def _process_ping(self, ping_url, ping_reply, auth, tags, pool_name, http_host):
if ping_reply is None:
ping_reply = 'pong'
sc_tags = ["ping_url:{0}".format(ping_url)] + tags
if http_host is not None:
sc_tags += ["http_host:{0}".format(http_host)]
try:
# TODO: adding the 'full' parameter gets you per-process detailed
# informations, which could be nice to parse and output as metrics
resp = requests.get(ping_url, auth=auth,
headers=headers(self.agentConfig, http_host=http_host))
resp.raise_for_status()
if ping_reply not in resp.text:
raise Exception("Received unexpected reply to ping {0}".format(resp.text))
except Exception as e:
self.log.error("Failed to ping FPM pool {0} on URL {1}."
"\nError {2}".format(pool_name, ping_url, e))
self.service_check(self.SERVICE_CHECK_NAME,
AgentCheck.CRITICAL, tags=sc_tags, message=str(e))
else:
self.service_check(self.SERVICE_CHECK_NAME, AgentCheck.OK, tags=sc_tags)
def check(self, instance):
if 'monitor_agent_url' not in instance:
raise Exception(
'Fluentd instance missing "monitor_agent_url" value.')
try:
url = instance.get('monitor_agent_url')
plugin_ids = instance.get('plugin_ids', [])
parsed_url = urlparse.urlparse(url)
monitor_agent_host = parsed_url.hostname
monitor_agent_port = parsed_url.port or 24220
service_check_tags = [
'fluentd_host:%s' % monitor_agent_host,
'fluentd_port:%s' % monitor_agent_port
]
r = requests.get(url, headers=headers(self.agentConfig))
r.raise_for_status()
status = r.json()
for p in status['plugins']:
for m in self.GAUGES:
if p.get(m) is None:
continue
if p.get('plugin_id') in plugin_ids:
self.gauge('fluentd.%s' % (m), p.get(m),
["plugin_id:%s" % p.get('plugin_id')])
except Exception, e:
msg = "No stats could be retrieved from %s : %s" % (url, str(e))
self.service_check(self.SERVICE_CHECK_NAME,
AgentCheck.CRITICAL,
tags=service_check_tags,
message=msg)
raise e
def _get_data(self, instance):
url = instance.get('nginx_status_url')
ssl_validation = instance.get('ssl_validation', True)
auth = None
if 'user' in instance and 'password' in instance:
auth = (instance['user'], instance['password'])
# Submit a service check for status page availability.
parsed_url = urlparse.urlparse(url)
nginx_host = parsed_url.hostname
nginx_port = parsed_url.port or 80
service_check_name = 'nginx.can_connect'
service_check_tags = ['host:%s' % nginx_host, 'port:%s' % nginx_port]
try:
self.log.debug(u"Querying URL: {0}".format(url))
r = requests.get(url, auth=auth, headers=headers(self.agentConfig),
verify=ssl_validation, timeout=self.default_integration_http_timeout)
r.raise_for_status()
except Exception:
self.service_check(service_check_name, AgentCheck.CRITICAL,
tags=service_check_tags)
raise
else:
self.service_check(service_check_name, AgentCheck.OK,
tags=service_check_tags)
body = r.content
resp_headers = r.headers
return body, resp_headers.get('content-type', 'text/plain')
def _get_data(self, instance):
url = instance.get('nginx_status_url')
ssl_validation = instance.get('ssl_validation', True)
auth = None
if 'user' in instance and 'password' in instance:
auth = (instance['user'], instance['password'])
# Submit a service check for status page availability.
parsed_url = urlparse.urlparse(url)
nginx_host = parsed_url.hostname
nginx_port = parsed_url.port or 80
service_check_name = 'nginx.can_connect'
service_check_tags = ['host:%s' % nginx_host, 'port:%s' % nginx_port]
try:
self.log.debug(u"Querying URL: {0}".format(url))
r = requests.get(url, auth=auth, headers=headers(self.agentConfig),
verify=ssl_validation, timeout=self.default_integration_http_timeout)
r.raise_for_status()
except Exception:
self.service_check(service_check_name, AgentCheck.CRITICAL,
tags=service_check_tags)
raise
else:
self.service_check(service_check_name, AgentCheck.OK,
tags=service_check_tags)
body = r.content
resp_headers = r.headers
return body, resp_headers.get('content-type', 'text/plain')
def check(self, instance):
if 'apache_status_url' not in instance:
self.log.warn("Missing 'apache_status_url' in Apache config")
return
tags = instance.get('tags', [])
try:
req = urllib2.Request(instance['apache_status_url'], None,
headers(self.agentConfig))
request = urllib2.urlopen(req)
response = request.read()
# Loop through and extract the numerical values
for line in response.split('\n'):
values = line.split(': ')
if len(values) == 2: # match
metric, value = values
metric_name = self.METRIC_TRANSLATION.get(metric, metric)
try:
if metric_name == 'apache.net.bytes':
self.gauge(metric_name, float(value) * 1024, tags=tags)
else:
self.gauge(metric_name, float(value), tags=tags)
except ValueError:
continue
except:
self.log.exception('Unable to get Apache status')
def _process_status(self, status_url, auth, tags):
data = {}
try:
resp = requests.get(status_url,
auth=auth,
headers=headers(self.agentConfig),
params={'json': True})
resp.raise_for_status()
data = resp.json()
except Exception as e:
self.log.error("Failed to get metrics from {0}.\nError {1}".format(
status_url, e))
raise
pool_name = data.get('pool', 'default')
metric_tags = tags + ["pool:{0}".format(pool_name)]
for key, mname in self.GAUGES.iteritems():
if key not in data:
self.log.warn(
"Gauge metric {0} is missing from FPM status".format(key))
continue
self.gauge(mname, int(data[key]), tags=metric_tags)
for key, mname in self.MONOTONIC_COUNTS.iteritems():
if key not in data:
self.log.warn(
"Counter metric {0} is missing from FPM status".format(
key))
continue
self.monotonic_count(mname, int(data[key]), tags=metric_tags)
return pool_name
def _get_metrics(self, url, tags):
req = urllib2.Request(url, None, headers(self.agentConfig))
request = urllib2.urlopen(req)
response = request.read()
# Thanks to http://hostingfu.com/files/nginx/nginxstats.py for this code
# Connections
parsed = re.search(r'Active connections:\s+(\d+)', response)
if parsed:
connections = int(parsed.group(1))
self.gauge("nginx.net.connections", connections, tags=tags)
# Requests per second
parsed = re.search(r'\s*(\d+)\s+(\d+)\s+(\d+)', response)
if parsed:
conn = int(parsed.group(1))
requests = int(parsed.group(3))
self.rate("nginx.net.conn_opened_per_s", conn, tags=tags)
self.rate("nginx.net.request_per_s", requests, tags=tags)
# Connection states, reading, writing or waiting for clients
parsed = re.search(r'Reading: (\d+)\s+Writing: (\d+)\s+Waiting: (\d+)', response)
if parsed:
reading, writing, waiting = map(int, parsed.groups())
self.gauge("nginx.net.reading", reading, tags=tags)
self.gauge("nginx.net.writing", writing, tags=tags)
self.gauge("nginx.net.waiting", waiting, tags=tags)
def check(self, instance):
if "monitor_agent_url" not in instance:
raise Exception('Fluentd instance missing "monitor_agent_url" value.')
try:
url = instance.get("monitor_agent_url")
plugin_ids = instance.get("plugin_ids", [])
parsed_url = urlparse.urlparse(url)
monitor_agent_host = parsed_url.hostname
monitor_agent_port = parsed_url.port or 24220
service_check_tags = ["fluentd_host:%s" % monitor_agent_host, "fluentd_port:%s" % monitor_agent_port]
req = urllib2.Request(url, None, headers(self.agentConfig))
res = urllib2.urlopen(req).read()
status = json.loads(res)
for p in status["plugins"]:
for m in self.GAUGES:
if p.get(m) is None:
continue
if p.get("plugin_id") in plugin_ids:
self.gauge("fluentd.%s" % (m), p.get(m), ["plugin_id:%s" % p.get("plugin_id")])
except Exception, e:
msg = "No stats could be retrieved from %s : %s" % (url, str(e))
self.service_check(self.SERVICE_CHECK_NAME, AgentCheck.CRITICAL, tags=service_check_tags, message=msg)
raise e
def check(self, instance):
if 'lighttpd_status_url' not in instance:
raise Exception("Missing 'lighttpd_status_url' in Lighttpd config")
tags = instance.get('tags', [])
req = urllib2.Request(instance['lighttpd_status_url'], None,
headers(self.agentConfig))
request = urllib2.urlopen(req)
response = request.read()
# Loop through and extract the numerical values
for line in response.split('\n'):
values = line.split(': ')
if len(values) == 2: # match
metric, value = values
try:
value = float(value)
except ValueError:
continue
# Special case: kBytes => bytes
if metric == 'Total kBytes':
value = value * 1024
# Send metric as a gauge, if applicable
if metric in self.GAUGES:
metric_name = self.GAUGES[metric]
self.gauge(metric_name, value, tags=tags)
# Send metric as a rate, if applicable
if metric in self.RATES:
metric_name = self.RATES[metric]
self.rate(metric_name, value, tags=tags)
def check(self, instance):
if 'monitor_agent_url' not in instance:
raise Exception('Fluentd instance missing "monitor_agent_url" value.')
try:
url = instance.get('monitor_agent_url')
plugin_ids = instance.get('plugin_ids', [])
# Fallback with `tag_by: plugin_id`
tag_by = instance.get('tag_by')
tag_by = tag_by if tag_by in self._AVAILABLE_TAGS else 'plugin_id'
parsed_url = urlparse.urlparse(url)
monitor_agent_host = parsed_url.hostname
monitor_agent_port = parsed_url.port or 24220
service_check_tags = ['fluentd_host:%s' % monitor_agent_host, 'fluentd_port:%s'
% monitor_agent_port]
r = requests.get(url, headers=headers(self.agentConfig))
r.raise_for_status()
status = r.json()
for p in status['plugins']:
tag = "%s:%s" % (tag_by, p.get(tag_by))
for m in self.GAUGES:
if p.get(m) is None:
continue
# Filter unspecified plugins to keep backward compatibility.
if len(plugin_ids) == 0 or p.get('plugin_id') in plugin_ids:
self.gauge('fluentd.%s' % (m), p.get(m), [tag])
except Exception, e:
msg = "No stats could be retrieved from %s : %s" % (url, str(e))
self.service_check(self.SERVICE_CHECK_NAME, AgentCheck.CRITICAL,
tags=service_check_tags, message=msg)
raise
def _get_data(self, instance):
url = instance.get('nginx_status_url')
auth = None
if 'user' in instance and 'password' in instance:
auth = (instance['user'], instance['password'])
# Submit a service check for status page availability.
parsed_url = urlparse.urlparse(url)
nginx_host = parsed_url.hostname
nginx_port = parsed_url.port or 80
service_check_name = 'nginx.can_connect'
service_check_tags = ['host:%s' % nginx_host, 'port:%s' % nginx_port]
try:
r = requests.get(url, auth=auth, headers=headers(self.agentConfig))
r.raise_for_status()
except Exception:
self.service_check(service_check_name, AgentCheck.CRITICAL,
tags=service_check_tags)
raise
else:
self.service_check(service_check_name, AgentCheck.OK,
tags=service_check_tags)
body = r.content
resp_headers = r.headers
return body, resp_headers.get('content-type', 'text/plain')
def _get_data(self, instance):
url = instance.get("nginx_status_url")
ssl_validation = instance.get("ssl_validation", True)
auth = None
if "user" in instance and "password" in instance:
auth = (instance["user"], instance["password"])
# Submit a service check for status page availability.
parsed_url = urlparse.urlparse(url)
nginx_host = parsed_url.hostname
nginx_port = parsed_url.port or 80
service_check_name = "nginx.can_connect"
service_check_tags = ["host:%s" % nginx_host, "port:%s" % nginx_port]
try:
self.log.debug(u"Querying URL: {0}".format(url))
r = requests.get(url, auth=auth, headers=headers(self.agentConfig), verify=ssl_validation)
r.raise_for_status()
except Exception:
self.service_check(service_check_name, AgentCheck.CRITICAL, tags=service_check_tags)
raise
else:
self.service_check(service_check_name, AgentCheck.OK, tags=service_check_tags)
body = r.content
resp_headers = r.headers
return body, resp_headers.get("content-type", "text/plain")
def _fetch_data(self, instance):
if 'kong_status_url' not in instance:
raise Exception('missing "kong_status_url" value')
tags = instance.get('tags', [])
url = instance.get('kong_status_url')
parsed_url = urlparse.urlparse(url)
host = parsed_url.hostname
port = parsed_url.port
service_check_name = 'kong.can_connect'
service_check_tags = ['kong_host:%s' % host, 'kong_port:%s' % port]
try:
self.log.debug(u"Querying URL: {0}".format(url))
response = requests.get(url, headers=headers(self.agentConfig))
self.log.debug(u"Kong status `response`: {0}".format(response))
response.raise_for_status()
except Exception:
self.service_check(service_check_name, AgentCheck.CRITICAL,
tags=service_check_tags)
raise
else:
if response.status_code == 200:
self.service_check(service_check_name, AgentCheck.OK,
tags=service_check_tags)
else:
self.service_check(service_check_name, AgentCheck.CRITICAL,
tags=service_check_tags)
return self._parse_json(response.content, tags)
def _get_data(self, instance):
url = instance.get('nginx_status_url')
req = urllib2.Request(url, None, headers(self.agentConfig))
if 'user' in instance and 'password' in instance:
add_basic_auth(req, instance['user'], instance['password'])
# Submit a service check for status page availability.
parsed_url = urlparse.urlparse(url)
nginx_host = parsed_url.hostname
nginx_port = parsed_url.port or 80
service_check_name = 'nginx.can_connect'
service_check_tags = ['host:%s' % nginx_host, 'port:%s' % nginx_port]
try:
response = urllib2.urlopen(req)
except Exception:
self.service_check(service_check_name, AgentCheck.CRITICAL,
tags=service_check_tags)
raise
else:
self.service_check(service_check_name, AgentCheck.OK,
tags=service_check_tags)
body = response.read()
resp_headers = response.info()
return body, resp_headers.get('Content-Type', 'text/plain')
def check(self, instance):
if 'gostats_url' not in instance:
raise Exception("Missing 'gostats_url' in GoStats config")
try:
url = instance.get('gostats_url')
parsed_url = urlparse.urlparse(url)
gostats_host = parsed_url.hostname
gostats_port = parsed_url.port or 8080
r = requests.get(url, headers=headers(self.agentConfig))
r.raise_for_status()
status = r.json()
for metric in self.GAUGES:
if status.get(metric) is None:
continue
self.gauge('gostats.%s' % (metric), status.get(metric))
for metric in self.HISTGRAMS:
if status.get(metric) is None:
continue
for value in status.get(metric):
self.histogram('gostats.%s' % (metric), value)
except Exception, e:
self.log.error('error: %s' % str(e))
raise
def _get_data(self, instance):
url = instance.get('nginx_status_url')
req = urllib2.Request(url, None, headers(self.agentConfig))
if 'user' in instance and 'password' in instance:
add_basic_auth(req, instance['user'], instance['password'])
request = urllib2.urlopen(req)
return request.read()
def _fetch_data(self, instance):
if 'kong_status_url' not in instance:
raise Exception('missing "kong_status_url" value')
tags = instance.get('tags', [])
url = instance.get('kong_status_url')
parsed_url = urlparse.urlparse(url)
host = parsed_url.hostname
port = parsed_url.port or 80
service_check_name = 'kong.can_connect'
service_check_tags = ['kong_host:%s' % host, 'kong_port:%s' % port]
try:
self.log.debug(u"Querying URL: {0}".format(url))
response = requests.get(url, headers=headers(self.agentConfig))
self.log.debug(u"Kong status `response`: {0}".format(response))
response.raise_for_status()
except Exception:
self.service_check(service_check_name,
AgentCheck.CRITICAL,
tags=service_check_tags)
raise
else:
if response.status_code == 200:
self.service_check(service_check_name,
AgentCheck.OK,
tags=service_check_tags)
else:
self.service_check(service_check_name,
AgentCheck.CRITICAL,
tags=service_check_tags)
return self._parse_json(response.content, tags)
def _get_json(self, url, ssl_params, timeout):
try:
certificate = None
if 'ssl_certfile' in ssl_params and 'ssl_keyfile' in ssl_params:
certificate = (ssl_params['ssl_certfile'],
ssl_params['ssl_keyfile'])
verify = ssl_params.get(
'ssl_ca_certs',
True) if ssl_params['ssl_cert_validation'] else False
r = requests.get(url,
verify=verify,
cert=certificate,
timeout=timeout,
headers=headers(self.agentConfig))
except requests.exceptions.Timeout:
# If there's a timeout
self.service_check(self.SERVICE_CHECK_NAME,
AgentCheck.CRITICAL,
message="Timeout when hitting %s" % url,
tags=["url:{0}".format(url)])
raise
if r.status_code != 200:
self.service_check(self.SERVICE_CHECK_NAME,
AgentCheck.CRITICAL,
message="Got %s when hitting %s" %
(r.status_code, url),
tags=["url:{0}".format(url)])
raise Exception("Http status code {0} on url {1}".format(
r.status_code, url))
return r.json()
def _get_data(self,
url,
auth=None,
send_service_check=False,
service_check_tags=None):
""" Hit a given URL and return the parsed json
`auth` is a tuple of (username, password) or None
"""
req = urllib2.Request(url, None, headers(self.agentConfig))
if auth:
add_basic_auth(req, *auth)
try:
request = urllib2.urlopen(req)
except urllib2.URLError as e:
if send_service_check:
self.service_check(self.SERVICE_CHECK_CONNECT_NAME,
AgentCheck.CRITICAL,
tags=service_check_tags,
message=e.reason)
raise
except Exception as e:
if send_service_check:
self.service_check(self.SERVICE_CHECK_CONNECT_NAME,
AgentCheck.CRITICAL,
tags=service_check_tags,
message=str(e))
raise
response = request.read()
return json.loads(response)
def save_image(self, url):
"""
保存图片，并且实现去重
:param url: Image url
:return:
"""
local_path = f"./images/{sys.argv[1]}/"
os.makedirs(local_path, exist_ok=True)
content = ""
try:
response = requests.get(url, headers=headers(), verify=False)
if response.status_code == 200:
# print(f"{'---'*20} Image url request success {response.status_code}{'---'*40}")
content = response.content
except Exception as e:
# print(f"----Request image url failure----")
return
if len(content) == 0:
return
name = image_name(content)
image_url = re.findall(r"thumburl=(.*)", url)[0]
name_suffix = self.re_name_suffix.findall(image_url)
if len(name_suffix) == 1:
suffix = name_suffix[0]
if os.path.exists(local_path + f"{name}{suffix}"):
print(f"图片 {name}{suffix} 已经存在")
# time.sleep(SECONDS)
return
with open(local_path + f"{name}{suffix}", "ab") as f:
f.write(content)
f.flush()
print(f"下载 {name}.jpg 成功")
def check(self, logger, agentConfig):
if 'rabbitMQStatusUrl' not in agentConfig or \
'rabbitMQUser' not in agentConfig or \
'rabbitMQPass' not in agentConfig or \
agentConfig['rabbitMQStatusUrl'] == 'http://www.example.com:55672/json':
return False
try:
logger.debug('getRabbitMQStatus: attempting authentication setup')
manager = urllib2.HTTPPasswordMgrWithDefaultRealm()
manager.add_password(None, agentConfig['rabbitMQStatusUrl'], agentConfig['rabbitMQUser'], agentConfig['rabbitMQPass'])
handler = urllib2.HTTPBasicAuthHandler(manager)
opener = urllib2.build_opener(handler)
urllib2.install_opener(opener)
logger.debug('getRabbitMQStatus: attempting urlopen')
req = urllib2.Request(agentConfig['rabbitMQStatusUrl'], None, headers(agentConfig))
# Do the request, log any errors
request = urllib2.urlopen(req)
response = request.read()
return json.loads(response)
except:
logger.exception('Unable to get RabbitMQ status')
return False
def _process_status(self, status_url, auth, tags):
data = {}
try:
# TODO: adding the 'full' parameter gets you per-process detailed
# informations, which could be nice to parse and output as metrics
resp = requests.get(status_url, auth=auth,
headers=headers(self.agentConfig),
params={'json': True})
resp.raise_for_status()
data = resp.json()
except Exception as e:
self.log.error("Failed to get metrics from {0}.\nError {1}".format(status_url, e))
raise
pool_name = data.get('pool', 'default')
metric_tags = tags + ["pool:{0}".format(pool_name)]
for key, mname in self.GAUGES.iteritems():
if key not in data:
self.log.warn("Gauge metric {0} is missing from FPM status".format(key))
continue
self.gauge(mname, int(data[key]), tags=metric_tags)
for key, mname in self.MONOTONIC_COUNTS.iteritems():
if key not in data:
self.log.warn("Counter metric {0} is missing from FPM status".format(key))
continue
self.monotonic_count(mname, int(data[key]), tags=metric_tags)
# return pool, to tag the service check with it if we have one
return pool_name
def _get_data(self, url, config, send_sc=True):
""" Hit a given URL and return the parsed json
"""
# Load basic authentication configuration, if available.
if config.username and config.password:
auth = (config.username, config.password)
else:
auth = None
try:
resp = requests.get(
url,
timeout=config.timeout,
headers=headers(self.agentConfig),
auth=auth
)
resp.raise_for_status()
except Exception as e:
if send_sc:
self.service_check(
self.SERVICE_CHECK_CONNECT_NAME,
AgentCheck.CRITICAL,
message="Error {0} when hitting {1}".format(e, url),
tags=config.service_check_tags
)
raise
return resp.json()
def check(self, instance):
if 'gostats_url' not in instance:
raise Exception("Missing 'gostats_url' in GoStats config")
try:
url = instance.get('gostats_url')
parsed_url = urlparse.urlparse(url)
gostats_host = parsed_url.hostname
gostats_port = parsed_url.port or 8080
r = requests.get(url, headers=headers(self.agentConfig))
r.raise_for_status()
status = r.json()
for metric in self.GAUGES:
if status.get(metric) is None:
continue
self.gauge('gostats.%s' % (metric), status.get(metric))
for metric in self.HISTGRAMS:
if status.get(metric) is None:
continue
for value in status.get(metric):
self.histogram('gostats.%s' % (metric), value)
except Exception, e:
self.log.error('error: %s' % str(e))
raise
def check(self, instance):
if 'apache_status_url' not in instance:
self.log.warn("Missing 'apache_status_url' in Apache config")
return
tags = instance.get('tags', [])
req = urllib2.Request(instance['apache_status_url'], None,
headers(self.agentConfig))
request = urllib2.urlopen(req)
response = request.read()
# Loop through and extract the numerical values
for line in response.split('\n'):
values = line.split(': ')
if len(values) == 2: # match
metric, value = values
try:
value = float(value)
except ValueError:
continue
# Special case: kBytes => bytes
if metric == 'Total kBytes':
value = value * 1024
# Send metric as a gauge, if applicable
if metric in self.GAUGES:
metric_name = self.GAUGES[metric]
self.gauge(metric_name, value, tags=tags)
# Send metric as a rate, if applicable
if metric in self.RATES:
metric_name = self.RATES[metric]
self.rate(metric_name, value, tags=tags)
def _get_metrics(self, url, tags):
req = urllib2.Request(url, None, headers(self.agentConfig))
request = urllib2.urlopen(req)
response = request.read()
# Thanks to http://hostingfu.com/files/nginx/nginxstats.py for this code
# Connections
parsed = re.search(r'Active connections:\s+(\d+)', response)
if parsed:
connections = int(parsed.group(1))
self.gauge("nginx.net.connections", connections, tags=tags)
# Requests per second
parsed = re.search(r'\s*(\d+)\s+(\d+)\s+(\d+)', response)
if parsed:
requests = int(parsed.group(3))
self.rate("nginx.net.request_per_s", requests, tags=tags)
# Connection states, reading, writing or waiting for clients
parsed = re.search(r'Reading: (\d+)\s+Writing: (\d+)\s+Waiting: (\d+)',
response)
if parsed:
reading, writing, waiting = map(int, parsed.groups())
assert connections == reading + writing + waiting
self.gauge("nginx.net.reading", reading, tags=tags)
self.gauge("nginx.net.writing", writing, tags=tags)
self.gauge("nginx.net.waiting", waiting, tags=tags)
def _process_ping(self, ping_url, ping_reply, auth, tags, pool_name,
http_host):
if ping_reply is None:
ping_reply = 'pong'
sc_tags = ["ping_url:{0}".format(ping_url)]
try:
# TODO: adding the 'full' parameter gets you per-process detailed
# informations, which could be nice to parse and output as metrics
resp = requests.get(ping_url,
auth=auth,
headers=headers(self.agentConfig,
http_host=http_host))
resp.raise_for_status()
if ping_reply not in resp.text:
raise Exception("Received unexpected reply to ping {0}".format(
resp.text))
except Exception as e:
self.log.error("Failed to ping FPM pool {0} on URL {1}."
"\nError {2}".format(pool_name, ping_url, e))
self.service_check(self.SERVICE_CHECK_NAME,
AgentCheck.CRITICAL,
tags=sc_tags,
message=str(e))
else:
self.service_check(self.SERVICE_CHECK_NAME,
AgentCheck.OK,
tags=sc_tags)
def check(self, logger, agentConfig):
if 'rabbitmq_status_url' not in agentConfig or \
'rabbitmq_user' not in agentConfig or \
'rabbitmq_pass' not in agentConfig or \
agentConfig['rabbitmq_status_url'] == 'http://www.example.com:55672/json':
return False
try:
logger.debug('getRabbitMQStatus: attempting authentication setup')
manager = urllib2.HTTPPasswordMgrWithDefaultRealm()
manager.add_password(None, agentConfig['rabbitmq_status_url'], agentConfig['rabbitmq_user'], agentConfig['rabbitmq_pass'])
handler = urllib2.HTTPBasicAuthHandler(manager)
opener = urllib2.build_opener(handler)
urllib2.install_opener(opener)
logger.debug('getRabbitMQStatus: attempting urlopen')
req = urllib2.Request(agentConfig['rabbitmq_status_url'], None, headers(agentConfig))
# Do the request, log any errors
request = urllib2.urlopen(req)
response = request.read()
return json.loads(response)
except:
logger.exception('Unable to get RabbitMQ status')
return False
def _get_data(self, instance):
url = instance.get('nginx_status_url')
req = urllib2.Request(url, None, headers(self.agentConfig))
if 'user' in instance and 'password' in instance:
add_basic_auth(req, instance['user'], instance['password'])
# Submit a service check for status page availability.
parsed_url = urlparse.urlparse(url)
nginx_host = parsed_url.hostname
nginx_port = parsed_url.port or 80
service_check_name = 'nginx.can_connect'
service_check_tags = ['host:%s' % nginx_host, 'port:%s' % nginx_port]
try:
response = urllib2.urlopen(req)
except Exception:
self.service_check(service_check_name, AgentCheck.CRITICAL,
tags=service_check_tags)
raise
else:
self.service_check(service_check_name, AgentCheck.OK,
tags=service_check_tags)
body = response.read()
resp_headers = response.info()
return body, resp_headers.get('Content-Type', 'text/plain')
def check(self, instance):
if 'monitor_agent_url' not in instance:
raise Exception('Fluentd instance missing "monitor_agent_url" value.')
try:
url = instance.get('monitor_agent_url')
plugin_ids = instance.get('plugin_ids', [])
parsed_url = urlparse.urlparse(url)
monitor_agent_host = parsed_url.hostname
monitor_agent_port = parsed_url.port or 24220
service_check_tags = ['fluentd_host:%s' % monitor_agent_host, 'fluentd_port:%s' % monitor_agent_port]
r = requests.get(url, headers=headers(self.agentConfig))
r.raise_for_status()
status = r.json()
for p in status['plugins']:
for m in self.GAUGES:
if p.get(m) is None:
continue
if p.get('plugin_id') in plugin_ids:
self.gauge('fluentd.%s' % (m), p.get(m), ["plugin_id:%s" % p.get('plugin_id')])
except Exception, e:
msg = "No stats could be retrieved from %s : %s" % (url, str(e))
self.service_check(self.SERVICE_CHECK_NAME, AgentCheck.CRITICAL, tags=service_check_tags, message=msg)
raise e
def _process_ping(self, ping_url, ping_reply, auth, tags, pool_name):
if ping_reply is None:
ping_reply = 'pong'
sc_tags = ["ping_url:{0}".format(ping_url)]
try:
resp = requests.get(ping_url,
auth=auth,
headers=headers(self.agentConfig))
resp.raise_for_status()
if ping_reply not in resp.text:
raise Exception("Received unexpected reply to ping {0}".format(
resp.text))
except Exception as e:
self.log.error("Failed to ping FPM pool {0} on URL {1}."
"\nError {2}".format(pool_name, ping_url, e))
self.service_check(self.SERVICE_CHECK_NAME,
AgentCheck.CRITICAL,
tags=sc_tags,
message=str(e))
else:
self.service_check(self.SERVICE_CHECK_NAME,
AgentCheck.OK,
tags=sc_tags)
def _perform_request(self, url, ssl_validation, auth):
r = requests.get(url,
auth=auth,
headers=headers(self.agentConfig),
verify=ssl_validation,
timeout=self.default_integration_http_timeout)
r.raise_for_status()
return r
def _check_connectivity_to_master(self, instance):
url = instance.get('gitlab_url')
if url is None:
# Simply ignore this service check if not configured
return
parsed_url = urlparse.urlparse(url)
gitlab_host = parsed_url.hostname
gitlab_port = parsed_url.port or 80
service_check_tags = ['gitlab_host:%s' % gitlab_host, 'gitlab_port:%s' % gitlab_port]
## Load the ssl configuration
ssl_params = {
'ssl_cert_validation': _is_affirmative(instance.get('ssl_cert_validation', True)),
'ssl_ca_certs': instance.get('ssl_ca_certs'),
}
for key, param in ssl_params.items():
if param is None:
del ssl_params[key]
verify_ssl = ssl_params.get('ssl_ca_certs', True) if ssl_params['ssl_cert_validation'] else False
## Timeout settings
timeouts = (int(instance.get('connect_timeout', GitlabRunnerCheck.DEFAULT_CONNECT_TIMEOUT)),
int(instance.get('receive_timeout', GitlabRunnerCheck.DEFAULT_RECEIVE_TIMEOUT)))
## Auth settings
auth = None
if 'gitlab_user' in instance and 'gitlab_password' in instance:
auth = (instance['gitlab_user'], instance['gitlab_password'])
try:
self.log.debug('checking connectivity against %s' % url)
r = requests.get(url, auth=auth, verify=verify_ssl, timeout=timeouts,
headers=headers(self.agentConfig))
if r.status_code != 200:
self.service_check(self.MASTER_SERVICE_CHECK_NAME, PrometheusCheck.CRITICAL,
message="Got %s when hitting %s" % (r.status_code, url),
tags=service_check_tags)
raise Exception("Http status code {0} on url {1}".format(r.status_code, url))
else:
r.raise_for_status()
except requests.exceptions.Timeout:
# If there's a timeout
self.service_check(self.MASTER_SERVICE_CHECK_NAME, PrometheusCheck.CRITICAL,
message="Timeout when hitting %s" % url,
tags=service_check_tags)
raise
except Exception as e:
self.service_check(self.MASTER_SERVICE_CHECK_NAME, PrometheusCheck.CRITICAL,
message="Error hitting %s. Error: %s" % (url, e.message),
tags=service_check_tags)
raise
else:
self.service_check(self.MASTER_SERVICE_CHECK_NAME, PrometheusCheck.OK, tags=service_check_tags)
self.log.debug("gitlab check succeeded")
def _get_data(self, instance):
url = instance.get('nginx_status_url')
req = urllib2.Request(url, None, headers(self.agentConfig))
if 'user' in instance and 'password' in instance:
add_basic_auth(req, instance['user'], instance['password'])
response = urllib2.urlopen(req)
body = response.read()
resp_headers = response.info()
return body, resp_headers.get('Content-Type', 'text/plain')
def _get_data(self, instance):
url = instance.get("nginx_status_url")
req = urllib2.Request(url, None, headers(self.agentConfig))
if "user" in instance and "password" in instance:
add_basic_auth(req, instance["user"], instance["password"])
response = urllib2.urlopen(req)
body = response.read()
resp_headers = response.info()
return body, resp_headers.get("Content-Type", "text/plain")
def _get_response_from_url(self, url, instance, params=None):
self.log.debug('Fetching uwsgi stats at url: %s' % url)
request_headers = headers(self.agentConfig)
response = requests.get(url, params=params, headers=request_headers,
timeout=int(instance.get('timeout', self.TIMEOUT)))
response.raise_for_status()
return response
def _get_data(self, instance):
url = instance.get('nginx_status_url')
req = urllib2.Request(url, None, headers(self.agentConfig))
if 'user' in instance and 'password' in instance:
add_basic_auth(req, instance['user'], instance['password'])
response = urllib2.urlopen(req)
body = response.read()
resp_headers = response.info()
return body, resp_headers.get('Content-Type', 'text/plain')
def _get_stats(self, url):
"Hit a given URL and return the parsed json"
self.log.debug('Fetching Couchbase stats at url: %s' % url)
req = urllib2.Request(url, None, headers(self.agentConfig))
# Do the request, log any errors
request = urllib2.urlopen(req)
response = request.read()
return json.loads(response)
def check(self, instance):
if 'lighttpd_status_url' not in instance:
raise Exception("Missing 'lighttpd_status_url' variable in Lighttpd config")
url = self.assumed_url.get(instance['lighttpd_status_url'], instance['lighttpd_status_url'])
tags = instance.get('tags', [])
self.log.debug("Connecting to %s" % url)
req = urllib2.Request(url, None, headers(self.agentConfig))
if 'user' in instance and 'password' in instance:
add_basic_auth(req, instance['user'], instance['password'])
request = urllib2.urlopen(req)
headers_resp = request.info().headers
server_version = self._get_server_version(headers_resp)
response = request.read()
metric_count = 0
# Loop through and extract the numerical values
for line in response.split('\n'):
values = line.split(': ')
if len(values) == 2: # match
metric, value = values
try:
value = float(value)
except ValueError:
continue
# Special case: kBytes => bytes
if metric == 'Total kBytes':
value = value * 1024
# Send metric as a gauge, if applicable
if metric in self.GAUGES:
metric_count += 1
metric_name = self.GAUGES[metric]
self.gauge(metric_name, value, tags=tags)
# Send metric as a rate, if applicable
if metric in self.RATES:
metric_count += 1
metric_name = self.RATES[metric]
self.rate(metric_name, value, tags=tags)
# Send metric as a counter, if applicable
if metric in self.COUNTERS:
metric_count += 1
metric_name = self.COUNTERS[metric]
self.increment(metric_name, value, tags=tags)
if metric_count == 0:
url_suffix = self.URL_SUFFIX_PER_VERSION[server_version]
if self.assumed_url.get(instance['lighttpd_status_url'], None) is None and url[-len(url_suffix):] != url_suffix:
self.assumed_url[instance['lighttpd_status_url']] = '%s%s' % (url, url_suffix)
self.warning("Assuming url was not correct. Trying to add %s suffix to the url" % url_suffix)
self.check(instance)
else:
raise Exception("No metrics were fetched for this instance. Make sure that %s is the proper url." % instance['lighttpd_status_url'])
def _get_stats(self, url):
"Hit a given URL and return the parsed json"
self.log.debug('Fetching Couchbase stats at url: %s' % url)
req = urllib2.Request(url, None, headers(self.agentConfig))
# Do the request, log any errors
request = urllib2.urlopen(req)
response = request.read()
return json.loads(response)
def _get_data(self, url, auth=None):
""" Hit a given URL and return the parsed json
`auth` is a tuple of (username, password) or None
"""
req = urllib2.Request(url, None, headers(self.agentConfig))
if auth:
add_basic_auth(req, *auth)
request = urllib2.urlopen(req)
response = request.read()
return json.loads(response)
def _get_data(self, url, auth=None):
""" Hit a given URL and return the parsed json
`auth` is a tuple of (username, password) or None
"""
req = urllib2.Request(url, None, headers(self.agentConfig))
if auth:
add_basic_auth(req, *auth)
request = urllib2.urlopen(req)
response = request.read()
return json.loads(response)
def _check_health_endpoint(self, instance, check_type, tags):
if check_type not in self.ALLOWED_SERVICE_CHECKS:
raise CheckException("Health endpoint %s is not a valid endpoint" % check_type)
url = instance.get('gitlab_url')
if url is None:
# Simply ignore this service check if not configured
self.log.debug("gitlab_url not configured, service check %s skipped" % check_type)
return
service_check_tags = self._service_check_tags(url)
service_check_tags.extend(tags)
verify_ssl = self._verify_ssl(instance)
## Timeout settings
timeouts = (int(instance.get('connect_timeout', GitlabCheck.DEFAULT_CONNECT_TIMEOUT)),
int(instance.get('receive_timeout', GitlabCheck.DEFAULT_RECEIVE_TIMEOUT)))
## Auth settings
auth = None
if 'gitlab_user' in instance and 'gitlab_password' in instance:
auth = (instance['gitlab_user'], instance['gitlab_password'])
# These define which endpoint is hit and which type of check is actually performed
# TODO: parse errors and report for single sub-service failure?
service_check_name = "gitlab.%s" % check_type
check_url = "%s/-/%s" % (url, check_type)
try:
self.log.debug('checking %s against %s' % (check_type, check_url))
r = requests.get(check_url, auth=auth, verify=verify_ssl, timeout=timeouts,
headers=headers(self.agentConfig))
if r.status_code != 200:
self.service_check(service_check_name, PrometheusCheck.CRITICAL,
message="Got %s when hitting %s" % (r.status_code, check_url),
tags=service_check_tags)
raise Exception("Http status code {0} on check_url {1}".format(r.status_code, check_url))
else:
r.raise_for_status()
except requests.exceptions.Timeout:
# If there's a timeout
self.service_check(service_check_name, PrometheusCheck.CRITICAL,
message="Timeout when hitting %s" % check_url,
tags=service_check_tags)
raise
except Exception as e:
self.service_check(service_check_name, PrometheusCheck.CRITICAL,
message="Error hitting %s. Error: %s" % (check_url, e.message),
tags=service_check_tags)
raise
else:
self.service_check(service_check_name, PrometheusCheck.OK, tags=service_check_tags)
self.log.debug("gitlab check %s succeeded" % check_type)
def _get_stats(self, url, instance):
""" Hit a given URL and return the parsed json. """
self.log.debug('Fetching Couchbase stats at url: %s' % url)
req = urllib2.Request(url, None, headers(self.agentConfig))
if 'user' in instance and 'password' in instance:
add_basic_auth(req, instance['user'], instance['password'])
timeout = float(instance.get('timeout', DEFAULT_TIMEOUT))
request = urllib2.urlopen(req, timeout=timeout)
response = request.read()
return json.loads(response)
def _get_stats(self, url, instance):
"Hit a given URL and return the parsed json"
self.log.debug('Fetching Couchbase stats at url: %s' % url)
req = urllib2.Request(url, None, headers(self.agentConfig))
if 'user' in instance and 'password' in instance:
add_basic_auth(req, instance['user'], instance['password'])
# Do the request, log any errors
request = urllib2.urlopen(req)
response = request.read()
return json.loads(response)
def _fetch_data(self, url, username, password):
auth = (username, password)
url = "%s%s" % (url, STATS_URL)
self.log.debug("HAProxy Fetching haproxy search data from: %s" % url)
r = requests.get(url, auth=auth, headers=headers(self.agentConfig))
r.raise_for_status()
return r.content.splitlines()
def _get_stats(self, url, instance):
""" Hit a given URL and return the parsed json. """
self.log.debug('Fetching Couchbase stats at url: %s' % url)
req = urllib2.Request(url, None, headers(self.agentConfig))
if 'user' in instance and 'password' in instance:
add_basic_auth(req, instance['user'], instance['password'])
timeout = float(instance.get('timeout', DEFAULT_TIMEOUT))
request = urllib2.urlopen(req, timeout=timeout)
response = request.read()
return json.loads(response)
def _get_response_from_url(self, url, instance, params=None):
self.log.debug('Fetching Celery stats at url: %s' % url)
auth=None
if 'username' and 'password' in instance:
auth = (instance['username'], instance['password'])
request_headers = headers(self.agentConfig)
response = requests.get(url, params=params, auth=auth, headers=request_headers,
timeout=int(instance.get('timeout', self.TIMEOUT)))
response.raise_for_status()
return response
def _get_stats(self, agentConfig, url):
"Hit a given URL and return the parsed json"
try:
req = urllib2.Request(url, None, headers(agentConfig))
# Do the request, log any errors
request = urllib2.urlopen(req)
response = request.read()
return json.loads(response)
except:
self.logger.exception('Unable to get CouchDB statistics')
return None
def _get_stats(self, url, instance):
"Hit a given URL and return the parsed json"
self.log.debug('Fetching Couchdb stats at url: %s' % url)
req = urllib2.Request(url, None, headers(self.agentConfig))
if 'user' in instance and 'password' in instance:
add_basic_auth(req, instance['user'], instance['password'])
# Do the request, log any errors
request = urllib2.urlopen(req)
response = request.read()
return json.loads(response)
def _get_stats(self, agentConfig, url):
"Hit a given URL and return the parsed json"
try:
req = urllib2.Request(url, None, headers(agentConfig))
# Do the request, log any errors
request = urllib2.urlopen(req)
response = request.read()
return json.loads(response)
except:
self.logger.exception('Unable to get CouchDB statistics')
return None
def _get_stats(self, url, instance):
"Hit a given URL and return the parsed json"
self.log.debug('Fetching Couchdb stats at url: %s' % url)
auth = None
if 'user' in instance and 'password' in instance:
auth = (instance['user'], instance['password'])
r = requests.get(url, auth=auth, headers=headers(self.agentConfig),
timeout=int(instance.get('timeout', self.TIMEOUT)))
r.raise_for_status()
return r.json()
def _get_stats(self, url, instance):
""" Hit a given URL and return the parsed json. """
self.log.debug("Fetching Couchbase stats at url: %s" % url)
timeout = float(instance.get("timeout", DEFAULT_TIMEOUT))
auth = None
if "user" in instance and "password" in instance:
auth = (instance["user"], instance["password"])
r = requests.get(url, auth=auth, headers=headers(self.agentConfig), timeout=timeout)
r.raise_for_status()
return r.json()
def _fetch_url_data(self, url, username, password, verify):
''' Hit a given http url and return the stats lines '''
# Try to fetch data from the stats URL
auth = (username, password)
url = "%s%s" % (url, STATS_URL)
self.log.debug("Fetching haproxy stats from url: %s" % url)
response = requests.get(url, auth=auth, headers=headers(self.agentConfig), verify=verify, timeout=self.default_integration_http_timeout)
response.raise_for_status()
return response.content.splitlines()
def _fetch_url_data(self, url, username, password, verify):
''' Hit a given http url and return the stats lines '''
# Try to fetch data from the stats URL
auth = (username, password)
url = "%s%s" % (url, STATS_URL)
self.log.debug("Fetching haproxy stats from url: %s" % url)
response = requests.get(url, auth=auth, headers=headers(self.agentConfig), verify=verify, timeout=self.default_integration_http_timeout)
response.raise_for_status()
return response.content.splitlines()
def _fetch_data(self, url, username, password):
''' Hit a given URL and return the parsed json '''
# Try to fetch data from the stats URL
auth = (username, password)
url = "%s%s" % (url, STATS_URL)
self.log.debug("HAProxy Fetching haproxy search data from: %s" % url)
r = requests.get(url, auth=auth, headers=headers(self.agentConfig))
r.raise_for_status()
return r.content.splitlines()
def _perform_request(self, url, path, ssl_params, timeout):
certificate = None
if 'ssl_certfile' in ssl_params and 'ssl_keyfile' in ssl_params:
certificate = (ssl_params['ssl_certfile'],
ssl_params['ssl_keyfile'])
verify = ssl_params.get(
'ssl_ca_certs',
True) if ssl_params['ssl_cert_validation'] else False
return requests.get(url + path,
verify=verify,
cert=certificate,
timeout=timeout,
headers=headers(self.agentConfig))
