XGrid getXGridOutput.py

Back to C-Program on the XGrid

---

Modified --Thiebaut 22:46, 20 October 2008 (UTC)

#! /usr/bin/python
# getXGridOutput.py
# D. Thiebaut
#
# A filter program receiving (through a pipe, typically) the output of the command
# 
#    xgrid -job submit xxxx
#
# and displaying the result of the computation, including execution time and 
# the output of each process.
# 
# Example:
# 
# runQueensAll.sh 15 | getXGridOutput.py 
# Job 1250 stopped: Execution time: 2.000000 seconds
# First Queen in column 0 of 15x15 board:  69516 solutions found
#
# Job 1251 stopped: Execution time: 2.000000 seconds
# First Queen in column 1 of 15x15 board:  98156 solutions found
# ...
# ...
# Job 1264 stopped: Execution time: 2.000000 seconds
# First Queen in column 14 of 15x15 board:  69516 solutions found
#
# Total execution time: 6.000000 seconds
#

import sys
import os
import time

jobStatus = {}
RUNNING   = 1
STOPPED   = 0
STARTTIMES= []
ENDTIMES  = []

def printResults( jobId ):
    """Prints the output of job #jobId.  Format of time: 2008-09-27 19:09:27"""
    global STARTTIMES, ENDTIMES
    [status, startTime, endTime] = jobStatus[ jobId ]
    time1 = time.mktime( time.strptime( startTime, "%Y-%m-%d %H:%M:%S" ) )
    time2 = time.mktime( time.strptime( endTime, "%Y-%m-%d %H:%M:%S" ) )
    STARTTIMES.append( time1 )
    ENDTIMES.append( time2 )
    print "Job %d stopped: Execution time: %f seconds"  % ( jobId,  time2-time1 )
    dothis = "xgrid -job results -id %d" % jobId
    handle = os.popen( dothis )
    while 1:
        line = handle.readline()
        if line=="": break
        print line,

def deleteJob( jobId ):
    """ deletes a job given its jobId """
    #--- prepare command ---
    dothis = "xgrid -job delete -id %d" % jobId
    handle = os.popen( dothis )

    #--- read output from xgrid ---
    while 1:
        line = handle.readline()
        if line=="": break
        if line=="}": break

    return 1


def getJobsStatus( jobIds ):
    """Polls all the jobs and see if they are still running.
    returns the number of stopped jobs"""

    global RUNNING, STOPPED
    noStoppedJobs = 0

    for jobId in jobIds:

        #--- don't poll the xgrid if the job is already stoppped---
        status = jobStatus[ jobId ]
        if status[ 0 ]==STOPPED:
            noStoppedJobs += 1
            continue

        #--- poll xgrid ---
        dothis = "xgrid -job attributes -id %d" % jobId
        handle = os.popen( dothis )

        #--- read the output from xgrid ---
        while 1:
            line = handle.readline()
            if line=="": break
            if line.find( "dateStarted" )!=-1:
                status[1] = line[ line.find( "=" )+1 :-8 ] 
                status[1] = status[1].strip()
            if line.find( "dateStopped" )!=-1:
                status[2] =  line[ line.find( "=" )+1 :-8 ] 
                status[2] = status[2].strip()
            #---- if the job is now finished record it and display its output ---
            if line.find( "jobStatus" )!=-1 and \
                    line.find( "Finished" ) != -1:
                status[0] = STOPPED 
                noStoppedJobs += 1
                printResults( jobId )
                deleteJob( jobId )

        #--- record status  ---
        jobStatus[ jobId ] = status

    #--- returned # of stopped jobs ---
    return noStoppedJobs

def main():
    """ ---------------------------- M A I N ---------------------------
    """
    count = 1
    jobIds = []
    for line in sys.stdin:
        if line.find( "jobIdentifier" ) != -1:
            jobIds.append( int( line[ line.find("=")+1 :-2 ] ) )

    #--- create job status records ---
    for jobId in jobIds:
        jobStatus[jobId] = [ RUNNING, 0, 0 ] # running, start time, end time

    #--- keep on going until all jobs have stopped ---
    while True:
        #--- get job status ---
        noStoppedJobs = getJobsStatus( jobIds )
        if noStoppedJobs == len( jobIds ):
            break
        time.sleep( 0.1 ) # wait 0.1 second before polling the xgrid again
    
    #--- print total execution time ---
    #print "ENDTIMES = ", ENDTIMES
    #print "STARTTIMES = ", STARTTIMES
    print "\nTotal execution time: %f seconds\n\n" % ( max( ENDTIMES )-min( STARTTIMES ) )



main()