Debugging Python workers

[Michal Romaniuk <mi...@imperial.ac.uk>]

Hi,

I'm trying to run a Python job that keeps failing. It's a map followed
by collect, and works fine if I use Python's built-in map instead of Spark.

I tried to replace the mapping function with an identity (lambda x: x)
and that works fine with Spark, so Spark seems to be configured correctly.

The error I get is:

org.apache.spark.SparkException (org.apache.spark.SparkException: Python
worker exited unexpectedly (crashed))

The problem is that I can't see what went wrong in the Python code. Any
ideas on how to debug this?

Thanks,
Michal

Re: Debugging Python workers

[Jey Kottalam <je...@cs.berkeley.edu>]

Hi Michal,

Try taking a look at the stderr output of the failed executor. You can
do this by logging into the Spark worker that had the failed executor,
and look for the executor ID in the "work" subdirectory. You should
see a file named "stderr" that has the error messages from the failed
executor.

Hope that helps,
-Jey

On Thu, Nov 21, 2013 at 1:01 PM, Michal Romaniuk
<mi...@imperial.ac.uk> wrote:
> Hi,
>
> I'm trying to run a Python job that keeps failing. It's a map followed
> by collect, and works fine if I use Python's built-in map instead of Spark.
>
> I tried to replace the mapping function with an identity (lambda x: x)
> and that works fine with Spark, so Spark seems to be configured correctly.
>
> The error I get is:
>
> org.apache.spark.SparkException (org.apache.spark.SparkException: Python
> worker exited unexpectedly (crashed))
>
> The problem is that I can't see what went wrong in the Python code. Any
> ideas on how to debug this?
>
> Thanks,
> Michal
>