Improve haproxy monitoring
Added: - hrsp_1xx, hrsp_2xx, hrsp_3xx, hrsp_4xx, hrsp_5xx, hrsp_other, hrsp_total for backands and frontends - qtime, ctime, rtime, ttime metrics for backand servers - Backend Servers In UP State
Taras Kozlovskyi committed
Dec 27, 2017 at 16:39 UTC
c2eed148f4857f97bb8a35d5f7e2206fc5572eb0
1 file changed
+128
-7
python.d/haproxy.chart.py
+128
-7
@@ -1,6 +1,6 @@
1
# -*- coding: utf-8 -*-
2
# Description: haproxy netdata python.d module
3
-# Author: l2isbad
3
+# Author: l2isbad, ktarasz
4
5
from collections import defaultdict
6
from re import compile as re_compile
@@ -20,8 +20,13 @@ priority = 60000
20
retries = 60
21
22
# charts order (can be overridden if you want less charts, or different order)
23
-ORDER = ['fbin', 'fbout', 'fscur', 'fqcur', 'bbin', 'bbout', 'bscur', 'bqcur',
24
- 'health_sdown', 'health_bdown', 'health_idle']
23
+ORDER = ['fbin', 'fbout', 'fscur', 'fqcur',
24
+ 'fhrsp_1xx', 'fhrsp_2xx', 'fhrsp_3xx', 'fhrsp_4xx', 'fhrsp_5xx', 'fhrsp_other', 'fhrsp_total',
25
+ 'bbin', 'bbout', 'bscur', 'bqcur',
26
+ 'bhrsp_1xx', 'bhrsp_2xx', 'bhrsp_3xx', 'bhrsp_4xx', 'bhrsp_5xx', 'bhrsp_other', 'bhrsp_total',
27
+ 'bqtime', 'bttime', 'brtime', 'bctime',
28
+ 'health_sup', 'health_sdown', 'health_bdown', 'health_idle']
29
+
30
CHARTS = {
31
'fbin': {
32
'options': [None, "Kilobytes In", "KB/s", 'frontend', 'haproxy_f.bin', 'line'],
@@ -39,6 +44,34 @@ CHARTS = {
44
'options': [None, "Session In Queue", "sessions", 'frontend', 'haproxy_f.qcur', 'line'],
45
'lines': [
46
]},
47
+ 'fhrsp_1xx': {
48
+ 'options': [None, "HTTP responses with 1xx code", "responses", 'frontend', 'haproxy_f.hrsp_1xx', 'line'],
49
+ 'lines': [
50
+ ]},
51
+ 'fhrsp_2xx': {
52
+ 'options': [None, "HTTP responses with 2xx code", "responses", 'frontend', 'haproxy_f.hrsp_2xx', 'line'],
53
+ 'lines': [
54
+ ]},
55
+ 'fhrsp_3xx': {
56
+ 'options': [None, "HTTP responses with 3xx code", "responses", 'frontend', 'haproxy_f.hrsp_3xx', 'line'],
57
+ 'lines': [
58
+ ]},
59
+ 'fhrsp_4xx': {
60
+ 'options': [None, "HTTP responses with 4xx code", "responses", 'frontend', 'haproxy_f.hrsp_4xx', 'line'],
61
+ 'lines': [
62
+ ]},
63
+ 'fhrsp_5xx': {
64
+ 'options': [None, "HTTP responses with 5xx code", "responses", 'frontend', 'haproxy_f.hrsp_5xx', 'line'],
65
+ 'lines': [
66
+ ]},
67
+ 'fhrsp_other': {
68
+ 'options': [None, "HTTP responses with other codes (protocol error)", "responses", 'frontend', 'haproxy_f.hrsp_other', 'line'],
69
+ 'lines': [
70
+ ]},
71
+ 'fhrsp_total': {
72
+ 'options': [None, "HTTP responses", "responses", 'frontend', 'haproxy_f.hrsp_total', 'line'],
73
+ 'lines': [
74
+ ]},
75
'bbin': {
76
'options': [None, "Kilobytes In", "KB/s", 'backend', 'haproxy_b.bin', 'line'],
77
'lines': [
@@ -55,11 +88,64 @@ CHARTS = {
88
'options': [None, "Sessions In Queue", "sessions", 'backend', 'haproxy_b.qcur', 'line'],
89
'lines': [
90
]},
91
+ 'bhrsp_1xx': {
92
+ 'options': [None, "HTTP responses with 1xx code", "responses", 'backend', 'haproxy_b.hrsp_1xx', 'line'],
93
+ 'lines': [
94
+ ]},
95
+ 'bhrsp_2xx': {
96
+ 'options': [None, "HTTP responses with 2xx code", "responses", 'backend', 'haproxy_b.hrsp_2xx', 'line'],
97
+ 'lines': [
98
+ ]},
99
+ 'bhrsp_3xx': {
100
+ 'options': [None, "HTTP responses with 3xx code", "responses", 'backend', 'haproxy_b.hrsp_3xx', 'line'],
101
+ 'lines': [
102
+ ]},
103
+ 'bhrsp_4xx': {
104
+ 'options': [None, "HTTP responses with 4xx code", "responses", 'backend', 'haproxy_b.hrsp_4xx', 'line'],
105
+ 'lines': [
106
+ ]},
107
+ 'bhrsp_5xx': {
108
+ 'options': [None, "HTTP responses with 5xx code", "responses", 'backend', 'haproxy_b.hrsp_5xx', 'line'],
109
+ 'lines': [
110
+ ]},
111
+ 'bhrsp_other': {
112
+ 'options': [None, "HTTP responses with other codes (protocol error)", "responses", 'backend',
113
+ 'haproxy_b.hrsp_other', 'line'],
114
+ 'lines': [
115
+ ]},
116
+ 'bhrsp_total': {
117
+ 'options': [None, "HTTP responses (total)", "responses", 'backend', 'haproxy_b.hrsp_total', 'line'],
118
+ 'lines': [
119
+ ]},
120
+ 'bqtime': {
121
+ 'options': [None, "The average queue time over the 1024 last requests", "ms", 'backend', 'haproxy_b.qtime', 'line'],
122
+ 'lines': [
123
+ ]},
124
+ 'bctime': {
125
+ 'options': [None, "The average connect time over the 1024 last requests", "ms", 'backend',
126
+ 'haproxy_b.ctime', 'line'],
127
+ 'lines': [
128
+ ]},
129
+ 'brtime': {
130
+ 'options': [None, "The average response time over the 1024 last requests", "ms", 'backend',
131
+ 'haproxy_b.rtime', 'line'],
132
+ 'lines': [
133
+ ]},
134
+ 'bttime': {
135
+ 'options': [None, "The average total session time over the 1024 last requests", "ms", 'backend',
136
+ 'haproxy_b.ttime', 'line'],
137
+ 'lines': [
138
+ ]},
139
'health_sdown': {
140
'options': [None, "Backend Servers In DOWN State", "failed servers", 'health',
141
'haproxy_hs.down', 'line'],
142
'lines': [
143
]},
144
+ 'health_sup': {
145
+ 'options': [None, "Backend Servers In UP State", "health servers", 'health',
146
+ 'haproxy_hs.up', 'line'],
147
+ 'lines': [
148
+ ]},
149
'health_bdown': {
150
'options': [None, "Is Backend Alive? 1 = DOWN", "failed backend", 'health', 'haproxy_hb.down', 'line'],
151
'lines': [
@@ -71,10 +157,27 @@ CHARTS = {
157
]}
158
}
159
160
+
161
METRICS = {'bin': {'algorithm': 'incremental', 'divisor': 1024},
162
'bout': {'algorithm': 'incremental', 'divisor': 1024},
163
'scur': {'algorithm': 'absolute', 'divisor': 1},
77
- 'qcur': {'algorithm': 'absolute', 'divisor': 1}}
164
+ 'qcur': {'algorithm': 'absolute', 'divisor': 1},
165
+ 'hrsp_1xx': {'algorithm': 'incremental', 'divisor': 1},
166
+ 'hrsp_2xx': {'algorithm': 'incremental', 'divisor': 1},
167
+ 'hrsp_3xx': {'algorithm': 'incremental', 'divisor': 1},
168
+ 'hrsp_4xx': {'algorithm': 'incremental', 'divisor': 1},
169
+ 'hrsp_5xx': {'algorithm': 'incremental', 'divisor': 1},
170
+ 'hrsp_other': {'algorithm': 'incremental', 'divisor': 1},
171
+ }
172
+
173
+
174
+BACKEND_METRICS = {
175
+ 'qtime': {'algorithm': 'absolute', 'divisor': 1},
176
+ 'ctime': {'algorithm': 'absolute', 'divisor': 1},
177
+ 'rtime': {'algorithm': 'absolute', 'divisor': 1},
178
+ 'ttime': {'algorithm': 'absolute', 'divisor': 1}
179
+}
180
+
181
182
REGEX = dict(url=re_compile(r'idle = (?P<idle>[0-9]+)'),
183
socket=re_compile(r'Idle_pct: (?P<idle>[0-9]+)'))
@@ -139,11 +242,20 @@ class Service(UrlService, SocketService):
242
243
for backend in self.data['backend']:
244
name, idx = backend['# pxname'], backend['# pxname'].replace('.', '_')
245
+ stat_data['hsup_' + idx] = len([server for server in self.data['servers']
246
+ if server_status(server, name, 'UP')])
247
stat_data['hsdown_' + idx] = len([server for server in self.data['servers']
143
- if server_down(server, name)])
248
+ if server_status(server, name, 'DOWN')])
249
stat_data['hbdown_' + idx] = 1 if backend.get('status') == 'DOWN' else 0
250
+ for metric in BACKEND_METRICS:
251
+ stat_data['_'.join(['backend', metric, idx])] = backend.get(metric) or 0
252
+ hrsp_total = 0
253
for metric in METRICS:
254
stat_data['_'.join(['backend', metric, idx])] = backend.get(metric) or 0
255
+ if metric.startswith('hrsp_'):
256
+ hrsp_total += int(backend.get(metric) or 0)
257
+ if hrsp_total:
258
+ stat_data['_'.join(['backend', 'hrsp_total', idx])] = hrsp_total
259
return stat_data
260
261
def _get_info_data(self, regex):
@@ -173,12 +285,21 @@ class Service(UrlService, SocketService):
285
self.definitions['f' + metric]['lines'].append(['_'.join(['frontend', metric, idx]),
286
name, METRICS[metric]['algorithm'], 1,
287
METRICS[metric]['divisor']])
288
+ self.definitions['fhrsp_total']['lines'].append(['_'.join(['frontend', 'hrsp_total', idx]),
289
+ name, 'incremental', 1, 1])
290
for back in self.data['backend']:
291
name, idx = back['# pxname'], back['# pxname'].replace('.', '_')
292
for metric in METRICS:
293
self.definitions['b' + metric]['lines'].append(['_'.join(['backend', metric, idx]),
294
name, METRICS[metric]['algorithm'], 1,
295
METRICS[metric]['divisor']])
296
+ self.definitions['bhrsp_total']['lines'].append(['_'.join(['backend', 'hrsp_total', idx]),
297
+ name, 'incremental', 1, 1])
298
+ for metric in BACKEND_METRICS:
299
+ self.definitions['b' + metric]['lines'].append(['_'.join(['backend', metric, idx]),
300
+ name, BACKEND_METRICS[metric]['algorithm'], 1,
301
+ BACKEND_METRICS[metric]['divisor']])
302
+ self.definitions['health_sup']['lines'].append(['hsup_' + idx, name, 'absolute'])
303
self.definitions['health_sdown']['lines'].append(['hsdown_' + idx, name, 'absolute'])
304
self.definitions['health_bdown']['lines'].append(['hbdown_' + idx, name, 'absolute'])
305
@@ -210,8 +331,8 @@ def parse_data_(data):
331
return result or None
332
333
213
-def server_down(server, backend_name):
214
- return server.get('# pxname') == backend_name and server.get('status') == 'DOWN'
334
+def server_status(server, backend_name, status='DOWN'):
335
+ return server.get('# pxname') == backend_name and server.get('status') == status
336
337
338
def url_remove_params(url):