pingmon.py

import rrdtool
import os
import sys
import subprocess
import re
import time
import datetime

class GraphPing(object):
    def __init__(self, target, reset=False):
        self.filename_rrd = os.path.join('/tmp', 'ping-%s.rrd' % \
            target)
        self.filename_graph = os.path.join('/tmp', 'ping-%s.png' % \
            target)

        self.target = target
        self.step = 30
        self.heartbeat = self.step * 2

        # 7 days of logs for the step of 2 by min
        self.steps_db = 2 * (7 * 24 * 60 * 60)

        if not os.path.isfile(self.filename_rrd) or reset:
            self.create()

    def create(self):
        rrdtool.create(self.filename_rrd,
            '--step', str(self.step),
            '--start', 'N',
            'DS:pl:GAUGE:%d:0:100' % (self.heartbeat),
            'DS:rtt:GAUGE:%d:0:10000000' % (self.heartbeat),
            'RRA:AVERAGE:0.5:1:%d' % (self.steps_db),
            'RRA:MAX:0.5:1:%d' % (self.steps_db))

    def update(self, pl, rtt):
        rrdtool.update(self.filename_rrd,
            'N:%s:%s' % (pl, rtt))

    def graph(self):
        rrdtool.graph(self.filename_graph,
            '-a', 'PNG',
            '--title', 'ping %s' % (self.target),
            '--vertical-label', 'RTT (ms)',
            '-h', '225', '-w', '600',
            '--start', '-3600',
            '--lower-limit', '0',
            '--watermark', datetime.datetime.now().isoformat(),
            #'--slope-mode',
            #'--alt-y-grid',
            #'--rigid',
            '--x-grid', 'MINUTE:5:HOUR:1:MINUTE:30:0:%R',
            'DEF:roundtrip=%s:rtt:AVERAGE' % (self.filename_rrd),
            'DEF:packetloss=%s:pl:AVERAGE' % (self.filename_rrd),
            'CDEF:PLNone=packetloss,0,0,LIMIT,UN,UNKN,INF,IF',
            'CDEF:PL10=packetloss,1,10,LIMIT,UN,UNKN,INF,IF',
            'CDEF:PL25=packetloss,10,25,LIMIT,UN,UNKN,INF,IF',
            'CDEF:PL50=packetloss,25,50,LIMIT,UN,UNKN,INF,IF',
            'CDEF:PL75=packetloss,50,75,LIMIT,UN,UNKN,INF,IF',
            'CDEF:PL100=packetloss,75,100,LIMIT,UN,UNKN,INF,IF',
            'AREA:roundtrip#0000ff:RTT (ms)',
            'GPRINT:roundtrip:LAST:Cur\: %5.2lf',
            'GPRINT:roundtrip:AVERAGE:Avg\: %5.2lf',
            'GPRINT:roundtrip:MAX:Max\: %5.2lf',
            r'GPRINT:roundtrip:MIN:Min\: %5.2lf\n',
            'COMMENT:pkt loss (%)',
            'AREA:PLNone#ffffff:0-1%:STACK',
            'AREA:PL10#ffff00:1-10%:STACK',
            'AREA:PL25#ffcc00:10-25%:STACK',
            'AREA:PL50#ff8000:25-50%:STACK',
            'AREA:PL75#ff0000:50-75%:STACK',
            'AREA:PL100#bf0000:75-100%:STACK')

    def ping(self, count, maxtime):
        params = [ "ping", 
            "-c", str(count), 
            "-w", str(maxtime), 
            self.target]

        p = subprocess.Popen(params, 
            stdout=subprocess.PIPE, 
            stderr=subprocess.PIPE)
        out, err = p.communicate()

        # get RTT avg
        m = re.search('(\d+.\d+)/(\d+.\d+)/(\d+.\d+)', out)
        rtt = "0"
        if m: rtt = m.group(2)

        # get packet loss
        m = re.search('(\d+)% packet loss', out)
        pl = "100"
        if m: pl = m.group(1)
        return rtt, pl

    def run(self):
        while True:
            start = time.time()

            rtt, pl = self.ping(self.step, self.step)

            print "%s, %s, %s, %s" % \
                (datetime.datetime.now().isoformat(), 
                self.target, rtt, pl)
            self.update(pl, rtt)
            self.graph()

            stop = time.time()
            remaning = self.step - (stop - start)
            if remaning > 0:
                time.sleep(remaning)

if __name__ == "__main__":
    graph = GraphPing(sys.argv[1], 
        reset=False)
    graph.run()