ping for datadog

ping.py

#!/usr/bin/env python

import subprocess
import re
import time
from checks.network_checks import NetworkCheck, Status


class PingCheck(NetworkCheck):
    def __init__(self, name, init_config, agentConfig, instances):
        NetworkCheck.__init__(self, name, init_config, agentConfig, instances)
        for instance in instances:
            if not instance.get('isp', None):
                raise Exception("All instances should have a 'isp' parameter")
            # for initialize loss cnt
            self.__simple_increment(instance, 'loss_cnt', 0)

    def _check(self, instance):
        start = time.time()
        ping_timeout = int(self.init_config.get('ping_timeout', 4))
        max_retry = int(self.init_config.get('check_interval', 15)) / ping_timeout
        p = Ping(instance['addr'], ping_timeout)
        failure_count = 0
        for i in range(self.init_config.get('check_times', 1)):
            try:
                p.run()
                self.__simple_histogram(instance, 'rtt', p._rtt)
                time.sleep(self.init_config.get('ping_interval', 0.1))
                ret = (Status.UP, "UP")
            except:
                self.__simple_increment(instance, 'loss_cnt')
                failure_count += 1
                ret = (Status.DOWN, "DOWN")
                if failure_count >= max_retry:
                    break
            finally:
                self.__simple_increment(instance, 'total_cnt')
        elapsed_time = time.time() - start
        self.log.info("name:%s, elapsed_time:%s[sec]"
                      % (instance['name'], round(elapsed_time, 2)))
        return ret

    def __simple_increment(self, instance, category, value=1):
        self.increment(
               '%s.%s' % (self.init_config.get('basename', 'ping'),
                          category),
               value,
               tags=['isp:%s' % instance['isp'],
                     'locate:%s' % instance['name']]
        )

    def __simple_histogram(self, instance, category, value):
        self.histogram(
               '%s.%s' % (self.init_config.get('basename', 'ping'),
                          category),
               value,
               tags=['isp:%s' % instance['isp'],
                     'locate:%s' % instance['name']]
        )

    def report_as_service_check(self, sc_name, status, instance, msg=None):
        pass

    def _create_status_event(self, sc_name, status, msg, instance):
        # TODO 5.3 remove that
        pass


class Ping(object):
    def __init__(self, host, timeout):
        self._host = host
        self._timeout = timeout

    def run(self):
        ping = subprocess.Popen(
                ["ping", "-c", "1", "-W", str(self._timeout), self._host],
                stdout=subprocess.PIPE,
                stderr=subprocess.PIPE
                )
        out, error = ping.communicate()
        if error:
            raise Exception('[NG]: ServerName->%s, Msg->"%s"'
                            % (self._host, error.rstrip()))
        else:
            try:
                self._ttl = int(re.search("(?<=ttl=)[\d\.]+", out).group())
                self._rtt = float(re.search("(?<=time=)[\d\.]+", out).group())
            except AttributeError:
                raise Exception('[NG]: ServerName->%s, Msg->"%s"'
                                % (self._host, 'cannot connect'))