Merge branch 'master' of git.wrede.ca:andreas/heartbeat

sequence error

hbc

@@ -1,5 +1,9 @@
 #!/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 
 import sys
 import time
 import socket
@@ -9,12 +13,12 @@ import getopt
 import string
 import select
 import errno
+import traceback
+from  lockfile import FileLock
 
 PORT=50003
 INTERVAL=10
-False=0
-True=1
-DBG = False
+DBG = True
 
 sock=None
 
@@ -23,60 +27,6 @@ class NullDevice:
         pass
 
 
-
-class Flock:
-	def __init__(self, lock_file):
-		self.lock_file = lock_file
-		self.fd = None
-		self.opid = None
-
-	def lock(self):
-		while 1:
-			self.fd = None
-			try:
-				self.fd = os.open(self.lock_file, os.O_CREAT | os.O_EXCL | os.O_RDWR)
-			except OSError, e:
-				if e.errno != errno.EEXIST:
-					raise
-			except:
-				raise
-		
-			if not self.fd:
-				if self.oproc():
-					if DBG: print "process is alive"
-					os.remove(self.lock_file)
-					continue
-				else:
-					if DBG: print "no pid process??"
-	
-			if self.fd:
-				f=os.fdopen(self.fd, 'w').write("%s" % os.getpid())
-			return self.fd
-	
-	
-	def unlock(self):
-		os.remove(self.lock_file)
-		self.fd=None
-		
-
-	def setopid(self):
-		try:
-			self.opid=open(self.lock_file).readline()
-		except:
-			pass
-
-
-	def oproc(self):
-		self.setopid()
-		if not self.opid:
-			return False
-		try:
-			os.kill(int(self.opid), 0)
-			return True
-		except:
-			pass
-		return False
-
 def handler(signum, frame):
 	global up
 	if up == 0:
@@ -110,7 +60,40 @@ def socksend(msg, tohost):
 	if sock == None:
 		sock=getsock(tohost[0])
 	sock.sendto(msg, tohost)
+	if verbose: print "msg %s sent" % msg
 
+def process():
+	ackcount=0
+	lastT=time.time()
+	while up:
+		sleep=(lastT+interval) - time.time() 
+		if verbose: print "sleep %s" % sleep
+		if sleep > 0:
+			try:
+				r=select.select([sock.fileno()],[],[],sleep)
+	#			time.sleep(interval)
+			except:
+				break
+			if verbose: print r
+			if sock.fileno() in r[0]:
+				data, addr = sock.recvfrom(1024)
+				if  data == "ACK":
+					ackcount+=1
+				else:
+					try:
+						os.system(data)
+					except:
+						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)
+				socksend(msg, (hb_host, hb_port))
+			except:
+				pass
+	
 
 msgonly=False
 helpflag=False
@@ -238,6 +221,7 @@ if len(msgboot) > 0:
 		else:
 			break
 	
+if verbose: print "msgboot done msgonly=%s" % msgonly
 if msgonly:
 		sys.exit(0)
 
@@ -252,59 +236,33 @@ if daemon:
 		sys.exit(0)
 
 
-	os.close(0)
-	os.close(1)
-	os.close(2)
-	sys.stdin.close()
-	sys.stdout = NullDevice()
-	sys.stderr = NullDevice()
+	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) 
 
 
-while True:
-	lock=Flock('/tmp/hbc.pid')
-	if lock.lock():
-		break
-	if not lock.oproc():
-		sys.exit(1)
-	os.kill(lock.opid,15)
-	time.sleep(1)
+lock = FileLock('/tmp/hbc.pid')
+lock.acquire()
 
 up=1
 signal.signal(signal.SIGTERM, handler)
 signal.signal(signal.SIGHUP, handler)
-ackcount=0
-lastT=time.time()
-while up:
-	sleep=(lastT+interval) - time.time() 
-	if verbose: print "sleep %s" % sleep
-	if sleep > 0:
-		try:
-			r=select.select([sock.fileno()],[],[],sleep)
-#			time.sleep(interval)
-		except:
-			break
-		if verbose: print r
-		if sock.fileno() in r[0]:
-			data, addr = sock.recvfrom(1024)
-			if  data == "ACK":
-				ackcount+=1
-			else:
-				try:
-					os.system(data)
-				except:
-					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)
-			socksend(msg, (hb_host, hb_port))
-		except:
-			pass
+
+
+
+try:
+	process()
+except:
+	data='hbc died:\n'+traceback.format_exc()
+	open("/tmp/hbc.log","a").write(data)
+	sys.exit(1)
 
 up=0
 msg="shutdown=1;name=%s;acks=%s" % (iam, ackcount)
@@ -313,4 +271,4 @@ for hb_host in hb_hosts:
 	socksend(msg, (hb_host, hb_port))
 time.sleep(1)
 sock.close()
-lock.unlock()
+lock.release()