condor_node_check

#!/usr/bin/env python

#  Worker node health check script, to run from condor startd

from mc_util import *
import sys, os
import memcache
import socket
import subprocess

hostname = socket.gethostname() 
node = shortname(hostname)

#
# define directories to test
# 
SCRATCH = "/scratch"
TMP = "/tmp"
SLASH = "/"
FNAME = "node_check"


def setOffline(msg):
   mc_set( node+".status", "offline")
   mc_set( node+".message", msg)
   print "NodeOnline = False"
   print "NodeOfflineReason = %s" % msg

def setOnline():
   mc_set( node+".status", "online")
   print "NodeOnline = True"


# Test if the node has been set offline manually
MANUALSTATUS = mc_get( node+".manualstatus" )

if MANUALSTATUS == 'offline':
    MANUALREASON = mc_get( node+".manualreason" )
    setOffline("Manual offline: %s " % MANUALREASON)
    sys.exit()
#
# TESTS
#
def can_write( mypath, myfile ):
  if os.path.exists( "%s" % (mypath) ):
    if not os.path.isfile( "%s/%s" % (mypath, myfile) ):
      if ( open( "%s/%s" % (mypath, myfile), 'w').close() ):
        setOffline("cannot write to %s" % mypath) 
        sys.exit()
      else:
        if debug: print "can write to %s" % mypath
        os.remove( "%s/%s" % (mypath, myfile))

def correct_perms( mypath, myperms):
  if os.path.exists( "%s" % (mypath) ):
    PERMS = os.stat(mypath).st_mode
    if not ( PERMS & 0777 == 1777 ):
      setOffline("permissions on %s should be 1777, are %s" % mypath, PERMS)
      if debug: print "permissions on %s should be 1777, are %s" % (mypath, PERMS)
      sys.exit()

def disk_free ( mypath, amount ):  
  if os.path.exists( "%s" % (mypath) ):
    df = subprocess.Popen(["/bin/df", "-Ph", "%s" % mypath], stdout=subprocess.PIPE)
    output = df.communicate()[0]
    device, size, used, available, percent, mountpoint = output.split("\n")[1].split()

    available = available.split(".")[0]

    if ( available < amount ):
      setOffline("%s disk space available is %s GB" % mypath, available)
      sys.exit()


##
# run the checks
## 
# can_write( SCRATCH )
# can_write( TMP )
# correct_perms( SCRATCH )
# disk_free ( SCRATCH, 5 )
# disk_free ( SLASH, 5 )

setOnline()
sys.exit()