get_return_value > format(target_id, ".", name), value) > py4j.protocol.Py4JJavaError: An error occurred while calling > z:org.apache.spark.api.python.PythonRDD.runJob. > : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 > in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0 > (TID 0, localhost): java.net.SocketException: Connection reset by peer: > socket write error > Then i ran the same code for a small text file, this time .take() worked fine. > How can i solve this problem? -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[ https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15144793#comment-15144793 ] Christopher Bourez commented on SPARK-12261: Dear Niall Your solution works very well :) Thank you a lot > pyspark crash for large dataset > --- > > Key: SPARK-12261 > URL: https://issues.apache.org/jira/browse/SPARK-12261 > Project: Spark > Issue Type: Bug > Components: PySpark >Affects Versions: 1.5.2 > Environment: windows >Reporter: zihao > > I tried to import a local text(over 100mb) file via textFile in pyspark, when > i ran data.take(), it failed and gave error messages including: > 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times; > aborting job > Traceback (most recent call last): > File "E:/spark_python/test3.py", line 9, in > lines.take(5) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299, > in take > res = self.context.runJob(self, takeUpToNumLeft, p) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line > 916, in runJob > port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd, > partitions) > File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in > __call__ > answer, self.gateway_client, self.target_id, self.name) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line > 36, in deco > return f(*a, **kw) > File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in > get_return_value > format(target_id, ".", name), value) > py4j.protocol.Py4JJavaError: An error occurred while calling > z:org.apache.spark.api.python.PythonRDD.runJob. > : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 > in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0 > (TID 0, localhost): java.net.SocketException: Connection reset by peer: > socket write error > Then i ran the same code for a small text file, this time .take() worked fine. > How can i solve this problem? -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[ https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15144734#comment-15144734 ] Niall McCarroll commented on SPARK-12261: - As a workaround you might try the following change to python/pyspark/worker.py Add the following 2 lines to the end of the process function defined inside the main function for obj in iterator: pass ... so the process function now looks like this (in spark 1.5.2 at least): def process(): iterator = deserializer.load_stream(infile) serializer.dump_stream(func(split_index, iterator), outfile) for obj in iterator: pass After the change you will need to rebuild your pyspark.zip in the python/lib folder to include the change. The issue may be that the worker process is completing before the executor has written all the data to it. The thread writing the data down the socket throws an exception and if this happens before the executor marks the task as complete it will cause trouble. The idea is to try to get the worker to pull all the data from the executor even if its not needed to lazily compute the output. This is very inefficient of course so it is a workaround rather than a proper solution. > pyspark crash for large dataset > --- > > Key: SPARK-12261 > URL: https://issues.apache.org/jira/browse/SPARK-12261 > Project: Spark > Issue Type: Bug > Components: PySpark >Affects Versions: 1.5.2 > Environment: windows >Reporter: zihao > > I tried to import a local text(over 100mb) file via textFile in pyspark, when > i ran data.take(), it failed and gave error messages including: > 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times; > aborting job > Traceback (most recent call last): > File "E:/spark_python/test3.py", line 9, in > lines.take(5) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299, > in take > res = self.context.runJob(self, takeUpToNumLeft, p) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line > 916, in runJob > port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd, > partitions) > File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in > __call__ > answer, self.gateway_client, self.target_id, self.name) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line > 36, in deco > return f(*a, **kw) > File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in > get_return_value > format(target_id, ".", name), value) > py4j.protocol.Py4JJavaError: An error occurred while calling > z:org.apache.spark.api.python.PythonRDD.runJob. > : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 > in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0 > (TID 0, localhost): java.net.SocketException: Connection reset by peer: > socket write error > Then i ran the same code for a small text file, this time .take() worked fine. > How can i solve this problem? -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[ https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15145077#comment-15145077 ] Niall McCarroll commented on SPARK-12261: - In various windows environments I've tried, I see failures intermittently in some but not all. Using spark standalone 1.5.2 and 1.6.0. I don't think its related to memory resources. I added some logging to the python and scala sides to trace the protocol. The change above is a simple (but inefficient) workaround based on what I think the underlying issue might be. I hope it should not impact functionality. I am more than happy to work on this and and try and come up with a proper solution if it does turn out to be a bug. > pyspark crash for large dataset > --- > > Key: SPARK-12261 > URL: https://issues.apache.org/jira/browse/SPARK-12261 > Project: Spark > Issue Type: Bug > Components: PySpark >Affects Versions: 1.5.2 > Environment: windows >Reporter: zihao > > I tried to import a local text(over 100mb) file via textFile in pyspark, when > i ran data.take(), it failed and gave error messages including: > 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times; > aborting job > Traceback (most recent call last): > File "E:/spark_python/test3.py", line 9, in > lines.take(5) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299, > in take > res = self.context.runJob(self, takeUpToNumLeft, p) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line > 916, in runJob > port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd, > partitions) > File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in > __call__ > answer, self.gateway_client, self.target_id, self.name) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line > 36, in deco > return f(*a, **kw) > File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in > get_return_value > format(target_id, ".", name), value) > py4j.protocol.Py4JJavaError: An error occurred while calling > z:org.apache.spark.api.python.PythonRDD.runJob. > : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 > in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0 > (TID 0, localhost): java.net.SocketException: Connection reset by peer: > socket write error > Then i ran the same code for a small text file, this time .take() worked fine. > How can i solve this problem? -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[ https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15136229#comment-15136229 ] Christopher Bourez commented on SPARK-12261: I'm still here if you need any more info about how to reproduce the case > pyspark crash for large dataset > --- > > Key: SPARK-12261 > URL: https://issues.apache.org/jira/browse/SPARK-12261 > Project: Spark > Issue Type: Bug > Components: PySpark >Affects Versions: 1.5.2 > Environment: windows >Reporter: zihao > > I tried to import a local text(over 100mb) file via textFile in pyspark, when > i ran data.take(), it failed and gave error messages including: > 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times; > aborting job > Traceback (most recent call last): > File "E:/spark_python/test3.py", line 9, in > lines.take(5) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299, > in take > res = self.context.runJob(self, takeUpToNumLeft, p) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line > 916, in runJob > port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd, > partitions) > File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in > __call__ > answer, self.gateway_client, self.target_id, self.name) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line > 36, in deco > return f(*a, **kw) > File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in > get_return_value > format(target_id, ".", name), value) > py4j.protocol.Py4JJavaError: An error occurred while calling > z:org.apache.spark.api.python.PythonRDD.runJob. > : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 > in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0 > (TID 0, localhost): java.net.SocketException: Connection reset by peer: > socket write error > Then i ran the same code for a small text file, this time .take() worked fine. > How can i solve this problem? -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[ https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15136515#comment-15136515 ] Josh Rosen commented on SPARK-12261: [~srowen], I reopened this after it was updated with reproduction instructions because it seemed like it might be legitimate and would be worth investigating. However, I don't have any spare cycles to investigate this myself. > pyspark crash for large dataset > --- > > Key: SPARK-12261 > URL: https://issues.apache.org/jira/browse/SPARK-12261 > Project: Spark > Issue Type: Bug > Components: PySpark >Affects Versions: 1.5.2 > Environment: windows >Reporter: zihao > > I tried to import a local text(over 100mb) file via textFile in pyspark, when > i ran data.take(), it failed and gave error messages including: > 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times; > aborting job > Traceback (most recent call last): > File "E:/spark_python/test3.py", line 9, in > lines.take(5) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299, > in take > res = self.context.runJob(self, takeUpToNumLeft, p) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line > 916, in runJob > port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd, > partitions) > File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in > __call__ > answer, self.gateway_client, self.target_id, self.name) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line > 36, in deco > return f(*a, **kw) > File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in > get_return_value > format(target_id, ".", name), value) > py4j.protocol.Py4JJavaError: An error occurred while calling > z:org.apache.spark.api.python.PythonRDD.runJob. > : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 > in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0 > (TID 0, localhost): java.net.SocketException: Connection reset by peer: > socket write error > Then i ran the same code for a small text file, this time .take() worked fine. > How can i solve this problem? -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[
https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15121114#comment-15121114
]
Christopher Bourez commented on SPARK-12261:
I recompiled Spark on Windows but the problem remains.
The first 3 times I launch the textFile command followed by a take(1), it
works, but then does not work anymore. The memory (between python and the JVM)
sounds not to be release.
I tried to re-init
sc.stop(), del sc, sc = SparkContext('local','test')
import gc, gc.collect()
...
does not change. Memory not released.
It sounds that OOM are quite common on Windows/Pyspark
> pyspark crash for large dataset
> ---
>
> Key: SPARK-12261
> URL: https://issues.apache.org/jira/browse/SPARK-12261
> Project: Spark
> Issue Type: Bug
>Affects Versions: 1.5.2
> Environment: windows
>Reporter: zihao
>
> I tried to import a local text(over 100mb) file via textFile in pyspark, when
> i ran data.take(), it failed and gave error messages including:
> 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times;
> aborting job
> Traceback (most recent call last):
> File "E:/spark_python/test3.py", line 9, in
> lines.take(5)
> File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299,
> in take
> res = self.context.runJob(self, takeUpToNumLeft, p)
> File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line
> 916, in runJob
> port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd,
> partitions)
> File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in
> __call__
> answer, self.gateway_client, self.target_id, self.name)
> File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line
> 36, in deco
> return f(*a, **kw)
> File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in
> get_return_value
> format(target_id, ".", name), value)
> py4j.protocol.Py4JJavaError: An error occurred while calling
> z:org.apache.spark.api.python.PythonRDD.runJob.
> : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0
> in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0
> (TID 0, localhost): java.net.SocketException: Connection reset by peer:
> socket write error
> Then i ran the same code for a small text file, this time .take() worked fine.
> How can i solve this problem?
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
-
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[ https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15121182#comment-15121182 ] Christopher Bourez commented on SPARK-12261: There is a strange "remove broadcast variable" operation at the end of the 3 third sc.textFile().take(1) method execution; and then the next executions fail. Can there be a link with this problem : https://spark-project.atlassian.net/browse/SPARK-1065 ? > pyspark crash for large dataset > --- > > Key: SPARK-12261 > URL: https://issues.apache.org/jira/browse/SPARK-12261 > Project: Spark > Issue Type: Bug >Affects Versions: 1.5.2 > Environment: windows >Reporter: zihao > > I tried to import a local text(over 100mb) file via textFile in pyspark, when > i ran data.take(), it failed and gave error messages including: > 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times; > aborting job > Traceback (most recent call last): > File "E:/spark_python/test3.py", line 9, in > lines.take(5) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299, > in take > res = self.context.runJob(self, takeUpToNumLeft, p) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line > 916, in runJob > port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd, > partitions) > File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in > __call__ > answer, self.gateway_client, self.target_id, self.name) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line > 36, in deco > return f(*a, **kw) > File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in > get_return_value > format(target_id, ".", name), value) > py4j.protocol.Py4JJavaError: An error occurred while calling > z:org.apache.spark.api.python.PythonRDD.runJob. > : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 > in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0 > (TID 0, localhost): java.net.SocketException: Connection reset by peer: > socket write error > Then i ran the same code for a small text file, this time .take() worked fine. > How can i solve this problem? -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[
https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15116907#comment-15116907
]
Christopher Bourez commented on SPARK-12261:
To reproduce you can follow the steps :
- create an Aws Workspace with Windows 7 (that I can share you if you'd like)
with Standard instance, 2GiB RAM
On this instance :
- download spark (1.5 or 1.6 same pb) with hadoop 2.6
- install java 8 jdk
- download python 2.7.8
- downloaded the sample file
https://s3-eu-west-1.amazonaws.com/christopherbourez/public/test.csv
- launch Pyspark : bin\pyspark --master local[1]
- run command : sc.textFile("test.csv").take(1) => fails (very few times worked)
- run sc.textFile("test.csv", 2000).take(1) => works
Sample file is 13M, has been created randomly
for i in {0..30}; do
VALUE="$RANDOM"
for j in {0..6}; do
VALUE="$VALUE;$RANDOM";
done
echo $VALUE >> test.csv
done
Running Pyspark with more memory
bin\pyspark --master local[1] --conf spark.driver.memory=3g
displays more memory in http://localhost:4040/executors
but does not change the problem
Full video of the problem :
https://s3-eu-west-1.amazonaws.com/christopherbourez/public/video.mov
> pyspark crash for large dataset
> ---
>
> Key: SPARK-12261
> URL: https://issues.apache.org/jira/browse/SPARK-12261
> Project: Spark
> Issue Type: Bug
>Affects Versions: 1.5.2
> Environment: windows
>Reporter: zihao
>
> I tried to import a local text(over 100mb) file via textFile in pyspark, when
> i ran data.take(), it failed and gave error messages including:
> 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times;
> aborting job
> Traceback (most recent call last):
> File "E:/spark_python/test3.py", line 9, in
> lines.take(5)
> File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299,
> in take
> res = self.context.runJob(self, takeUpToNumLeft, p)
> File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line
> 916, in runJob
> port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd,
> partitions)
> File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in
> __call__
> answer, self.gateway_client, self.target_id, self.name)
> File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line
> 36, in deco
> return f(*a, **kw)
> File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in
> get_return_value
> format(target_id, ".", name), value)
> py4j.protocol.Py4JJavaError: An error occurred while calling
> z:org.apache.spark.api.python.PythonRDD.runJob.
> : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0
> in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0
> (TID 0, localhost): java.net.SocketException: Connection reset by peer:
> socket write error
> Then i ran the same code for a small text file, this time .take() worked fine.
> How can i solve this problem?
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
-
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[ https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15117345#comment-15117345 ] Christopher Bourez commented on SPARK-12261: The solution "Increase driver memory" does not work. > pyspark crash for large dataset > --- > > Key: SPARK-12261 > URL: https://issues.apache.org/jira/browse/SPARK-12261 > Project: Spark > Issue Type: Bug >Affects Versions: 1.5.2 > Environment: windows >Reporter: zihao > > I tried to import a local text(over 100mb) file via textFile in pyspark, when > i ran data.take(), it failed and gave error messages including: > 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times; > aborting job > Traceback (most recent call last): > File "E:/spark_python/test3.py", line 9, in > lines.take(5) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299, > in take > res = self.context.runJob(self, takeUpToNumLeft, p) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line > 916, in runJob > port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd, > partitions) > File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in > __call__ > answer, self.gateway_client, self.target_id, self.name) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line > 36, in deco > return f(*a, **kw) > File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in > get_return_value > format(target_id, ".", name), value) > py4j.protocol.Py4JJavaError: An error occurred while calling > z:org.apache.spark.api.python.PythonRDD.runJob. > : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 > in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0 > (TID 0, localhost): java.net.SocketException: Connection reset by peer: > socket write error > Then i ran the same code for a small text file, this time .take() worked fine. > How can i solve this problem? -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-12261) pyspark crash for large dataset
[ https://issues.apache.org/jira/browse/SPARK-12261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15115433#comment-15115433 ] Christopher Bourez commented on SPARK-12261: I think the issue is not resolved I installed spark 1.6 on many different computers. On Windows, PySpark textfile method, followed by take(1), does not work on a file of 13M. If I set numpartitions to 2000 or take a smaller file, the method works well. The Pyspark is set with all RAM memory of the computer thanks to the command --conf spark.driver.memory=5g in local mode. On Mac OS, I'm able to launch the exact same program with Pyspark with 16G RAM for a file of much bigger in comparison, of 5G. Memory is correctly allocated, removed etc On Ubuntu, no trouble, I can also launch a cluster http://christopher5106.github.io/big/data/2016/01/19/computation-power-as-you-need-with-EMR-auto-termination-cluster-example-random-forest-python.html The error message on Windows is : java.net.SocketException: Connection reset by peer: socket write error What could be the reason to have the windows spark textfile method fail ? > pyspark crash for large dataset > --- > > Key: SPARK-12261 > URL: https://issues.apache.org/jira/browse/SPARK-12261 > Project: Spark > Issue Type: Bug >Affects Versions: 1.5.2 > Environment: windows >Reporter: zihao > > I tried to import a local text(over 100mb) file via textFile in pyspark, when > i ran data.take(), it failed and gave error messages including: > 15/12/10 17:17:43 ERROR TaskSetManager: Task 0 in stage 0.0 failed 1 times; > aborting job > Traceback (most recent call last): > File "E:/spark_python/test3.py", line 9, in > lines.take(5) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\rdd.py", line 1299, > in take > res = self.context.runJob(self, takeUpToNumLeft, p) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\context.py", line > 916, in runJob > port = self._jvm.PythonRDD.runJob(self._jsc.sc(), mappedRDD._jrdd, > partitions) > File "C:\Anaconda2\lib\site-packages\py4j\java_gateway.py", line 813, in > __call__ > answer, self.gateway_client, self.target_id, self.name) > File "D:\spark\spark-1.5.2-bin-hadoop2.6\python\pyspark\sql\utils.py", line > 36, in deco > return f(*a, **kw) > File "C:\Anaconda2\lib\site-packages\py4j\protocol.py", line 308, in > get_return_value > format(target_id, ".", name), value) > py4j.protocol.Py4JJavaError: An error occurred while calling > z:org.apache.spark.api.python.PythonRDD.runJob. > : org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 > in stage 0.0 failed 1 times, most recent failure: Lost task 0.0 in stage 0.0 > (TID 0, localhost): java.net.SocketException: Connection reset by peer: > socket write error > Then i ran the same code for a small text file, this time .take() worked fine. > How can i solve this problem? -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
