Troubleshooting R-UDF, trying to decode the log files

riccardo_pasquini

Hello
we're running a function which properly works when executed not concurrently but it seems to crash when multiple concurrent instances are executed.
It has been developed avoiding shared state and it is invoked on a data set of about 6M rows x 50columns.
I'm not able to decode the UDxFencedProcesses.log, can you help me to understand the meaning of problematic traces and which causes can bring to them?

Jim_Knicely

Hi,

Which version of Vertica are you using? How many nodes are in your cluster? Have you tried running the UDx concurrently on a smaller data set? By crash, what do you mean? Do you get an error?

Thanks!

riccardo_pasquini

Hi Jim

thank you for answering; to add some context:
we're using Vertica 9.0.1 delivered as a cluster of 3 nodes. Unit tests include concurrent executions on smaller data sets which terminate succesfully. The behavior is that the SQL statement invoking such function hangs, without recovering its execution...

Jim_Knicely

Are you running the SQL that calls the function locally on one of the nodes (i.e. via vsql) or remotely using some other tool like DbVisualizer?

When the query is "hung", can you check the SESSIONS table to make sure the session is still connected and running the query?

SELECT current_statement FROM sessions WHERE session_id = <>;

And check these system tables for any obvious issues:

• V_MONITOR.ERROR_MESSAGES
• V_MONITOR.RESOURCE_POOL_STATUS
• V_MONITOR.RESOURCE_ACQUISITIONS
• V_MONITOR.RESOURCE_QUEUES
• V_MONITOR.RESOURCE_REJECTIONS

Also, the vertica.log (on each node) need to be reviewed for errors / unexpected session disconnect (for the session running the SQL).