question Re: Spark-submit suddenly doesn't recognize modules on the cluster in Support Questions

Spark-submit suddenly doesn't recognize modules on the cluster

imule — Fri, 30 Sep 2022 09:04:10 GMT

Hi,

We're experiencing a peculiar issue these past few days - our python scripts started failing when run with spark-submit. On first glance our logs show that the scripts are encountering syntax errors whenever we are using code related to modules, but further troubleshooting showed that in actuality the modules are the issue. When we comment out pieces of code that throw syntax errors, we instead receive import errors ("No module named...").

We have two versions of python on our cluster but it appears that spark-submit still is using the proper python version with all our modules installed on it. Our scripts run just fine through pyspark, for some reason however, spark-submit does not recognize the imported modules when we run scripts through it.

What is more, YARN doesn't seem to recognize these jobs as failed, they are not logged at all, probably because they crash as soon as we start importing modules. So basically we do not have access to YARN logs for these jobs.

Any insight would be greatly appreciated.

Thanks.

Re: Spark-submit suddenly doesn't recognize modules on the cluster

RangaReddy — Fri, 30 Sep 2022 09:09:58 GMT

Hi @imule

Add the following parameter to your spark-submit

--conf spark.yarn.appMasterEnv.PYSPARK_PYTHON=<python3_path> --conf spark.yarn.appMasterEnv.PYSPARK_DRIVER_PYTHON=<python3_path>

Note:

1. Ensure python3_path exists in all nodes.

2. Ensure required modules are installed in each node.

Re: Spark-submit suddenly doesn't recognize modules on the cluster

Shelton — Sun, 02 Oct 2022 17:29:35 GMT

@imule

Firstly I would like to ask if there were any changes in the cluster i.e Patching or rpm ?
if the spark-submit was running successfully before you need to know this could be linked to the python version

On the edge/gateway node

# python -V

# Python 3.7.7

# conda deactivate

# Python 2.7.5

Then try relaunching your spark-submit

Re: Spark-submit suddenly doesn't recognize modules on the cluster

VidyaSargur — Fri, 07 Oct 2022 08:58:44 GMT

@imule, Has the reply helped resolve your issue? If so, please mark the appropriate reply as the solution, as it will make it easier for others to find the answer in the future.