Monday, November 14, 2011

SULU 49c

Summary

* Corrected the dataset name pattern in the new LFC convention for user
jobs that was introduced in pilot version 49a. Previously single char dirs
were not allowed.

* Added new JEMstub module from Tim Dos Santos. JEM is now using the
taskId field in the decision taking.

* The pilot is now reporting outbound connections discovered by runGen to
the logger.

* Changed napping time from 60 s (too short) to 5/10 minutes in [pilot]
stage-in/out retries.

* Now aborting stage-in if only one input file and the file is a DBRelease
file that in turn is locally available and should not be subject to any
LFC checks. Problem seen in user jobs.

* Improved exception handling around dq2.listFilesInDataset(). A problem
with the central catalog last week led to "Pilot has no child processes"
errors in the pilot. The problematic dq2 function did not inherit from the
Exception class which was assumed in the pilot.

Version info

Created abortStageIn() used by mover_get_data() (Mover)
Corrected dsname pattern in getUserLFCDir(), previously single char dirs were not allowed due to bug in pattern (SiteMover)
Added new JEMstub module from Tim (JEMstub)
Adding log extracts to node structure if "outbound connections" found in log string, in getNodeStructure() (PandaServerClient)
Now reporting outbound connections to logger, in buildLogExtracts() (JobLog)
Changed napping time from 60 s to 5/10 minutes in mover_get_data()/mover_put_data() (Mover)
Improved exception handling in getFileInDataset() (a problem with the central catalog caused a few thousand jobs to
fail but was not correctly labeled by the pilot, instead led to a "Pilot has no child processes" error) (SiteMover)
Improved exception handling in getFileInDataset() (SiteMover)

No comments: