source: palm/trunk/SCRIPTS/palm_wdd @ 1612

#!/usr/bin/python
# -*- coding: utf-8 -*-
#--------------------------------------------------------------------------------#
# This file is part of PALM.
#
# PALM is free software: you can redistribute it and/or modify it under the terms
# of the GNU General Public License as published by the Free Software Foundation,
# either version 3 of the License, or (at your option) any later version.
#
# PALM is distributed in the hope that it will be useful, but WITHOUT ANY
# WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
# A PARTICULAR PURPOSE.  See the GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License along with
# PALM. If not, see <http://www.gnu.org/licenses/>.
#
# Copyright 1997-2015  Leibniz Universitaet Hannover
#--------------------------------------------------------------------------------#
#
# Current revisions:
# -----------------
# 
# 
# Former revisions:
# -----------------
# $Id: palm_wdd 1612 2015-07-07 12:25:21Z maronga $
#
# 1611 2015-07-07 12:23:22Z maronga
# Initial revision
# 
#
# Description:
# ------------
# PALM watchdog client for monitoring batch jobs on a variety of hosts specified
# by the user. The watchdog server requires python 2.7 or higher installed on
# host to be monitored.
# 
# Instructions:
# -------------
# 1) Modify the header section of palm_wd
# 2) Move .wd.olddata and .wd.newdata to your palm directory 
#    (e.g. /home/user/current_version/.wd.newdata etc.)
# 3) Modify a copy of palm_wdd for each host to be monitored and move it to the
#    respective hosts
# 4) Start the client either from mrungui or from shell by "nohup palm_wd&"
#
# To do:
# ------
# 1) Add "Options", "Help" and "Manual"
# 2) Move user settings to a configuration file
#------------------------------------------------------------------------------!

import os
import sys
from subprocess import check_output

# START OF HEADER

# configuration for host
cmd_readqueue      = "showq | egrep "
cmd_tmpdir         = "/gfs1/tmp/"
cmd_canceljob      = "canceljob"
cmd_checkjob       = "checkjob"
cmd_realname_grep  = "AName"
cmd_starttime      = "showstart"
cmd_starttime_grep = "start in"

# END OF HEADER


action   = str(sys.argv[1])
data     = str(sys.argv[2])
if ( len(sys.argv) > 3 ):
   data2 =  str(sys.argv[3])


cmd_readqueue = cmd_readqueue + data
cmd_tmpdir    = cmd_tmpdir + data



# reading queuing system
def ReadQueue(username):
  
#  collect queuing information
   try:
      out = check_output(cmd_readqueue, shell=True)
      job_list = out.splitlines()
      out = None
#  do nothing for empty results list
   except:
      job_list = []
  
 
   job_data_tmp = []
   for j in range(0,len(job_list)):

      # Write temporary data array containing the job information.
      job_data_tmp.append(j)
      job_data_tmp[j] = job_list[j].split(" ")
      job_data_tmp[j] = filter(None, job_data_tmp[j])

      cmd_realname = cmd_checkjob + " " + job_data_tmp[j][0] + "|grep " + cmd_realname_grep 

      # retrieve real job name for all jobs
      try:  
         out = check_output(cmd_realname, shell=True)
         job_realname = out.split(" ")[1].rstrip()
      except:
         job_realname = "error"


      # for running jobs, determine progress                  
      if ( job_data_tmp[j][2] == "Running" ):
  
         # collect progress information
         cmd_progress = "cat " + cmd_tmpdir + "/" + username + "." + job_data_tmp[j][0].partition(".")[2] + "/PROGRESS"
         try: 

            devnull = open(os.devnull, 'w')
            out = check_output(cmd_progress, shell=True, stderr=devnull)
            progress_lines = out.splitlines()
            job_progress = progress_lines[1].split(" ")[1]
            out = None
         except:
            job_progress = "0" 

      else:
         job_progress = "0" 

      # return the job data
      job_starttime = GetStartTime(job_data_tmp[j][0])
      print job_realname + " " + job_data_tmp[j][0] + " " + job_data_tmp[j][3] + " " + job_data_tmp[j][2] + " " + job_progress + " "  + job_data_tmp[j][4] + " " + job_starttime


# check details of specific job
def CheckJob(jobid):

   cmd_checkjob_tmp = cmd_checkjob + " " + jobid

   try:  
      out = check_output(cmd_checkjob_tmp, shell=True)
      job_details = out
   except:
      job_details = "No details available." 

   return job_details


# cancel a specific job
def CancelJob(jobid):

   cmd_canceljob_tmp = cmd_canceljob + " " + jobid

   try:  
      out = check_output(cmd_canceljob_tmp, shell=True)
      job_canceled = out
   except:
      job_canceled = "Action failed." 

   return job_canceled


# retrieve estimated start time of job
def GetStartTime(jobid):

   cmd_starttime_tmp = cmd_starttime + " " + jobid + "|grep \"" + cmd_starttime_grep + "\"" 

   try:  
      out = check_output(cmd_starttime_tmp, shell=True)
      job_starttime = out.split()[5]
   except:
      job_starttime = "Action failed." 

   return job_starttime


def DoStopNow(username,jobid):

   # collect progress information
   cmd_dostop = "touch " + cmd_tmpdir + "/" + username + "." + jobid.partition(".")[2] + "/DO_STOP_NOW"
   try: 
      devnull = open(os.devnull, 'w')
      out = check_output(cmd_dostop, shell=True, stderr=devnull)
      out = None
   except:
      return_message = "Action failed."  
      return return_message

def DoRestartNow(username,jobid):

   # collect progress information
   cmd_dorestart = "touch " + cmd_tmpdir + "/" + username + "." + jobid.partition(".")[2] + "/DO_RESTART_NOW"
   try: 
      devnull = open(os.devnull, 'w')
      out = check_output(cmd_dorestart, shell=True, stderr=devnull)
      out = None
   except:
      return_message = "Action failed."  
      return return_message


# START OF MAIN
if ( action == "queue" ):
   ReadQueue(data)
elif ( action == "check"):
   print CheckJob(data)
elif ( action == "cancel"):
   print CancelJob(data)
elif ( action == "start"):
   print GetStartTime(data)
elif ( action == "stop"):
   print DoStopNow(data,data2)
elif ( action == "restart"):
   print DoRestartNow(data,data2)
else:
   print "Error. Action " + action + " unknown."