major changes to both hbc and hbd

hbc

@@ -1,9 +1,13 @@
 #!/usr/bin/env python
-# $Id: hbc,v 1.2 2005/07/15 14:25:06 andreas Exp $
+# $Id: hbc,v 1.3 2005/07/19 20:31:05 andreas Exp $
 import sys, time, socket, os, signal, getopt, string
 
 PORT=50003
-INTERVAL=20
+INTERVAL=10
+
+class NullDevice:
+    def write(self, s):
+        pass
 
 
 def handler(signum, frame):
@@ -12,9 +16,10 @@ def handler(signum, frame):
 		return
 	sys.exit(0)
 
-helpflag=0
+msgonly=False
-verbose=0
+helpflag=False
-daemon=0
+verbose=False
+daemon=False
 optlist=[]  
 args=[]
 msgboot=[]
@@ -24,7 +29,7 @@ configfile="%s/.hbrc" % home
 try:
     optlist, args = getopt.getopt(sys.argv[1:], 'bc:dhm:v')
 except:
-    helpflag=1
+    helpflag=True
 
 for o,a in optlist:
 	if o == '-b':
@@ -32,24 +37,37 @@ for o,a in optlist:
 	elif o == '-c':
 		configfile=a
 	elif o == '-d':
-		daemon=1
+		daemon=True
 	elif o == '-h':
-		helpflag=1
+		helpflag=True
 	elif o == '-m':
 		msgboot.append("service=%s" % "service")
+		a.replace(';',':')
 		msgboot.append("msg=%s" % a)
+		msgonly=True
 	elif o == '-v':
-		verbose+=1
+		verbose=True
 
 
 if helpflag:
 	print "hbc HeartBeatClient"
 	print "usage: hbc [-bdhv] [-c configfile] [-m msg][host1 [..]]"
 	print
+	print "	-b	indicate machine boot"
+	print "	-c configfile"
+	print "	-d daemonize"
+	print "	-h this help"
+	print "	-m send a message"
+	print "	-v verbose"
+	print
 	print """ config file can contain 
 hb_hosts=('host1', 'host2', ..._
 hb_port=50003
 interval=20
+logfile=...
+logfmt={|test|msg}
+grace=SECONDS
+reportstrict={True|False}
 """
 
 	sys.exit(1)
@@ -76,6 +94,8 @@ if f:
 		r=l[:-1].split('=')
 		if r[0] == 'hb_hosts':
 			hb_hosts=eval(r[1])
+			if verbose:
+				print "notice:  cfg hb_hosts: %s" % hb_hosts
 		elif r[0] == 'interval':
 			interval=eval(r[1])
 		elif r[0] == 'hb_port':
@@ -99,7 +119,7 @@ if verbose:
 	print "notice: interval: %s" % interval
 	print "notice: iam: %s" % iam
 
-if daemon:
+if not msgonly:
 	msgboot.append("interval=%s" % interval)
 
 if len(msgboot) > 0:
@@ -118,15 +138,27 @@ if len(msgboot) > 0:
 		else:
 			break
 	
+if msgonly:
+		sys.exit(0)
 
-if not daemon:
+if daemon:
-	sys.exit(0)
+	pid=os.fork()
+	if pid > 0:
+		if verbose:
+			print "daemoinizing... pid=%d" % pid
+		sys.exit(0)
+
+
+	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) 
 
-pid=os.fork()
-if pid > 0:
-	if verbose:
-		print "daemoinizing... pid=%d" % pid
-	sys.exit(0)
 
 msg="interval=%s;name=%s" % (interval, iam)
 up=1
@@ -152,3 +184,5 @@ for hb_host in hb_hosts:
 	sock.sendto(msg, (hb_host, hb_port))
 time.sleep(1)
 sock.close()
+
+

hbd

@@ -1,23 +1,23 @@
 #!/usr/bin/env python
-# $Id: hbd,v 1.1 2005/07/14 19:26:01 andreas Exp $
+# $Id: hbd,v 1.2 2005/07/19 20:31:05 andreas Exp $
 # Wait for heartbeat messages and act on them (or their absence)
 #
-import time, os, string, sys, socket, curses, atexit, select, SocketServer
+import time, os, string, sys, socket, curses, atexit, select, SocketServer, getopt
 
 
-LOGF="/home/andreas/public_html/messages/andreas"
+LOGFILE="/home/andreas/public_html/messages/andreas"
 
 hosts={}
 num=0
 upcount=0
 PORT=50003
 TPORT=50004
-THOST="10.99.1.4"
+THOST=""
 DEBUG=False
 verbose=False
 
 INTERVAL=10
-GRACE=10
+GRACE=2
 
 visual=0
 
@@ -29,11 +29,14 @@ msgw=None
 msgwB=None
 msgwHeight=10
 
+class NullDevice:
+    def write(self, s):
+        pass
+
 class Host:
   up="up"
   down="down"
   overdue="overdue"
-  OVERDUE="OVERDUE"
 
   def __init__(self, name, addr):
 	global num
@@ -46,12 +49,13 @@ class Host:
 	self.uppercent="n/a"
 	self.state="up"
 	self.statetime=self.lastbeat
-	self.interval=INTERVAL
+	self.interval=0
 	num+=1
 
   def getstate(self):
 	return self.state
 
+  # set new state, return number of secs in previous state
   def newstate(self, state):
 	self.state=state
 	now=time.time()
@@ -109,9 +113,8 @@ def on_exit():
 	print "exit"
 
 
-def initlog():
+def initlog(logfile):
-	global logf
+	return open(logfile,"a")
-	logf=open(LOGF,"a")
 #
 #
 #
@@ -135,7 +138,7 @@ def initwin():
 	msgw.setscrreg(0, msgwHeight-1)
 	msgw.scrollok(1)
 
-	stdscr.addstr(0,0, "WatchArnsberg Version 1.0", curses.A_BOLD)
+	stdscr.addstr(0,0, "hbd Version 1.0", curses.A_BOLD)
 	stdscr.refresh()
 	msgwB.refresh()
 #
@@ -146,11 +149,13 @@ def checkoverdue():
 	for h in hosts:
 		if hosts[h].state == Host.down:
 			continue
-		if hosts[h].state == Host.up and now-hosts[h].lastbeat > hosts[h].interval+GRACE:
+		if reportstrict:
+			timeout=hosts[h].interval+grace
+		else:
+			timeout=hosts[h].interval*5+grace
+		if hosts[h].state == Host.up and now-hosts[h].lastbeat > timeout:
+			if reportstrict: log("%s is overdue" % h)
 			hosts[h].newstate(Host.overdue)
-		elif hosts[h].state == Host.overdue and now-hosts[h].lastbeat > hosts[h].interval*5+GRACE:
-			log("%s is overdue" % h)
-			hosts[h].newstate(Host.OVERDUE)
 
 #
 #
@@ -165,7 +170,7 @@ def displaytime():
 		attr=0
 		if verbose and  hosts[h].state != Host.down:
 			d=dur(now-hosts[h].lastbeat)
-		if hosts[h].state == Host.OVERDUE:
+		if hosts[h].state == Host.overdue:
 			attr=curses.A_BOLD
 		win.addstr(hosts[h].num+1, 25, "%8s" %  d, attr)
 		win.addstr(hosts[h].num+1, 53,  "%3s" % hosts[h].uppercent )
@@ -223,7 +228,10 @@ def display():
 def log(m, service="heartbeat"):
 	msg=time.strftime("%b %d %H:%M:%S")+": "+m+"\n"
 	msgs.append(msg)
-	m2="%d|%s|%s\n" % (now, service, m)
+	if logfmt == "msg":
+		m2="%d|%s|%s\n" % (now, service, m)
+	else:
+		m2=msg
 	logf.write(m2)
 	logf.flush()
 	if msgw != None:
@@ -236,18 +244,17 @@ def log(m, service="heartbeat"):
 
 #
 #
-def fromaddr(name, addr, boot):
+def fromaddr(name, addr, boot, interval):
 	global htab
 
 	if not htab.has_key(addr):
 		addhost(name, addr)
 	host=hosts[htab[addr]]
 	host.lastbeat=now
-	if host.getstate() != Host.up:
+	if host.getstate() != Host.up and interval > 0:
 		lasts=host.state
 		d=host.newstate(Host.up)
-		if lasts != 'overdue':
+		log("%s, back after being %s for %s" % (host.name, lasts, dur(d)))
-			log("%s, back after being %s for %s" % (host.name, lasts, dur(d)))
 	host.upcount+=1
 
 #
@@ -261,7 +268,7 @@ def readsock():
 	shutdown=0
 	name="unknown"
 	msg=None
-	interval=INTERVAL
+	interval=0
 	for pair in pairs:
 		l=string.split(pair,"=")
 		key=l[0]
@@ -285,11 +292,12 @@ def readsock():
 		log("%s booted" % name)
 	if msg:
 		log("%s msg: %s" % (name, msg),service=service)
-	fromaddr(name, addr[0], boot)
+	fromaddr(name, addr[0], boot, interval)
 	if shutdown:
 		log("%s shutdown" % name)
 		hosts[name].newstate(Host.down)
-	hosts[name].interval=interval
+	if interval > 0:
+		hosts[name].interval=interval
 
 
 #
@@ -300,7 +308,7 @@ def updatestats():
 	upcount+=1
 	for h in hosts:
 		if upcount > 0:
-			hosts[h].uppercent="%3.0f" % ((hosts[h].upcount*hosts[h].interval*100.0)/(upcount*INTERVAL))
+			hosts[h].uppercent="%3.0f" % ((hosts[h].upcount*hosts[h].interval*100.0)/(upcount*interval))
 #
 #
 #
@@ -339,7 +347,7 @@ class HtmlHandler(SocketServer.BaseRequestHandler):
 		cause="OK"
 	self.request.send("HTTP/1.0 %s %s\r\n" % (code, cause))
 	self.request.send("Date: %s\r\n" % time.strftime("%a, %d %b %Y %H:%M:%S GMT",time.gmtime(now)))
-	self.request.send("Server: WatchArnsberg\r\n")
+	self.request.send("Server: hbd\r\n")
 	self.request.send("Last-Modified: %s\r\n" % time.strftime("%a, %d %b %Y %H:%M:%S GMT",time.gmtime(now)))
 	self.request.send("Accept-Ranges: bytes\r\n")
 	self.request.send("Connection: close\r\n")
@@ -353,7 +361,7 @@ class HtmlHandler(SocketServer.BaseRequestHandler):
 		res.append('<h1>%s</h1>' % (cause))
 		res.append('<p>The requested URL %s was not found on this server.</p>' % uri)
 		res.append('<hr>')
-		res.append('<address>WatchArnsberg (Unix) Server at somewhere.planix.com Port %d</address>' % TPORT)
+		res.append('<address>hbd (Unix) Server at %s Port %d</address>' % (hbd_host, hbd_port))
 		res.append('</body></html>')
 		
 	else:
@@ -383,11 +391,106 @@ class HtmlHandler(SocketServer.BaseRequestHandler):
 # Main
 #
 
+helpflag=False
+forground=False
+optlist=[]  
+args=[]
+home=os.environ['HOME']
+configfile="%s/.hbrc" % home
+try:
+	optlist, args = getopt.getopt(sys.argv[1:], 'c:dfhv')
+except:
+    helpflag=True
 
-initlog()
+for o,a in optlist:
+	if o == '-c':
+		configfile=a
+	if o == '-d':
+		visual=True
+	elif o == '-f':
+		forground=True
+	elif o == '-h':
+		helpflag=True
+	elif o == '-v':
+		verbose=True
+
+
+if helpflag:
+	print "hbc HeartBeatDaemon"
+	print "usage: hbd [-dfhv] [-c configfile]"
+	print
+	print "	-c configfile"
+	print "	-d display"
+	print "	-f run in foreground"
+	print "	-h this help"
+	print "	-v verbose"
+	print
+	print """ config file can contain 
+logfile=/var/log/heartbeat.log
+logfmt=[text|msg]
+hb_port=50003
+interval=20
+hbd_port=50004
+hbd_host=www.domain.com
+grace=1
+"""
+
+	sys.exit(1)
+
+if visual:
+	forground=True
+#
+# set defaults
+
+hb_port=PORT
+hbd_host=THOST
+hbd_port=TPORT
+logfile=LOGFILE
+logfmt="text"
+interval=INTERVAL
+grace=GRACE
+reportstrict=False
+
+try:
+	f=open(configfile,"r")	
+	if verbose: print "notice: using config file %s" % configfile
+except:
+	print "warning: running without conifig file: %s" % configfile
+	f=None
+
+if f:
+	while 1:
+		l=f.readline()
+		if len(l) == 0:
+			break
+		r=l[:-1].split('=')
+		if r[0] == 'interval':
+			interval=eval(r[1])
+		elif r[0] == 'grace':
+			grace=eval(r[1])
+		elif r[0] == 'hbd_port':
+			hbd_port=eval(r[1])
+		elif r[0] == 'hbd_host':
+			hbd_host=r[1]
+		elif r[0] == 'hb_port':
+			hb_port=eval(r[1])
+		elif r[0] == 'logfile':
+			logfile=r[1]
+		elif r[0] == 'logfmt':
+			logfmt=r[1]
+		elif r[0] == 'reportstrict':
+			reportstrict=r[1]
+	f.close()
+
+if len(args) != 0:
+	print "error: args"
+	sys.exit(1)
+
+
+logf=initlog(logfile)
 
 now=time.time()
-startsec=int(now) % INTERVAL
+startsec=int(now) % interval
 
 htab={}
 if visual:
@@ -401,12 +504,30 @@ if DEBUG: log("Starting")
 ilist=[]
 
 sock=socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
-sock.bind(("",PORT))
+sock.bind(("",hb_port))
 ilist.append(sock)
 
-serv=SocketServer.TCPServer((THOST,TPORT),HtmlHandler)
+serv=SocketServer.TCPServer((hbd_host,hbd_port),HtmlHandler)
 ilist.append(serv.fileno())
 
+if not forground:
+	pid=os.fork()
+	if pid > 0:
+		if verbose:
+			print "daemoinizing... pid=%d" % pid
+		sys.exit(0)
+
+	verbose=False
+	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) 
+
 next=int(now)+1
 sleep=next - now
 while 1:
@@ -441,7 +562,7 @@ while 1:
 			serv.handle_request()
 	if now >= next:
 		next=now+1
-		if int(now) % INTERVAL == startsec:
+		if int(now) % interval == startsec:
 			updatestats()
 		checkoverdue()
 		if visual: