On Apr 25, 2012, at 8:25 AM, Louise-Amélie Schmitt wrote:
Hi,
Thanks a lot it actually helped. It is not exactly as straightforward in drmaa.py but somehow I could manage.
However, it was not the problem. For some reason, the user needs to write files from the node to job_working_directory/00X/XXXX/ and the latter is not world-writable. I had to make everyone chmod everything to 777 to make it work. Did I miss something?
That doesn't seem right. The subdirectory for your job underneath job_working_directory/00X/XXX/ should be chowned to the real user before the job is submitted and then chowned back to the galaxy user once the job is complete. --nate
Best, L-A
Le 24/04/2012 15:17, Alban Lermine a écrit :
Hi L-A,
I run Galaxy as real user on our cluster with pbs (free version).
We first configure LDAP authentification for having email account related to unix account (just cut the @curie.fr) Then I have modify pbs.py (in <GALAXY_DIR>galaxy-dist/lib/galaxy/jobs/runners)
I have just disconnected the pbs submission through python library and replace it by a system call (just like I send jobs to the cluster with command line), here is the code used:
galaxy_job_id = job_wrapper.job_id log.debug("(%s) submitting file %s" % ( galaxy_job_id, job_file ) ) log.debug("(%s) command is: %s" % ( galaxy_job_id, command_line ) )
# Submit job with system call instead of using python PBS library - Permit to run jobs as .. with sudo -u cmd prefix
galaxy_job_idSTR = str(job_wrapper.job_id) galaxy_tool_idSTR = str(job_wrapper.tool.id) galaxy_job_name = galaxy_job_idSTR+"_"+galaxy_tool_idSTR+"_"+job_wrapper.user torque_options = runner_url.split("/") queue = torque_options[3] ressources = torque_options[4] user_mail = job_wrapper.user.split("@") username = user_mail[0]
torque_cmd = "sudo -u username echo "+"\""+command_line+"\" | qsub -o "+ofile+" -e "+efile+" -M "+job_wrapper.user+" -N "+galaxy_job_name+" -q "+queue+" "+ressources
submit_pbs_job = os.popen(torque_cmd)
job_id = submit_pbs_job.read().rstrip("\n")
#Original job launcher #job_id = pbs.pbs_submit(c, job_attrs, job_file, pbs_queue_name, None)
pbs.pbs_disconnect(c)
Second thing I have done is to wait error and output file from torque in the finish_job function (if not, I never receive the output, seems to be your problem..), here is the code used:
def finish_job( self, pbs_job_state ): """ Get the output/error for a finished job, pass to `job_wrapper.finish` and cleanup all the PBS temporary files. """ ofile = pbs_job_state.ofile efile = pbs_job_state.efile job_file = pbs_job_state.job_file
# collect the output try:
# With qsub system call, need to wait efile and ofile creation at the end of the job execution before reading them
efileExists = os.path.isfile(efile) ofileExists = os.path.isfile(ofile) efileExistsSTR = str(efileExists) ofileExistsSTR = str(ofileExists)
while efileExistsSTR == "False": time.sleep( 1 ) efileExists = os.path.isfile(efile) efileExistsSTR = str(efileExists)
while ofileExistsSTR == "False": time.sleep( 1 ) ofileExists = os.path.isfile(ofile) ofileExistsSTR = str(ofileExists)
# Back to original code
ofh = file(ofile, "r") efh = file(efile, "r") stdout = ofh.read( 32768 ) stderr = efh.read( 32768 ) except: stdout = '' stderr = 'Job output not returned by PBS: the output datasets were deleted while the job was running, the job was manually dequeued or there was a cluster error.' log.debug(stderr)
* Last step is to allow galaxy user to run sudo
Hope it can help you finding your problem..
See you,
Alban
___________________________________________________________ Please keep all replies on the list by using "reply all" in your mail client. To manage your subscriptions to this and other Galaxy lists, please use the interface at: