#!/usr/bin/env python

import os, sys, commands, time
import sos_classes, killAndChildren

""" 
sos_apocontrol:

sos_apocontrol controls the running of the sos daemon.  Everything is hard coded to APO--no
comfort is given as a general purpose configurable control program.

Written by Gary Kushner (LBL).  Oct 2009.

"""

####
##   Globals Block. 
####

#CONTROLDIR = "/tmp/sos/control"
#STARTCMD = "sos_apostart -t -x -s"
CONTROLDIR = "/home/eboss/sos/control"
STARTCMD = "sos_apostart -g"
DIEFILE = os.path.join(CONTROLDIR, sos_classes.Consts().dieFileName)


####
def usage():
	"""Display usage and exit"""
	
	usageCMD = os.path.basename(sys.argv[0])

	print usageCMD + "\n\nusage: start|allow|stopWait|stop|status|stopKill"
	print \
	"""
	start:
	  delete the die file and run sos_apostart
	
	allow:
	  delete the die file but don't start the daemons

	stop:
	  touch the die file.  With the die file, the daemon will not run even if restarted
	  manually or via cron.

	stopWait:
	  touch the die file.  With the die file, the daemon will not run even if restarted
	  manually or via cron.  stopWait will not return to the shell until all the 
	  processes are stopped.  Normally, stopWait will return in less than 5 minutes, 
	  however, if an idl process is active, stopWait can take more time.
	
	stopKill:
	  touch the die file.  With the die file, the daemon will not run even if restarted
	  manually or via cron.  stopKill will then kill all the processes by sending them
	  a SIGINT, sleeping, and then sending any stragglers a SIGKILL.  It is best to try
	  stopWait first. 

	status:
	  use ps to display running sos_runnerd processes

	preferred sequence upon failure of one or more runnerd processes:
	  sos_apocontrol status   --- if processes not running then
	  sos_apocontrol stopWait
	  sos_apocontrol status   --- check that zero processes are running
	  sos_apocontrol start

	"""
	sys.exit(1)


####
def start():
	if os.path.isfile(DIEFILE):
		os.unlink(DIEFILE)
	rc = os.system(STARTCMD)
	if rc != 0:
		print >> sys.stderr, "sos_apostart failed: " + str(rc)
	time.sleep(5)
	print " "
	print "Running status now.  There should be 4 processes running."
	status()

####
def allow():
	if os.path.isfile(DIEFILE):
		os.unlink(DIEFILE)
	print "die file removed -- sos_runnerd allowed to run."
	
		
####
def stop():
	open(DIEFILE, 'w').close()
	print "die file created -- processes should stop.  If an idl command is"
	print "executing, it will be allowed to finish before the process stops."

	
####
def stopWait():
	open(DIEFILE, 'w').close()
	print "die file created -- processes should stop.  If an idl command is"
	print "executing, it will be allowed to finish before the process stops."
	print "Waiting . . .",
	while True:
		sys.stdout.flush()
		rc = commands.getstatusoutput("ps ax | grep runnerd | grep -v grep | grep -Eo sos_runnerd.*$ | sed G")
		if len(rc[1]) == 0:
			break
		time.sleep(5)
		print ".",
		
		
####
def stopKill():
	open(DIEFILE, 'w').close()
	print "die file created -- processes should stop.  Pausing one minute to check..."
	time.sleep(70)
	
	rc = commands.getstatusoutput("ps ax -o pid,args | grep runnerd | grep -v grep | grep -Eo ^.*python | grep -Eo [0-9]+")
	if len(rc[1]) != 0:
		print "Killing sos and associated idl processes..."
		sys.stdout.flush()
		pids = rc[1].split("\n")
		for pid in pids:
			killAndChildren.kill(int(pid))

	rc = commands.getstatusoutput("ps ax -o pid,args | grep runnerd | grep -v grep | grep -Eo ^.*python | grep -Eo [0-9]+")
	if len(rc[1]) != 0:
		print "** Could not kill processes **"
	else:
		print "No processes running."
	sys.stdout.flush()
			
####
def status():
	ps_sos_daemons = "ps ax | grep runnerd | grep -v grep | grep -v 'svn commit' | grep -Eo sos_runnerd.*$ | sed G" 
	rc = commands.getstatusoutput(ps_sos_daemons)
	if len(rc[1]) == 0:
		print "No processes running!"
		return
	print str(rc[1].count("sos_runnerd")) + " processes running:\n"
	os.system(ps_sos_daemons)
		
	

####
def main(argv):
	"""control the apo sos daemon"""
	
	if len(argv) != 1:
		usage()
	cmd = argv[0]

	if cmd == "start":
		start()
	elif cmd == "allow":
		allow()
	elif cmd == "stop":
		stop()
	elif cmd == "stopWait":
		stopWait()
	elif cmd == "stopKill":
		stopKill()
	elif cmd == "status":
		status()
	else:
		usage()

### Start of script

if __name__=='__main__':
	main(sys.argv[1:])
	print os.path.basename(sys.argv[0]) + " finished running!"