#!/usr/bin/env python # $Id: hbd,v 1.38 2013/07/14 02:25:05 andreas Exp $ # Wait for heartbeat messages and act on them (or their absence) # VER = 1.36 import time, os, string, sys, socket, atexit, select, SocketServer, getopt, signal, cPickle, smtplib, traceback, urllib from subprocess import Popen, STDOUT, PIPE False = 0 True = 1 LOGFILE = "/home/andreas/public_html/messages/andreas" PICKFILE = "/var/tmp/hbd.pick" AEMAIL = ["andreas@wrede.ca"] NAME = "heatbeat" SMTPSERVER = "localhost" hosts = {} htab = {} msgs = [] num = 0 upcount = 0 PORT = 50003 TPORT = 50004 THOST = "" DEBUG = False verbose = False INTERVAL = 10 GRACE = 2 visual = 0 os.environ['TZ'] = 'EST5EDT' stdscr = None win = None msgw = None msgwB = None msgwHeight = 10 def handler(signum, frame): global up, sig if up == 0: return sig = signum if verbose: print "signal: %s up: %d" % (sig, up) up = 0 # sys.exit(0) def shortname(name): r = string.split(name, '.') return r[0] class NullDevice: def write(self, s): pass class Host: up = "up" down = "down" overdue = "overdue" def __init__(self, name, addr): global num self.name = shortname(name) self.addr = addr self.num = num self.lastbeat = time.time() self.upcount = 0 self.state = Host.up self.state = "up" self.statetime = self.lastbeat self.interval = 0 self.doesack = -1 self.cmds = [] num += 1 def getstate(self): return self.state def dispstate(self): if self.state in ["down", "overdue"]: state = "%s" % self.state else: state = "%s" % self.state if self.doesack != -1: return "%s(%s)" % (state, self.doesack) return state # set new state, return number of secs in previous state def newstate(self, state, when=0): self.state = state now = time.time()-when s = now-self.statetime self.statetime = now if visual: displaystatetime(self.name) return s def email(s, msg): ret = "OK" toaddrs = AEMAIL fromaddr = "aew.heartbeat@wrede.ca" subj = "Info from %s: %s" % (NAME, s) date = time.strftime("%a, %d %b %Y %H:%M:%S %z", time.localtime()) body = "To: %s\nFrom: %s\nSubject: %s\nDate: %s\n\n%s" % (toaddrs[0], fromaddr, subj, date, msg) try: server = smtplib.SMTP(SMTPSERVER) if DEBUG: server.set_debuglevel(1) server.sendmail(fromaddr, toaddrs, body) except smtplib.SMTPRecipientsRefused, errs: log("cannot send email: %s\n" % (errs)) ret = "Fail" except: print("smtp error: "+traceback.format_exc()) saveandrestart() try: server.quit() except: pass return ret # # nsupdate: set the DNS A record for a fqdn # return: None if ok, else error text def nsupdate(hostname, newip): D = {} D['domain'] = 'dy.wapanafa.org' D['fqdn'] = '%s.dy.wapanafa.org' % hostname D['dnsttl'] = '5' D['newip'] = newip D['ts'] = time.strftime('%Y-%m-%d.%H:%M:%S', time.gmtime()) nsup = """update delete %(fqdn)s A update add %(fqdn)s %(dnsttl)s A %(newip)s update delete %(fqdn)s TXT update add %(fqdn)s %(dnsttl)s TXT "Created: %(ts)s" send answer """ % D # log("DBG: nsup %s" % nsup) cmd = ["/usr/bin/nsupdate", "-k", "/etc/dhcpc/K%(domain)s.+157+00000." % D, "-v"] # log("DBG: cmd %s" % cmd) try: p = Popen(cmd, shell=False, bufsize=1, stdin=PIPE, stdout=PIPE, stderr=STDOUT) except OSError, e: return "nsupdate: execution failed: %s" % e except: return "nsupdate: some error occured" (output, err) = p.communicate(nsup) if output.find('status: NOERROR') >= 0: return None return output # # def dur(sec): sec = int(sec) h = sec / 3600 m = (sec - h * 3600) / 60 s = (sec - h * 3600) % 60 if h > 0: return "%d:%02d:%02d" % (h, m, s) if m > 0: return "%d:%02d" % (m, s) return "0:%02d" % s # # def addhost(name, addr): sname = shortname(name) if sname in hosts: # was: hosts.has_key(sname): del htab[hosts[sname].addr] hosts[sname].addr = addr if visual: displayaddr(sname) htab[addr] = sname m = "%s, changed address to %s" % (sname, addr) log(m) else: hosts[sname] = Host(sname, addr) s = hosts.keys() s.sort() x = 0 for n in s: hosts[n].num = x x += 1 htab[addr] = sname if visual: display() # def on_exit(): if visual: exitcurses() logf.close() print "exit" def initlog(logfile): return open(logfile, "a") # # def initwin(): global win, msgw, msgwB, msgwHeight maxY, maxX = stdscr.getmaxyx() begin_x = 0 begin_y = 2 height = len(htab)+2 if DEBUG: log("initwin called with %d" % height) win = curses.newwin(height, maxX, begin_y, begin_x) a = win.border(0, 0, 0, 0, 0, 0, curses.ACS_LTEE, curses.ACS_RTEE) msgwB = curses.newwin(0, 0, height+1, begin_x) msgwB.border(0, 0, 0, 0, curses.ACS_LTEE, curses.ACS_RTEE) msgwHeight = maxY-height-3 msgw = curses.newwin(msgwHeight, maxX-2, height+2, begin_x+1) msgw.setscrreg(0, msgwHeight-1) msgw.scrollok(1) stdscr.addstr(0, 0, "hbd Version 1.0", curses.A_BOLD) stdscr.refresh() msgwB.refresh() # def checkoverdue(): for h in hosts.keys(): if hosts[h].state == Host.down: continue timeout = hosts[h].interval+grace if hosts[h].state == Host.up and now-hosts[h].lastbeat > timeout: m = "%s is overdue" % h if h in watchhosts: email("overdue", m) 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) for h in hosts.keys(): d = hosts[h].getstate() attr = 0 if verbose and hosts[h].state != Host.down: d = dur(now-hosts[h].lastbeat) if hosts[h].state == Host.overdue: attr = curses.A_BOLD win.addstr(hosts[h].num+1, 25, "%8s" % d, attr) win.refresh() stdscr.refresh() # # def displaystatetime(h, refresh=1): win.addstr(hosts[h].num+1, 60, "%-17s" % time.strftime("%Y-%m-%d %H:%M:%S", time.localtime(hosts[h].statetime))) if refresh: win.refresh() # # def displayaddr(h, refresh=1): win.addstr(hosts[h].num+1, 35, "%-16s" % hosts[h].addr) if refresh: win.refresh() # # def displaybody(): for h in hosts.keys(): win.addstr(hosts[h].num+1, 1, "%-25s" % (h)) if hosts[h].addr is not None: displayaddr(h, 0) if hosts[h].statetime is not None: displaystatetime(h, 0) win.refresh() # # def displaymsgs(): global msgw, msgs y = 0 for m in msgs[len(msgs)-msgwHeight:]: msgw.addstr(y, 0, m) y += 1 msgw.refresh() # # def display(): if visual: initwin() displaytime() displaybody() displaymsgs() def log(m, service="heartbeat"): msg = time.strftime("%b %d %H:%M:%S", time.localtime(time.time()))+": "+m+"\n" msgs.append(msg) if logfmt == "msg": m2 = "%d|%s|%s\n" % (now, service, m) else: m2 = msg logf.write(m2) logf.flush() if msgw is not None: y, x = msgw.getyx() # if y > = msgwHeight-1: # msgw.scroll() msgw.addstr(msg) msgw.clrtoeol() msgw.refresh() pickleit() # # def fromaddr(name, addr, boot, interval, acks): global htab if not name in hosts: # was: hosts.has_key(name): addhost(name, addr) host = hosts[name] host.doesack = acks if host.addr != addr: if host.addr in htab: # was: htab.has_key(host.addr): del htab[host.addr] host.addr = addr htab[addr] = name m = "%s changed address to %s" % (host.name, addr) if name in dyndnshosts: err = nsupdate(name, addr) if err: m += ", DNS failed: %s" % err else: m += ", DNS updated." log(m) if name in watchhosts: email("address change", m) host.lastbeat = now if host.getstate() != Host.up and interval > 0: lasts = host.state d = host.newstate(Host.up) m = "%s, back after being %s for %s" % (host.name, lasts, dur(d)) log(m) if name in watchhosts: email("back", name) host.upcount += 1 # # def readsock(): global htab, win data, addr = sock.recvfrom(1024) pairs = string.split(data, ';') boot = 0 shutdown = 0 name = "unknown" service = "unknown" msg = None interval = 0 deltaT = 0.0 acks = -1 for pair in pairs: l = string.split(pair, "=") key = l[0] if len(l) != 2: val = "0" else: val = l[1] if key == 'boot': boot += 1 elif key == 'shutdown': shutdown += 1 elif key == 'interval': interval = int(val) elif key == 'name': name = shortname(val) elif key == 'msg': msg = val elif key == 'service': service = val elif key == 'time': try: deltaT = now-float(val) except: pass elif key == 'acks': try: acks = int(val) except: acks = -1 if boot: if acks == -1: a = "(%s)" % acks else: a = "" m = "%s booted, deltaT %0.2g sec %s" % (name, deltaT, a) log(m) if name in watchhosts: email("booted", m) if msg: m = "%s msg: %s" % (name, msg) log(m, service=service) if name in watchhosts: email("msg", m) fromaddr(name, addr[0], boot, interval, acks) if shutdown: m = "%s shutdown" % name log(m) if name in watchhosts: email("shutdown", m) try: hosts[name].newstate(Host.down) except: pass if interval > 0: try: hosts[name].interval = interval except: pass rmsg="ACK" if len(hosts[name].cmds) then rmsg=hosts[name].cmds[0] email("%s cmd exec" % name, "command '%s' initiated" % hosts[name].cmds[0]) del hosts[name].cmds[0] try: sock.sendto(rmsg, addr) except: pass # # # def initcurses(): global stdscr stdscr = curses.initscr() curses.noecho() curses.cbreak() stdscr.keypad(1) if DEBUG: sys.stderr.write("curses init done: %s\n" % stdscr) def exitcurses(): curses.nocbreak() stdscr.keypad(0) curses.echo() curses.endwin() # # class HtmlHandler(SocketServer.BaseRequestHandler): def handle(self): f = self.request.makefile() uri = '/unknown' while 1: line = string.strip(f.readline()) if len(line) == 0: break r = line.split() if r[0] == "GET": uri = r[1] html = r[2] upar=string.split(uri,"?") if len(upar) == 1: uarg=[] else: uargs=string.split(upar[1],"&") code = 404 cause = "Not Found" if uri == "/": code = 200 cause = "OK" elif upar[0] == "/c": # /c?h=melschserver&c=sudo%20ls uname="" ucmd="" if uarg[0][:2] == "h=": uname=uarg[0][2:] if uarg[1][:2] == "c=": ucmd=uarg[1][2:] if ucmd != "" and uname != "" and hosts.has_key(uname): hosts[uname].cmds.append(urllib.unquote(ucmd)) code = 200 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: 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") self.request.send("Content-Type: text/html; charset = ISO-8859-1\r\n\r\n") res = [] if code != 200: res.append('') res.append('
') res.append('') res.append('The requested URL %s was not found on this server.
' % uri) res.append('| Host | State | IP Addr | Last change |
|---|---|---|---|
| %-24s | %-7s | %-16s | %-17s |