use daemon and PIDLock

hbc

@@ -1,9 +1,13 @@
 #!/usr/bin/env python
 # $Id: hbc,v 1.9 2012/03/29 02:08:36 andreas Exp $
 
-# requre python-filelock on Linux
-# require py27-lockfile on *bsd
-# or run  sudo easy_install-2.7 lockfile 
+# require on Linux
+#	python-filelock 
+#	python-daemon vs 1.61 or >
+# on *bsd
+#	py27-lockfile 
+# 	py-27-daemon
+# or run  sudo easy_install-2.7 lockfile daemon
 import sys
 import time
 import socket
@@ -14,24 +18,29 @@ import string
 import select
 import errno
 import traceback
-from  lockfile import FileLock
+import lockfile 
+import daemon
+import daemon.pidfile
+import syslog
+
 
 PORT=50003
 INTERVAL=10
 DBG = True
 
-sock=None
+sock = None
+up = True
+ackcount = 0
 
 class NullDevice:
     def write(self, s):
         pass
 
 
-def handler(signum, frame):
-	global up
-	if up == 0:
-		return
-	sys.exit(0)
+def syslogtrace(note):
+	logm = '%s hbc died: \n%s' % (note, traceback.format_exc())
+	for l in logm.split('\n'):
+		syslog.syslog('  tb: %s' % l)
 
 
 def getsock(host):
@@ -46,35 +55,43 @@ def getsock(host):
 	else:
 		return None
 	if verbose:
-		print "socktype: %s" % af_type
+		syslog.syslog("socktype: %s" % af_type)
 	sock=socket.socket(af_type, socket.SOCK_DGRAM)
 	sock.setsockopt(socket.SOL_SOCKET,socket.SO_REUSEADDR, \
             sock.getsockopt(socket.SOL_SOCKET,socket.SO_REUSEADDR) | 1)
-	if verbose: print "get socket %s" % sock
+	if verbose: syslog.syslog("get socket %s" % sock)
 	
 	return sock
 
+
 def socksend(msg, tohost):
 	global sock
 
 	if sock == None:
 		sock=getsock(tohost[0])
 	sock.sendto(msg, tohost)
-	if verbose: print "msg %s sent" % msg
+	if verbose: syslog.syslog("msg %s sent" % msg)
+
 
 def process():
+	global up, sock, ackcount
+
+	if sock == None:
+		sock=getsock(tohost[0])
+
 	ackcount=0
 	lastT=time.time()
 	while up:
 		sleep=(lastT+interval) - time.time() 
-		if verbose: print "sleep %s" % sleep
+		if verbose: syslog.syslog("sleep %s" % sleep)
 		if sleep > 0:
 			try:
 				r=select.select([sock.fileno()],[],[],sleep)
-	#			time.sleep(interval)
 			except:
+				if up:
+					syslogtrace('select')
 				break
-			if verbose: print r
+			if verbose: syslog.syslog("r is %s" % str(r))
 			if sock.fileno() in r[0]:
 				data, addr = sock.recvfrom(1024)
 				if  data == "ACK":
@@ -83,22 +100,35 @@ def process():
 					try:
 						os.system(data)
 					except:
+						syslogtrace('System')
 						pass
 				continue	
 		lastT=time.time()
 		for hb_host in hb_hosts:
 			try:
 				msg="interval=%s;name=%s;time=%s;acks=%s" % (interval, iam, time.time(), ackcount)
-				if verbose: print "sock.send('%s', (%s, %s))" % (msg, hb_host, hb_port)
+				if verbose: syslog.syslog("sock.send('%s', (%s, %s))" % (msg, hb_host, hb_port))
 				socksend(msg, (hb_host, hb_port))
 			except:
 				pass
 	
 
+def cleanup(a, b):
+	global up, sock, ackcount
+	up = False
+	syslog.syslog('exit a=%s b=%s' % (str(a), str(b)))
+	msg="shutdown=1;name=%s;acks=%s" % (iam, ackcount)
+	for hb_host in hb_hosts:
+		if verbose: syslog.syslog("hbc: sock.send('%s', (%s, %s))" % (msg, hb_host, hb_port))
+		socksend(msg, (hb_host, hb_port))
+	time.sleep(1)
+	sock.close()
+
+
 msgonly=False
 helpflag=False
 verbose=False
-daemon=False
+fdaemon=False
 optlist=[]  
 args=[]
 msgboot=[]
@@ -116,7 +146,7 @@ for o,a in optlist:
 	elif o == '-c':
 		configfile=a
 	elif o == '-d':
-		daemon=True
+		fdaemon=True
 	elif o == '-h':
 		helpflag=True
 	elif o == '-m':
@@ -228,47 +258,29 @@ if msgonly:
 
 #
 # 
-if daemon:
-	pid=os.fork()
-	if pid > 0:
-		if verbose:
-			print "daemoinizing... pid=%d" % pid
-		sys.exit(0)
+syslog.openlog(logoption=syslog.LOG_PID, facility=syslog.LOG_DAEMON)
+if fdaemon:
+	print "daemoinizing... %s" % os.getpid()
 
+	context = daemon.DaemonContext(
+		working_directory='/tmp',
+		umask=0o002,
+		pidfile=daemon.pidfile.TimeoutPIDLockFile('/tmp/hbc.pid', acquire_timeout=-1),
+		)
 
-	if not DBG:
-		os.close(0)
-		os.close(1)
-		os.close(2)
-		sys.stdin.close()
-		sys.stdout = NullDevice()
-		sys.stderr = NullDevice()
-	os.chdir("/")
-	os.setsid() 
-	os.umask(0) 
+	context.signal_map = {
+		signal.SIGTERM: cleanup,
+		signal.SIGHUP: 'terminate',
+#		signal.SIGUSR1: reload_program_config,
+		}
 
-
-lock = FileLock('/tmp/hbc.pid')
-lock.acquire()
-
-up=1
-signal.signal(signal.SIGTERM, handler)
-signal.signal(signal.SIGHUP, handler)
-
-
-
-try:
+	context.files_preserve = [sock, sock.fileno()]
+	with context:
+		syslog.syslog('starting heartbeat to %s' %  ','.join(hb_hosts))
+		up = True
+		try:
 			process()
-except:
-	data='hbc died:\n'+traceback.format_exc()
-	open("/tmp/hbc.log","a").write(data)
-	sys.exit(1)
+		except:
+			syslogtrace('process')
+			cleanup(0, None)
 
-up=0
-msg="shutdown=1;name=%s;acks=%s" % (iam, ackcount)
-for hb_host in hb_hosts:
-	if verbose: print "hbc: sock.send('%s', (%s, %s))" % (msg, hb_host, hb_port)
-	socksend(msg, (hb_host, hb_port))
-time.sleep(1)
-sock.close()
-lock.release()