|
1
|
|
|
from datetime import datetime |
|
2
|
|
|
import logging |
|
3
|
|
|
from typing import Iterator |
|
4
|
|
|
|
|
5
|
|
|
from prometheus_client.core import InfoMetricFamily, StateSetMetricFamily, GaugeMetricFamily, Metric |
|
6
|
|
|
|
|
7
|
|
|
from requests import get |
|
8
|
|
|
from requests.compat import urljoin |
|
9
|
|
|
from requests.exceptions import ConnectionError, Timeout, HTTPError, RequestException |
|
10
|
|
|
|
|
11
|
|
|
from helpers import TimestampGaugeMetricFamily, TimestampCounterMetricFamily |
|
12
|
|
|
|
|
13
|
|
|
|
|
14
|
|
|
logger = logging.getLogger(__name__) |
|
15
|
|
|
|
|
16
|
|
|
|
|
17
|
|
|
def datestring_to_timestamp(date_str: str) -> float: |
|
18
|
|
|
return datetime.strptime(date_str, '%Y-%m-%dT%H:%M:%S.%f%z').timestamp() |
|
19
|
|
|
|
|
20
|
|
|
|
|
21
|
|
|
def _status(status: dict) -> (StateSetMetricFamily, GaugeMetricFamily): |
|
22
|
|
|
status_dict = {state: state == status['overall']['state'] for state in ['red', 'yellow', 'green']} |
|
23
|
|
|
since = datestring_to_timestamp(status['overall']['since']) |
|
24
|
|
|
status = StateSetMetricFamily('kibana_status', 'Kibana Status', value=status_dict) |
|
25
|
|
|
since = GaugeMetricFamily('kibana_status_since', 'Last change of status, in seconds since epoch', value=since) |
|
26
|
|
|
return status, since |
|
27
|
|
|
|
|
28
|
|
|
|
|
29
|
|
|
class Metrics(object): |
|
30
|
|
|
def __init__(self, metrics_dict: dict): |
|
31
|
|
|
self._timestamp = datestring_to_timestamp(metrics_dict['last_updated']) |
|
32
|
|
|
self._metrics_dict = metrics_dict |
|
33
|
|
|
|
|
34
|
|
|
def __iter__(self): |
|
35
|
|
|
yield from self._response_times() |
|
36
|
|
|
yield from self._requests() |
|
37
|
|
|
yield from self._process() |
|
38
|
|
|
yield from self._os() |
|
39
|
|
|
|
|
40
|
|
|
def _os(self) -> Iterator[Metric]: |
|
41
|
|
|
os_dict = self._metrics_dict['os'] |
|
42
|
|
|
|
|
43
|
|
|
yield from (TimestampGaugeMetricFamily('kibana_os_load_%s' % key, |
|
|
|
|
|
|
44
|
|
|
'Kibana OS load %s' % key, |
|
45
|
|
|
value=value, |
|
|
|
|
|
|
46
|
|
|
timestamp=self._timestamp) for key, value in os_dict['load'].items()) |
|
47
|
|
|
|
|
48
|
|
|
yield from (TimestampGaugeMetricFamily('kibana_os_memory_%s_bytes' % key.split('_')[0], |
|
49
|
|
|
'Kibana %s OS memory' % key.split('_')[0], |
|
50
|
|
|
value=value, |
|
51
|
|
|
timestamp=self._timestamp) for key, value in os_dict['memory'].items()) |
|
52
|
|
|
|
|
53
|
|
|
yield TimestampCounterMetricFamily('kibana_os_uptime_seconds', |
|
54
|
|
|
'Kibana OS uptime in seconds', |
|
55
|
|
|
value=os_dict['uptime_in_millis'] / 1000, |
|
56
|
|
|
timestamp=self._timestamp) |
|
57
|
|
|
|
|
58
|
|
|
def _response_times(self) -> Iterator[Metric]: |
|
59
|
|
|
rt_dict = self._metrics_dict['response_times'] |
|
60
|
|
|
|
|
61
|
|
|
yield TimestampGaugeMetricFamily('kibana_response_time_max_seconds', |
|
62
|
|
|
'Kibana maximum response time in seconds', |
|
63
|
|
|
value=rt_dict['max_in_millis'] / 1000, |
|
64
|
|
|
timestamp=self._timestamp) |
|
65
|
|
|
|
|
66
|
|
|
# Kibana statistics lib can sometimes return NaN for this value. |
|
67
|
|
|
# If that is the case, this is set to 0 in order to avoid gaps in the time series. |
|
68
|
|
|
# Reference: https://github.com/elastic/kibana/blob/6.7/src/server/status/lib/metrics.js#L73 |
|
69
|
|
|
# NaN is converted to `undefined` which then has the whole field removed from the response JSON |
|
70
|
|
|
yield TimestampGaugeMetricFamily('kibana_response_time_avg_seconds', |
|
71
|
|
|
'Kibana average response time in seconds', |
|
72
|
|
|
value=rt_dict.setdefault('avg_in_millis', 0) / 1000, |
|
73
|
|
|
timestamp=self._timestamp) |
|
74
|
|
|
|
|
75
|
|
|
def _requests(self) -> Iterator[Metric]: |
|
76
|
|
|
req_dict = self._metrics_dict['requests'] |
|
77
|
|
|
yield TimestampGaugeMetricFamily('kibana_requests_total', |
|
78
|
|
|
'Total requests serviced', |
|
79
|
|
|
value=req_dict['total'], |
|
80
|
|
|
timestamp=self._timestamp) |
|
81
|
|
|
|
|
82
|
|
|
yield TimestampGaugeMetricFamily('kibana_requests_disconnects', |
|
83
|
|
|
'Total requests disconnected', |
|
84
|
|
|
value=req_dict['disconnects'], |
|
85
|
|
|
timestamp=self._timestamp) |
|
86
|
|
|
|
|
87
|
|
|
per_status = TimestampGaugeMetricFamily('kibana_requests', |
|
88
|
|
|
'Total requests by status code', |
|
89
|
|
|
labels=['status_code'], |
|
90
|
|
|
timestamp=self._timestamp) |
|
91
|
|
|
|
|
92
|
|
|
for code, count in req_dict['status_codes'].items(): |
|
93
|
|
|
per_status.add_metric(labels=[code], value=count) |
|
94
|
|
|
|
|
95
|
|
|
yield per_status |
|
96
|
|
|
|
|
97
|
|
|
def _process(self) -> Iterator[Metric]: |
|
98
|
|
|
process_dict = self._metrics_dict['process'] |
|
99
|
|
|
|
|
100
|
|
|
yield TimestampGaugeMetricFamily('kibana_process_memory_heap_total_bytes', |
|
101
|
|
|
'Total heap size in bytes', |
|
102
|
|
|
value=process_dict['memory']['heap']['total_in_bytes'], |
|
103
|
|
|
timestamp=self._timestamp) |
|
104
|
|
|
yield TimestampGaugeMetricFamily('kibana_process_memory_heap_used_bytes', |
|
105
|
|
|
'Used heap size in bytes', |
|
106
|
|
|
value=process_dict['memory']['heap']['used_in_bytes'], |
|
107
|
|
|
timestamp=self._timestamp) |
|
108
|
|
|
|
|
109
|
|
|
yield TimestampGaugeMetricFamily('kibana_process_memory_heap_size_limit_bytes', |
|
110
|
|
|
'Heap size limit in bytes', |
|
111
|
|
|
value=process_dict['memory']['heap']['size_limit'], |
|
112
|
|
|
timestamp=self._timestamp) |
|
113
|
|
|
|
|
114
|
|
|
yield TimestampGaugeMetricFamily('kibana_process_memory_resident_set_size_bytes', |
|
115
|
|
|
'Memory resident set size', |
|
116
|
|
|
value=process_dict['memory']['resident_set_size_in_bytes'], |
|
117
|
|
|
timestamp=self._timestamp) |
|
118
|
|
|
|
|
119
|
|
|
yield TimestampCounterMetricFamily('kibana_process_uptime_seconds', |
|
120
|
|
|
'Kibana process uptime in seconds', |
|
121
|
|
|
value=process_dict['uptime_in_millis'] / 1000, |
|
122
|
|
|
timestamp=self._timestamp) |
|
123
|
|
|
|
|
124
|
|
|
|
|
125
|
|
|
class KibanaCollector(object): |
|
126
|
|
|
def __init__(self, host: str, path: str = '/api/status', kibana_login: str = None, kibana_password: str = None): |
|
127
|
|
|
self._url = urljoin(host, path) |
|
128
|
|
|
self._kibana_login = kibana_login |
|
129
|
|
|
self._kibana_password = kibana_password |
|
130
|
|
|
|
|
131
|
|
|
def _fetch_stats(self) -> dict: |
|
132
|
|
|
if self._kibana_login: |
|
133
|
|
|
auth = (self._kibana_login, self._kibana_password) |
|
134
|
|
|
else: |
|
135
|
|
|
auth = None |
|
136
|
|
|
r = get(self._url, auth=auth) |
|
137
|
|
|
r.raise_for_status() |
|
138
|
|
|
return r.json() |
|
139
|
|
|
|
|
140
|
|
|
def collect(self): |
|
141
|
|
|
kibana_up = GaugeMetricFamily('kibana_node_reachable', 'Kibana node was reached', value=0) |
|
142
|
|
|
try: |
|
143
|
|
|
stats = self._fetch_stats() |
|
144
|
|
|
except ConnectionError as e: |
|
145
|
|
|
logger.warning('Got a connection error while trying to contact Kibana:\n%s' % e) |
|
146
|
|
|
except Timeout as e: |
|
147
|
|
|
logger.warning('Got a timeout while trying to contact Kibana:\n%s' % e) |
|
148
|
|
|
except HTTPError as e: |
|
149
|
|
|
logger.warning('Got a HTTP error %s while trying to contact Kibana:\n%s' % (e.response.status_code, e)) |
|
150
|
|
|
except RequestException as e: |
|
151
|
|
|
logger.warning('Got a RequestException while trying to contact Kibana:\n%s' % e) |
|
152
|
|
|
else: |
|
153
|
|
|
kibana_up = GaugeMetricFamily('kibana_node_reachable', 'Kibana node was reached', value=1) |
|
154
|
|
|
yield InfoMetricFamily('kibana_version', 'Kibana Version', value=stats['version']) |
|
155
|
|
|
yield from _status(stats['status']) |
|
156
|
|
|
yield from Metrics(stats['metrics']) |
|
157
|
|
|
finally: |
|
158
|
|
|
yield kibana_up |
|
159
|
|
|
|
vladvasiliu /
kibana-prometheus-exporter-py
Loading history...