Getting error "Create Spark Context" on CDH 5.8.2 after upgrade to KNIME 3.5

Hi,
Recently I noticed that the Spark context nodes are failing while connecting to our CDH 5.8.2. We had set this up with spark-job-server-0.6.2.2-KNIME_cdh-5.11 last year and we recently updated to the latest KNIME 3.5. Not sure fi the job server needs to be updated etc.

KNIME Preferences are the same. Not sure if I’m missing anything here.

spark_error_log.txt (2.4 KB)

Attached is the log file for the error.

Thanks !
Mohammed Ayub

Hi Mohammed,
this seems to be a problem with the file system user rights. Can you have a look into the Spark Job Server log files for further details. You can find the Spark Job Server logs on the machine the job server is running in /var/log/spark-job-server. If you are using context per JVM you will find a separate directory for each created Spark context. Have a look at the log files within this directory to get further information.
Bye
Tobias

Sure. Here is the latest log file I pulled out from \var\log\spark-job-server. Not sure if some configurations are changed/missing.

spark-job-server-log.txt (100.7 KB)

I was using the Context per JVM and it had created a separate directory for each try. Attached is the more detailed spark-job-server.log file. Let me know if anything I may need to change here.

Thanks !

spark_job_server_config_details.txt (55.2 KB)

Hi,
is it possible that you have started the Spark Job Server as user root via the server_start.sh script? This is not recommended. You should preferably start the job server via the init scripts (see section: Starting the Spark Job Server in the Spark Job Server installation guide). The job server is then running with the spark-job-server user who has the appropriate rights in the HDFS.
Bye
Tobias

Thanks @tobias.koetter
Sorry for the later reply on this. Good news is I was successful in updating the KNIME Spark connections from 0.6.2.1 to 0.6.2.3. I set it up from scratch using the installation guide so the tricky part was deleting the existing contents from “/tmp/spark-job-server” directory and then starting job server using “/etc/init.d/${LINKNAME} start”.

Below is the pic for reference:

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.