always pickle current state when a log msg is set

drop strickt reporting - it's useless

hbd

@@ -1,5 +1,5 @@
 #!/usr/bin/env python
-# $Id: hbd,v 1.33 2012/09/22 17:51:45 andreas Exp $
+# $Id: hbd,v 1.34 2012/09/22 19:17:53 andreas Exp $
 # Wait for heartbeat messages and act on them (or their absence)
 #
 import time, os, string, sys, socket, atexit, select, SocketServer, getopt, signal, cPickle, smtplib, traceback
@@ -7,7 +7,7 @@ import time, os, string, sys, socket, atexit, select, SocketServer, getopt, sign
 False=0
 True=1
 LOGFILE="/home/andreas/public_html/messages/andreas"
-PICKFILE="/tmp/hbd.pick"
+PICKFILE="/var/tmp/hbd.pick"
 AEMAIL=["andreas@wrede.ca"]
 NAME="heatbeat"
 SMTPSERVER="localhost"
@@ -208,22 +208,17 @@ def checkoverdue():
 	for h in hosts.keys():
 		if hosts[h].state == Host.down:
 			continue
-		if reportstrict:
-			gr=grace
-		else:
-			gr=5*grace
-		timeout=hosts[h].interval+gr
+		timeout=hosts[h].interval+grace
 		if hosts[h].state == Host.up and now-hosts[h].lastbeat > timeout:
 			m="%s is overdue" % h
-			log(m)
 			if h in watchhosts:
 				email("overdue", m)
-			hosts[h].newstate(Host.overdue, gr)
+			hosts[h].newstate(Host.overdue, grace)
+			log(m)
 
 #
 #
 #
-#
 def displaytime():
 	maxY,maxX=stdscr.getmaxyx()
 	stdscr.addstr(0,maxX-8, time.strftime("%H:%M:%S", time.localtime(now)), curses.A_BOLD)
@@ -303,6 +298,7 @@ def log(m, service="heartbeat"):
 		msgw.addstr(msg)
 		msgw.clrtoeol()
 		msgw.refresh()
+	pickleit()
 
 #
 #
@@ -493,6 +489,10 @@ class HtmlHandler(SocketServer.BaseRequestHandler):
 def saveandrestart():
 	sock.close()
 	serv.socket.close()
+	log("restarting")
+	os.execv(sys.argv[0],[sys.argv[0]]+cmdargs)
+
+def pickleit():
 	pickf=open(PICKFILE, 'w')
 	pick=cPickle.Pickler(pickf)
 	pick.dump(hosts)
@@ -500,7 +500,6 @@ def saveandrestart():
 	pick.dump(msgs)
 	pickf.close()
 	
-	os.execv(sys.argv[0],[sys.argv[0]]+cmdargs)
 	
 #
 # Main
@@ -508,7 +507,6 @@ def saveandrestart():
 
 helpflag=False
 forground=False
-restart=None
 optlist=[]  
 args=[]
 home=os.environ['HOME']
@@ -554,7 +552,7 @@ hb_port=50003
 interval=20
 hbd_port=50004
 hbd_host=www.domain.com
-grace=1
+grace=2
 """
 
 	sys.exit(1)
@@ -571,7 +569,6 @@ logfile=LOGFILE
 logfmt="text"
 interval=INTERVAL
 grace=GRACE
-reportstrict=False
 watchhosts=[]
 drophosts=[]
 
@@ -603,8 +600,6 @@ if f:
 			logfile=r[1]
 		elif r[0] == 'logfmt':
 			logfmt=r[1]
-		elif r[0] == 'reportstrict':
-			reportstrict=r[1] in ["True","true","TRUE","1"]
 		elif r[0] == 'watchhosts':
 			watchhosts=eval(r[1])
 		elif r[0] == 'drophosts':
@@ -622,11 +617,13 @@ logf=initlog(logfile)
 if os.path.exists(PICKFILE):
 	pickf=open(PICKFILE, 'r')
 	pick=cPickle.Unpickler(pickf)
+	try:
 		hosts=pick.load()
 		htab=pick.load()
 		msgs=pick.load()
 		pickf.close()
-#	os.unlink(PICKFILE)
+	except:
+		os.unlink(PICKFILE)
 	for h in drophosts:
 		if hosts.has_key(h):
 			del hosts[h]
@@ -641,9 +638,7 @@ if visual:
 	display()
 	stdscr.nodelay(1)
 
-if verbose: 
-	if restart: log("Restarting")
-	else: log("Starting")
+log("Starting")
 atexit.register(on_exit)
 
 ilist=[]
@@ -658,7 +653,7 @@ ilist.append(sock)
 serv=SocketServer.TCPServer((hbd_host,hbd_port),HtmlHandler)
 ilist.append(serv.fileno())
 
-if not forground and not restart:
+if not forground:
 	pid=os.fork()
 	if pid > 0:
 		if verbose: print "daemoinizing... pid=%d" % pid
@@ -680,7 +675,7 @@ sig=0
 signal.signal(signal.SIGTERM, handler)
 signal.signal(signal.SIGHUP, handler)
 
-next=int(now)+1
+next=int(now)+15	# 15  seconds time to settle after (re-)start
 sleep=next - now
 while up:
 	if visual: