Hello DC1ers,
I've added a script to cleanly bring down the DC1 pipeline - thus ensuring
that all log files are forwarded correctly to the master node followed by
the deletion of the local logfiles. This clean shutdown also ensures that
the current image in the input queue completes its processing through
the pipeline.
The script is available from branch 502. It will work with any releases
numbered from 494 up. (It might work for 484 up but I don't know.)
The script is called by:
KillManager <policyDirectory> [<imageInputQueue>]
<policyDirectory> must always be provided.
If <imageInputQueue> is not supplied, the default is: "./queue"
Sample uses:
KillManager ~/policy
KillManager /gpfs_scratch1/rallsman/policy /gpfs_scratch1/rallsman/queue
Operational characteristics:
The manager's hostid and pid are derived from the central logging file.
A SIGTERM is sent to the manager.
A dummy file is sent to the input Queue to ensure the iteration loop
breaks out of its MPI barrier thus allowing termination processing
to proceed. (This is only an issue if the queue had previously
drained empty and the pipeline was 'waiting' on input.)
After the 'current' image has completed processing, the pipeline will
close down all mpi processes on all nodes and then exit.
This concludes the new script's documentation in its entirety.
Robyn
_______________________________________________
LSST-data mailing list
[email protected]
http://www.lsstmail.org/mailman/listinfo/lsst-data
