|
1
|
|
|
#!/usr/bin/python
|
|
2
|
|
|
# coding=utf-8
|
|
3
|
|
|
|
|
4
|
|
|
import requests
|
|
5
|
|
|
import httplib
|
|
6
|
|
|
import time
|
|
7
|
|
|
import sys
|
|
8
|
|
|
|
|
9
|
|
|
from system.logging import Logger
|
|
10
|
|
|
from utils import Utils
|
|
11
|
|
|
#
|
|
12
|
|
|
#
|
|
13
|
|
|
# Incident Statuses
|
|
14
|
|
|
#
|
|
15
|
|
|
# 0 = Scheduled
|
|
16
|
|
|
# 1 = Investigating
|
|
17
|
|
|
# 2 = Identified
|
|
18
|
|
|
# 3 = Watching
|
|
19
|
|
|
# 4 = Fixed
|
|
20
|
|
|
#
|
|
21
|
|
|
# Component Statuses
|
|
22
|
|
|
#
|
|
23
|
|
|
# 1 = Operational
|
|
24
|
|
|
# 2 = Performance Issues
|
|
25
|
|
|
# 3 = Partial Outage
|
|
26
|
|
|
# 4 = Major Outage
|
|
27
|
|
|
|
|
28
|
|
|
|
|
29
|
|
|
|
|
30
|
|
|
class Cachet(object):
|
|
31
|
|
|
httpErrors = {
|
|
32
|
|
|
# Cloudflare Errors
|
|
33
|
|
|
520: "Web server is returning an unknown error (Cloudflare)",
|
|
34
|
|
|
521: "Web server is down (Cloudflare)",
|
|
35
|
|
|
522: "Connection timed out (Cloudflare)",
|
|
36
|
|
|
523: "Origin is unreachable (Cloudflare)",
|
|
37
|
|
|
524: "A timeout occurred (Cloudflare)",
|
|
38
|
|
|
525: "SSL handshake failed (Cloudflare)",
|
|
39
|
|
|
526: "Invalid SSL certificate (Cloudflare)",
|
|
40
|
|
|
# Nginx Errors
|
|
41
|
|
|
444: "No Response (Nginx)",
|
|
42
|
|
|
494: "Request Header Too Large (Nginx)",
|
|
43
|
|
|
495: "Cert Error (Nginx)",
|
|
44
|
|
|
496: "No Cert (Nginx)",
|
|
45
|
|
|
497: "HTTP to HTTPS (Nginx)",
|
|
46
|
|
|
499: "Client Closed Request (Nginx)",
|
|
47
|
|
|
# Other
|
|
48
|
|
|
# 1xx
|
|
49
|
|
|
102: "Server has received and is processing the request",
|
|
50
|
|
|
103: "resume aborted PUT or POST requests",
|
|
51
|
|
|
122: "URI is longer than a maximum of 2083 characters",
|
|
52
|
|
|
# 2xx
|
|
53
|
|
|
207: "XML return with possible multiple seperate responses.",
|
|
54
|
|
|
208: "The results are previously returned.",
|
|
55
|
|
|
226: "The request has been fulfilled amd response is in instance manipulations.",
|
|
56
|
|
|
# 3xx
|
|
57
|
|
|
308: "Please connect again to the different URL using the same method.",
|
|
58
|
|
|
# 4xx
|
|
59
|
|
|
418: "I'm a teapot.",
|
|
60
|
|
|
420: "Method Failure",
|
|
61
|
|
|
421: "Enhance Your Calm",
|
|
62
|
|
|
422: "Unprocessable Entity",
|
|
63
|
|
|
423: "Locked",
|
|
64
|
|
|
424: "Failed Dependency",
|
|
65
|
|
|
426: "Upgrade Required",
|
|
66
|
|
|
428: "Precondition Required",
|
|
67
|
|
|
429: "Too Many Requests",
|
|
68
|
|
|
431: "Request Header Fields Too Large",
|
|
69
|
|
|
440: "Login Timeout (Microsoft)",
|
|
70
|
|
|
449: "Retry With (Microsoft)",
|
|
71
|
|
|
450: "Blocked by Windows Parental Controls",
|
|
72
|
|
|
451: "Unavailable For Legal Reasons",
|
|
73
|
|
|
# 5xx
|
|
74
|
|
|
506: "Variant Also Negotiates (RFC 2295)",
|
|
75
|
|
|
507: "Insufficient Storage (WebDAV; RFC 4918)",
|
|
76
|
|
|
508: "Loop Detected (WebDAV; RFC 5842)",
|
|
77
|
|
|
509: "Bandwidth Limit Exceeded (Apache bw/limited extension)",
|
|
78
|
|
|
510: "Not Extended (RFC 2774)",
|
|
79
|
|
|
511: "Network Authentication Required"
|
|
80
|
|
|
|
|
81
|
|
|
}
|
|
82
|
|
|
|
|
83
|
|
|
def __init__(self):
|
|
84
|
|
|
self.logs = Logger()
|
|
85
|
|
|
self.utils = Utils()
|
|
86
|
|
|
self.config = self.utils.readConfig()
|
|
87
|
|
|
self.base_url = self.config['api_url']
|
|
88
|
|
|
self.api_url = '%s/api/v1' % self.config['api_url']
|
|
89
|
|
|
self.api_token = self.config['api_token']
|
|
90
|
|
|
self.maxRetries = self.config['retries']
|
|
91
|
|
|
|
|
92
|
|
|
try:
|
|
93
|
|
|
if self.checkInitialPing() == 200:
|
|
94
|
|
|
self.checkSites()
|
|
95
|
|
|
else:
|
|
96
|
|
|
exit()
|
|
97
|
|
|
except Exception as e:
|
|
98
|
|
|
self.logs.error(e)
|
|
99
|
|
|
exit()
|
|
100
|
|
|
|
|
101
|
|
|
def checkSites(self):
|
|
102
|
|
|
# Count how many sites to monitor
|
|
103
|
|
|
monitor_count = len(self.config['monitoring'])
|
|
104
|
|
|
x = 0
|
|
105
|
|
|
|
|
106
|
|
|
# Loop through sites to monitor
|
|
107
|
|
|
while x < monitor_count:
|
|
108
|
|
|
|
|
109
|
|
|
isEnabled = self.config['monitoring'][x]['enabled']
|
|
110
|
|
|
status_codes = self.config['monitoring'][x]['expected_status_code']
|
|
111
|
|
|
url = self.config['monitoring'][x]['url']
|
|
112
|
|
|
request_method = self.config['monitoring'][x]['method']
|
|
113
|
|
|
c_id = self.config['monitoring'][x]['component_id']
|
|
114
|
|
|
localtime = time.asctime(time.localtime(time.time()))
|
|
115
|
|
|
current_status = self.utils.getComponentsByID(c_id).json()['data']['status']
|
|
116
|
|
|
incident_id = self.checkForIncident(c_id)
|
|
117
|
|
|
check_timeout = self.config['monitoring'][x]['timeout']
|
|
118
|
|
|
|
|
119
|
|
|
try:
|
|
120
|
|
|
if isEnabled:
|
|
121
|
|
|
if request_method.lower() == "get":
|
|
122
|
|
|
r = requests.get(url, verify=True, timeout=check_timeout)
|
|
123
|
|
|
# self.utils.postMetricsPointsByID(1, r.elapsed.total_seconds() * 1000)
|
|
124
|
|
View Code Duplication |
if r.status_code not in status_codes and r.status_code not in self.httpErrors:
|
|
125
|
|
|
error_code = '%s check **failed** - %s \n\n`%s %s HTTP StatusError: %s`' % (
|
|
126
|
|
|
url, localtime, request_method, url, httplib.responses[r.status_code])
|
|
127
|
|
|
c_status = 4
|
|
128
|
|
|
if not incident_id:
|
|
129
|
|
|
self.utils.postIncidents('%s: HTTP Status Error' % url, error_code, 1, 1,
|
|
130
|
|
|
component_id=c_id, component_status=c_status)
|
|
131
|
|
|
if current_status is not 4:
|
|
132
|
|
|
self.utils.putComponentsByID(c_id, status=c_status)
|
|
133
|
|
|
self.logs.warn("%s" % error_code.replace('\n', '').replace('`', ''))
|
|
134
|
|
|
elif r.status_code not in status_codes and r.status_code in self.httpErrors:
|
|
135
|
|
|
error_code = '%s check **failed** - %s \n\n`%s %s HTTP Status Error: %s`' % (
|
|
136
|
|
|
url, localtime, request_method, url, self.httpErrors[r.status_code])
|
|
137
|
|
|
c_status = 4
|
|
138
|
|
|
if not incident_id:
|
|
139
|
|
|
self.utils.postIncidents('%s: HTTP Status Error' % url, error_code, 1, 1,
|
|
140
|
|
|
component_id=c_id, component_status=c_status)
|
|
141
|
|
|
if current_status is not 4:
|
|
142
|
|
|
self.utils.putComponentsByID(c_id, status=c_status)
|
|
143
|
|
|
self.logs.warn("%s" % error_code.replace('\n', '').replace('`', ''))
|
|
144
|
|
View Code Duplication |
elif request_method.lower() == "post":
|
|
|
|
|
|
|
145
|
|
|
r = requests.get(url, verify=True, timeout=check_timeout)
|
|
146
|
|
|
if r.status_code not in status_codes and r.status_code not in self.httpErrors:
|
|
147
|
|
|
error_code = '%s check **failed** - %s \n\n`%s %s HTTP Status Error: %s`' % (
|
|
148
|
|
|
url, localtime, request_method, url, httplib.responses[r.status_code])
|
|
149
|
|
|
c_status = 4
|
|
150
|
|
|
if not incident_id:
|
|
151
|
|
|
self.utils.postIncidents('%s: HTTP Status Error' % url, error_code, 1, 1,
|
|
152
|
|
|
component_id=c_id, component_status=c_status)
|
|
153
|
|
|
if current_status is not 4:
|
|
154
|
|
|
self.utils.putComponentsByID(c_id, status=c_status)
|
|
155
|
|
|
self.logs.warn("%s" % error_code.replace('\n', '').replace('`', ''))
|
|
156
|
|
|
elif r.status_code not in status_codes and r.status_code in self.httpErrors:
|
|
157
|
|
|
error_code = '%s check **failed** - %s \n\n`%s %s HTTP Status Error: %s`' % (
|
|
158
|
|
|
url, localtime, request_method, url, self.httpErrors[r.status_code])
|
|
159
|
|
|
c_status = 4
|
|
160
|
|
|
if not incident_id:
|
|
161
|
|
|
self.utils.postIncidents('%s: HTTP Status Error' % url, error_code, 1, 1,
|
|
162
|
|
|
component_id=c_id, component_status=c_status)
|
|
163
|
|
|
if current_status is not 4:
|
|
164
|
|
|
self.utils.putComponentsByID(c_id, status=c_status)
|
|
165
|
|
|
self.logs.warn("%s" % error_code.replace('\n', '').replace('`', ''))
|
|
166
|
|
|
except requests.exceptions.HTTPError as e:
|
|
167
|
|
|
error_code = '%s check **failed** - %s \n\n`%s %s HTTP Error: %s`' % (
|
|
168
|
|
|
url, localtime, request_method, url, e)
|
|
169
|
|
|
c_status = 4
|
|
170
|
|
|
if not incident_id:
|
|
171
|
|
|
self.utils.postIncidents('%s: HTTP Error' % url, error_code, 1, 1, component_id=c_id,
|
|
172
|
|
|
component_status=c_status)
|
|
173
|
|
|
if current_status is not 4:
|
|
174
|
|
|
self.utils.putComponentsByID(c_id, status=c_status)
|
|
175
|
|
|
self.logs.warn(error_code.replace('\n', '').replace('`', ''))
|
|
176
|
|
|
except requests.exceptions.SSLError as e:
|
|
177
|
|
|
error_code = '%s check **failed** - %s \n\n`%s %s SSL Error: %s`' % (
|
|
178
|
|
|
url, localtime, request_method, url, e)
|
|
179
|
|
|
c_status = 4
|
|
180
|
|
|
if not incident_id:
|
|
181
|
|
|
self.utils.postIncidents('%s: SSL Error' % url, error_code, 1, 1, component_id=c_id,
|
|
182
|
|
|
component_status=c_status)
|
|
183
|
|
|
if current_status is not 4:
|
|
184
|
|
|
self.utils.putComponentsByID(c_id, status=c_status)
|
|
185
|
|
|
self.logs.warn(error_code.replace('\n', '').replace('`', ''))
|
|
186
|
|
|
except requests.exceptions.ConnectionError as e:
|
|
187
|
|
|
error_code = '%s check **failed** - %s \n\n`%s %s Connection Error: %s`' % (
|
|
188
|
|
|
url, localtime, request_method, url, e)
|
|
189
|
|
|
c_status = 4
|
|
190
|
|
|
if not incident_id:
|
|
191
|
|
|
self.utils.postIncidents('%s: Connection Error' % url, error_code, 1, 1, component_id=c_id,
|
|
192
|
|
|
component_status=c_status)
|
|
193
|
|
|
if current_status is not 4:
|
|
194
|
|
|
self.utils.putComponentsByID(c_id, status=c_status)
|
|
195
|
|
|
self.logs.warn(error_code.replace('\n', '').replace('`', ''))
|
|
196
|
|
|
except requests.exceptions.Timeout as e:
|
|
197
|
|
|
error_code = '%s check **failed** - %s \n\n`%s %s Request Timeout: %s`' % (
|
|
198
|
|
|
url, localtime, request_method, url, e)
|
|
199
|
|
|
c_status = 2
|
|
200
|
|
|
if not incident_id:
|
|
201
|
|
|
self.utils.postIncidents('%s: Request Timeout' % url, error_code, 1, 1, component_id=c_id,
|
|
202
|
|
|
component_status=c_status)
|
|
203
|
|
|
if current_status is not 4:
|
|
204
|
|
|
self.utils.putComponentsByID(c_id, status=c_status)
|
|
205
|
|
|
self.logs.warn(error_code.replace('\n', '').replace('`', ''))
|
|
206
|
|
|
except requests.exceptions.TooManyRedirects as e:
|
|
207
|
|
|
error_code = '%s check **failed** - %s \n\n`%s %s Too Many Redirects: %s`' % (
|
|
208
|
|
|
url, localtime, request_method, url, e)
|
|
209
|
|
|
c_status = 4
|
|
210
|
|
|
if not incident_id:
|
|
211
|
|
|
self.utils.postIncidents('%s: Too Many Redirects' % url, error_code, 1, 1, component_id=c_id,
|
|
212
|
|
|
component_status=c_status)
|
|
213
|
|
|
if current_status is not 4:
|
|
214
|
|
|
self.utils.putComponentsByID(c_id, status=c_status)
|
|
215
|
|
|
self.logs.warn(error_code.replace('\n', '').replace('`', ''))
|
|
216
|
|
|
except requests.exceptions.RetryError as e:
|
|
217
|
|
|
error_code = '%s check **failed** - %s \n\n`%s %s Retry Error: %s`' % (
|
|
218
|
|
|
url, localtime, request_method, url, e)
|
|
219
|
|
|
c_status = 4
|
|
220
|
|
|
if not incident_id:
|
|
221
|
|
|
self.utils.postIncidents('%s: Retry Error' % url, error_code, 1, 1, component_id=c_id,
|
|
222
|
|
|
component_status=c_status)
|
|
223
|
|
|
if current_status is not 4:
|
|
224
|
|
|
self.utils.putComponentsByID(c_id, status=c_status)
|
|
225
|
|
|
self.logs.warn(error_code.replace('\n', '').replace('`', ''))
|
|
226
|
|
|
except httplib.BadStatusLine as e:
|
|
227
|
|
|
self.logs.error("%s \nCould not fetch %s" % (e, url))
|
|
228
|
|
|
except Exception as e:
|
|
229
|
|
|
error_code = '%s check **failed** - %s \n\n`%s %s Unexpected Error: %s`' % (
|
|
230
|
|
|
url, localtime, request_method, url, e)
|
|
231
|
|
|
c_status = 4
|
|
232
|
|
|
if not incident_id:
|
|
233
|
|
|
self.utils.postIncidents('%s: Unexpected Error' % url, error_code, 1, 1, component_id=c_id,
|
|
234
|
|
|
component_status=c_status)
|
|
235
|
|
|
if current_status is not 4:
|
|
236
|
|
|
self.utils.putComponentsByID(c_id, status=c_status)
|
|
237
|
|
|
self.logs.error(error_code.replace('\n', '').replace('`', ''))
|
|
238
|
|
|
else:
|
|
239
|
|
|
if r.status_code in status_codes:
|
|
240
|
|
|
if current_status is not 1 and not incident_id:
|
|
241
|
|
|
self.utils.putComponentsByID(c_id, status=1)
|
|
242
|
|
|
self.logs.info("Issue with %s has been resolved" % url)
|
|
243
|
|
|
elif current_status is not 1 and incident_id:
|
|
244
|
|
|
incident_description = "Resolved at %s\n\n***\n\n%s" % (
|
|
245
|
|
|
localtime, self.getIncidentInfo(incident_id))
|
|
246
|
|
|
self.utils.putIncidentsByID(incident_id, message=incident_description, status=4,
|
|
247
|
|
|
component_id=c_id, component_status=1)
|
|
248
|
|
|
else:
|
|
249
|
|
|
self.logs.info("%s no issues found" % url)
|
|
250
|
|
|
|
|
251
|
|
|
x += 1
|
|
252
|
|
|
self.logs.info("############################")
|
|
253
|
|
|
|
|
254
|
|
|
def checkForIncident(self, component_id):
|
|
255
|
|
|
current_incidents = self.utils.getIncidents().json()
|
|
256
|
|
|
incidents = len(current_incidents['data'])
|
|
257
|
|
|
x = 0
|
|
258
|
|
|
|
|
259
|
|
|
while x < incidents:
|
|
260
|
|
|
incident_id = current_incidents['data'][x]['id']
|
|
261
|
|
|
incident_component_id = current_incidents['data'][x]['component_id']
|
|
262
|
|
|
incident_status = current_incidents['data'][x]['status']
|
|
263
|
|
|
|
|
264
|
|
|
if component_id == incident_component_id and incident_status is not 4:
|
|
265
|
|
|
return incident_id
|
|
266
|
|
|
x += 1
|
|
267
|
|
|
|
|
268
|
|
|
def getIncidentInfo(self, i_id):
|
|
269
|
|
|
incident = self.utils.getIncidentsByID(i_id).json()
|
|
270
|
|
|
i_description = incident['data']['message']
|
|
271
|
|
|
return i_description
|
|
272
|
|
|
|
|
273
|
|
|
def checkInitialPing(self):
|
|
274
|
|
|
rPing = requests.get(self.base_url, verify=True, timeout=10)
|
|
275
|
|
|
return rPing.status_code
|
|
276
|
|
|
|
gaz492 /
cachet-monitor
Loading history...