KeyError job_dict["state"] + strange DB Entry

[oleweidner]

This error was reported in bigjob-users by Scott Michael:

KeyError in agent:

  File "/N/soft/rhel6/python/2.7.3/lib/python2.7/site-packages/threadpool.py", line 156, in run
    result = request.callable(*request.args, **request.kwds)
  File "/N/soft/rhel6/python/2.7.3/lib/python2.7/site-packages/bigjob/bigjob_agent.py", line 720, in start_new_job_in_thread
    if(job_dict["state"]==str(bigjob.state.Unknown)):
KeyError: 'state'
Traceback (most recent call last):
  File "/N/soft/rhel6/python/2.7.3/lib/python2.7/site-packages/threadpool.py", line 156, in run
    result = request.callable(*request.args, **request.kwds)
  File "/N/soft/rhel6/python/2.7.3/lib/python2.7/site-packages/bigjob/bigjob_agent.py", line 720, in start_new_job_in_thread
    if(job_dict["state"]==str(bigjob.state.Unknown)):
KeyError: 'state'

Strange CU description in Redis:

instead of:

 {'Executable': '/N/dc2/projects/BDBS/cijohnson//dorunner.sh', 
'WorkingDirectory': 
'/N/dc2/projects/BDBS/cijohnson//./lp4.0058bm6.8292', 
'NumberOfProcesses': '1', 'start_time': '1390836474.31', 
'Environment': "['TASK_NO=4687']", 'state': 'Unknown', 'Arguments': 
"['/N/dc2/projects/BDBS/cijohnson/./lp4.0058bm6.8292 
tu1783717_58.in\\n']", 'Error': 'tu1783717_58.err', 'Output': 
'tu1783717_58.out', 'job-id': 
'sj-976b4976-8767-11e3-adde-001fc6d94bec', 'SPMDVariation': 'single'} 

this:

 {'a': 'd', 'c': '-', 'b': 'e', 'd': 'e', 'f': 'c', '-': '0', '3': 
'-', '1': 'e', '0': '1', 's': 'j', '7': 'f', '6': 'd', '9': '4', '8': 
'7'} 

[oleweidner]

The complete job script:

import os
import commands
import sys
import pilot
import traceback

""" DESCRIPTION: Tutorial 1: A Simple Workload 
Note: User must edit USER VARIABLES section
This example will not run if these values are not set.
"""

# ---------------- BEGIN REQUIRED PILOT SETUP -----------------

# Distributed Coordination Service - Redis server and password
REDIS_PWD   = "ILikeBigJob_wITH-REdIS"# Fill in the password to your redis server
REDIS_URL   = "redis://%[email protected]:6379" % REDIS_PWD

# Resource Information
HOSTNAME     = "localhost"# Remote Resource URL
USER_NAME    = "scamicha"# Username on the remote resource
SAGA_ADAPTOR = "pbs"# Name of the SAGA adaptor, e.g. fork, sge, pbs, slurm, etc.
# NOTE: See complete list of BigJob supported SAGA adaptors at:
# http://saga-project.github.io/BigJob/sphinxdoc/tutorial/table.html

# Fill in queue and allocation for the given resource 
# Note: Set fields to "None" if not applicable
QUEUE        = "batch" #Add queue you want to use
PROJECT      = "None"# Add project / allocation / account to charge

WALLTIME     = 1440# Maximum Runtime (minutes) for the Pilot Job

WORKDIR      = os.getenv("HOME")+"/agent" # Path of Resource Working Directory
# This is the directory where BigJob will store its output and error files

SPMD_VARIATION = "None"# Specify the WAYNESS of SGE clusters ONLY, valid input '12way' for example

PROCESSES_PER_NODE = 8# Valid on PBS clusters ONLY - this is the number of processors per node. One processor core is treated as one processor on PBS; e.g. a node with 8 cores has a maximum ppn=8

PILOT_SIZE = 128# Number of cores required for the Pilot-Job

# Job Information

datadir = "/N/dc2/projects/BDBS/cijohnson/"
files = []
os.chdir(datadir)
input = open('files.todo.01','r')
for line in input:
    files.append(line)
NUMBER_JOBS=len(files)

# Continue to USER DEFINED TASK DESCRIPTION to add 
# the required information about the individual tasks.

# ---------------- END REQUIRED PILOT SETUP -----------------
#

def main():
    try:
        # this describes the parameters and requirements for our pilot job
        pilot_description = pilot.PilotComputeDescription()
        pilot_description.service_url = "%s://%s@%s" %  (SAGA_ADAPTOR,USER_NAME,HOSTNAME)
        pilot_description.queue = QUEUE
        pilot_description.number_of_processes = PILOT_SIZE
        pilot_description.working_directory = WORKDIR
        pilot_description.walltime = WALLTIME
    pilot_description.processes_per_node = PROCESSES_PER_NODE
    pilot_description.spmd_variation = SPMD_VARIATION

        # create a new pilot job
        pilot_compute_service = pilot.PilotComputeService(REDIS_URL)
        pilotjob = pilot_compute_service.create_pilot(pilot_description)


        # submit tasks to pilot job
        tasks = list()
        for i in range(0,NUMBER_JOBS-1):
        directory = files[i].rsplit('/',1)[0]
        file = files[i].rsplit('/',1)[1]
    # -------- BEGIN USER DEFINED TASK DESCRIPTION --------- #
            task_desc = pilot.ComputeUnitDescription()
            task_desc.executable = datadir+'/dorunner.sh'
            task_desc.arguments = [datadir+directory+' '+file]
            task_desc.environment = {'TASK_NO': i}
            task_desc.number_of_processes = 1
        task_desc.spmd_variation = 'single' # Valid values are single or mpi
        task_desc.working_directory=datadir+'/'+directory
            task_desc.output = file.rsplit('.',1)[0]+".out" 
            task_desc.error = file.rsplit('.',1)[0]+".err"
    # -------- END USER DEFINED TASK DESCRIPTION --------- #

            task = pilotjob.submit_compute_unit(task_desc)
            print "* Submitted task '%s' with id '%s' to %s" % (i, task.get_id(), HOSTNAME)
            tasks.append(task)

        print "Waiting for tasks to finish..."
        pilotjob.wait()

        return(0)

    except Exception, ex:
            print "AN ERROR OCCURRED: %s" % ((str(ex)))
            # print a stack trace in case of an exception -
            # this can be helpful for debugging the problem
            traceback.print_exc()
            return(-1)

    finally:
        # alway try to shut down pilots, otherwise jobs might end up
        # lingering in the queue
        print ("Terminating BigJob...")
        pilotjob.cancel()
        pilot_compute_service.cancel()


if __name__ == "__main__":
    sys.exit(main())

[drelu]

Sorry, I cannot replicate this based on this script. I neither have the executable nor the input files. This needs to be narrowed down.

[scamicha]

Hi there,

I'm the user that originally wrote into the mailing list with this issue. I don't think you'll be able to exactly replicate this problem as I was attempting to run ~120K subjobs and the input data set is a little over 4TB in size. I'd be happy to get it to you but, it's probably technical infeasible at best. I was able to run the pilot job with the debug level at 5. The log file is located at https://iu.box.com/s/3611nik4aoop686vbrn9