#!/bin/bash

# Description:
#   This script aids management of condor compute nodes.  The script
#   takes a command as an argument and waits until condor_startd is
#   not running before executing the command.
# Use with: condor_off -peaceful <node>

# Original verions TDR Feb 29, 2008
# Change to looking at startd TDR Mar 17, 2008
# Seems that when you specify that the master be shutdown,
#   a gracefully instead of peaceful shutdown is done.
# So we don't want to shutdown the master and instead this
#   script should look at condor_startd
#  ends jobs.  Should do: condor_off -peaceful <nodename> and
#  then wait for condor_startd to go away

#Usage:
# condor_waiter {command}
# command should be specified as one argument (put it in quotes)

if [ "$#" -eq "0" ]; then
  echo
  echo Usage:
  echo condor_waiter '{command}'
  echo '  command should be one argument (put it in quotes)'
  exit 1
fi

# if not launched with --child option, relaunch in background
if [ "x$1" != "x--child" ]; then
  $0 --child $@ >> /tmp/condor_waiter.$$.out 2>&1 &
  exit 0
fi

# shift off the --child option
shift

# really want to look at startd, TDR Mar 17.
startd_pid=`ps h -o pid -C condor_startd`
if [ "$?" -eq "0" ]; then
    timestart=$(date -u +%s)
    logger -s -- condor_waiter $$: condor_startd $startd_pid is running waiting for it to end before executing: $@
    # wait for condor_master to go away
    # kill -0 is just a way to check if the process is alive
    while kill -0 $startd_pid 2> /dev/null; do sleep 10; done
    timeend=$(date -u +%s)
    logger -s -- condor_waiter $$: condor_startd $startd_pid has gone away after $((timeend-timestart)) secs will now execute: $@
else
    logger -s -- condor_waiter: condor_startd was not running will now execute: $@ 
fi

# finally do it
$@

exit 0