prelz / blah Goto Github PK

A quoting problem arises when trying to assign a value with quotes to a variable, e.g.:

+remote_environment = "PEGASUSHOME=/auto/rcf-40/vahi/SOFTWARE/pegasus/pegasus-4.4.0cvs KEY0='VALUE 0' KEY1=VALUE1A TEST1='""Karan vahi""'"

When the BNotifierSlurm cannot access the registry (e.g. because of wrong ownership, see GGUST Ticket #128229), it leaks memory and keeps a huge number of open files.

Make the sleep between the 2 'qacct' commands configurable via blah.config.
See https://ggus.eu/index.php?mode=ticket_info&ticket_id=125594

Hi,

I know (or at least I think) that 1.23 is not released so this is more for notifying the problem. If this is a known or expected problem, do not hesitate to close this.
I'm giving a try to the 1.23 release (for the officially released BLAH really does not work as it should with condor batch), and it works fine, but I've noticed that, when submitting multicore jobs, a '-h' argument is passed to the condor_submit.sh which is not understood by the script (which fails submission). This can be easily worked around, of course, but I guess you'd like to have this fixed before release.

Regards,
Andrea

The git describe command in CMakeLists.txt:

execute_process(COMMAND git describe
                WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}
                OUTPUT_VARIABLE GIT_TAG
                ERROR_QUIET)

works only when issued from the .git directory.
The WORKING_DIRECTORY should point to the .git directory instead of the current directory.

By default when you remove a job from condor queue it is also immedetialy removed from the condor_q jobs list. We have seen that in this condition BLAH/CREAM does not update the job status so the job remains in the last registered status for two months [blah default purge interval]. With high jobs cancel rate the DB can grow indefinitely, slowing the queries thus lead the CREAM CE to an unresponsive status.

Here at INFN-Bari T2 HTCondor cluster the JOB_IS_FINISHED_INTERVAL knob is set to the default 0,
because we can access to the job information using the condor_history command.
But from blah/CREAM point of view it is mandatory to have access to the removed job trough the condor_q command. For this reason the "leave_in_queue" knob in the file "src/scripts/condor_submit.sh" should care about the removed jobs too.