|
1
|
|
|
# -*- coding: utf-8 -*- |
|
2
|
|
|
"""Built-in Plumd Graphite Writer supporting plain text tcp/udp.""" |
|
3
|
|
|
import sys |
|
4
|
|
|
import time |
|
|
|
|
|
|
5
|
|
|
import socket |
|
6
|
|
|
import threading |
|
7
|
|
|
|
|
8
|
|
|
from plumd import Writer |
|
9
|
|
|
from plumd.util import Interval |
|
10
|
|
|
|
|
11
|
|
|
__author__ = 'Kenny Freeman' |
|
12
|
|
|
__email__ = '[email protected]' |
|
13
|
|
|
__license__ = "ISCL" |
|
14
|
|
|
__docformat__ = 'reStructuredText' |
|
15
|
|
|
|
|
16
|
|
|
|
|
17
|
|
|
PY3 = sys.version_info > (3,) |
|
18
|
|
|
|
|
19
|
|
|
|
|
20
|
|
|
class Graphite(Writer): |
|
21
|
|
|
"""Graphite TCP writer plugin.""" |
|
22
|
|
|
|
|
23
|
|
|
# todo: move things to render, writers, remove from here |
|
24
|
|
|
# also todo: print warning for queue size in render not in Graphite |
|
25
|
|
|
defaults = { |
|
26
|
|
|
'prefix': "servers", |
|
27
|
|
|
'host': '127.0.0.1', |
|
28
|
|
|
'port': 2003, |
|
29
|
|
|
'seperator': "=", # metadata seperator |
|
30
|
|
|
# any characters not in this value are removed from metrics/meta data |
|
31
|
|
|
'filter_chars': 'abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789_-', |
|
32
|
|
|
'meta_key': False, # include metadata keys in metrics |
|
33
|
|
|
'max_bytes': 1400, # send metrics in chunks of this many bytes |
|
34
|
|
|
'retry_time': 30, # delay between connection attempts |
|
35
|
|
|
'timeout': 5, # timeout for socket operations |
|
36
|
|
|
'maxqueue': 64, # maximum number of metrics to queue |
|
37
|
|
|
'warnqueue': 32, # print warning if queue size > this |
|
38
|
|
|
'tcp_ka_idle': 60, # send keepalives if idle (seconds) |
|
39
|
|
|
'tcp_ka_intvl': 5, # send unacked keepalives (seconds) |
|
40
|
|
|
'tcp_ka_count': 5 # up to this many unacked keepalives |
|
41
|
|
|
} |
|
42
|
|
|
|
|
43
|
|
|
def __init__(self, log, config): |
|
44
|
|
|
"""Graphite sender. |
|
45
|
|
|
|
|
46
|
|
|
:param log: A logger |
|
47
|
|
|
:type log: logging.RootLogger |
|
48
|
|
|
:param config: a plumd.config.Conf configuration helper instance. |
|
49
|
|
|
:type config: plumd.config.Conf |
|
50
|
|
|
""" |
|
51
|
|
|
super(Graphite, self).__init__(log, config) |
|
52
|
|
|
config.defaults(Graphite.defaults) |
|
53
|
|
|
self.addr = (config.get('host'), config.get('port')) |
|
54
|
|
|
self.socket = None |
|
55
|
|
|
self.evt_stop = threading.Event() |
|
56
|
|
|
|
|
57
|
|
|
def onstop(self): |
|
58
|
|
|
"""Set stop event to force any writes to fall through and exit.""" |
|
59
|
|
|
self.evt_stop.set() |
|
60
|
|
|
msg = "Graphite: onstop: {0}: setting stop event" |
|
61
|
|
|
self.log.info(msg.format(self.config.get('name'))) |
|
62
|
|
|
|
|
63
|
|
|
def connect(self): |
|
64
|
|
|
"""Connect to graphite and maintain a connection, return success of operation. |
|
65
|
|
|
|
|
66
|
|
|
:rtype: bool |
|
67
|
|
|
""" |
|
68
|
|
|
timeout = self.config.get('timeout') |
|
69
|
|
|
tcp_ka_idle = self.config.get('tcp_ka_idle') |
|
70
|
|
|
tcp_ka_intvl = self.config.get('tcp_ka_intvl') |
|
71
|
|
|
tcp_ka_count = self.config.get('tcp_ka_count') |
|
72
|
|
|
# disconnect if we're already connected |
|
73
|
|
|
if self.socket: |
|
74
|
|
|
self.disconnect() |
|
75
|
|
|
try: |
|
76
|
|
|
# create the socket + connection w/ timeout |
|
77
|
|
|
self.socket = socket.create_connection(self.addr, timeout=timeout) |
|
78
|
|
|
# enable tcp keepalive with more aggressive values than default |
|
79
|
|
|
self.socket.setsockopt(socket.SOL_SOCKET, socket.SO_KEEPALIVE, 1) |
|
80
|
|
|
self.socket.setsockopt(socket.SOL_TCP, socket.TCP_KEEPIDLE, tcp_ka_idle) |
|
81
|
|
|
self.socket.setsockopt(socket.SOL_TCP, socket.TCP_KEEPINTVL, tcp_ka_intvl) |
|
82
|
|
|
self.socket.setsockopt(socket.SOL_TCP, socket.TCP_KEEPCNT, tcp_ka_count) |
|
83
|
|
|
# set timeout for socket operations |
|
84
|
|
|
self.socket.settimeout(timeout) |
|
85
|
|
|
# log and return |
|
86
|
|
|
msg = "Graphite: connected: {0}" |
|
87
|
|
|
self.log.info(msg.format(self.addr)) |
|
88
|
|
|
return True |
|
89
|
|
|
except Exception as e: |
|
|
|
|
|
|
90
|
|
|
# log exception, ensure cleanup is done (disconnect) |
|
91
|
|
|
msg = "Graphite: {0}: connect: exception: {1}" |
|
92
|
|
|
self.log.error(msg.format(self.addr, e)) |
|
93
|
|
|
if self.socket: |
|
94
|
|
|
try: |
|
95
|
|
|
self.socket.close() |
|
96
|
|
|
except Exception as exc: |
|
|
|
|
|
|
97
|
|
|
self.socket = None |
|
98
|
|
|
self.socket = None |
|
99
|
|
|
return False |
|
100
|
|
|
|
|
101
|
|
|
def disconnect(self): |
|
102
|
|
|
"""Close the tcp connection to the graphite server.""" |
|
103
|
|
|
if self.socket: |
|
104
|
|
|
try: |
|
105
|
|
|
self.socket.close() |
|
106
|
|
|
self.socket = None |
|
107
|
|
|
except Exception as e: |
|
|
|
|
|
|
108
|
|
|
msg = "Graphite: disconnect: exception {0}".format(e) |
|
109
|
|
|
self.log.error(msg) |
|
110
|
|
|
self.socket = None |
|
111
|
|
|
|
|
112
|
|
|
def write(self, metrics): |
|
113
|
|
|
"""Send the already rendered metrics string to graphite with retry. |
|
114
|
|
|
|
|
115
|
|
|
:param metrics: the metrics string to send. |
|
116
|
|
|
:type metrics: str |
|
117
|
|
|
:rtype: bool |
|
118
|
|
|
""" |
|
119
|
|
|
# connect to graphite |
|
120
|
|
|
while (not self.evt_stop.is_set()) and (not self.socket) and (not self.connect()): |
|
121
|
|
|
# reconnect on an interval |
|
122
|
|
|
with Interval(self.config.get('retry_time'), self.evt_stop) as wait: |
|
123
|
|
|
msg = "re-connecting to graphite host {0}:{1} in {2:.1f} s" |
|
124
|
|
|
self.log.info(msg.format(self.addr[0], self.addr[1], wait.remaining)) |
|
125
|
|
|
# "this should never happen" |
|
126
|
|
|
if (not self.evt_stop.is_set()) and (not self.socket): |
|
127
|
|
|
msg = "graphite: socket not connected after connect, dropping metrics" |
|
128
|
|
|
self.log.error(msg) |
|
129
|
|
|
return |
|
130
|
|
|
# send metrics with retry |
|
131
|
|
|
while not self.evt_stop.is_set(): |
|
132
|
|
|
try: |
|
133
|
|
|
if PY3: |
|
134
|
|
|
self.socket.sendall(bytes(metrics, 'utf8')) |
|
135
|
|
|
else: |
|
136
|
|
|
self.socket.sendall(metrics) |
|
137
|
|
|
break |
|
138
|
|
|
except Exception as e: |
|
|
|
|
|
|
139
|
|
|
msg = "Graphite: {0}:{1}: write: exception: {2}" |
|
140
|
|
|
self.log.error(msg.format(self.addr[0], self.addr[1], e)) |
|
141
|
|
|
self.connect() |
|
142
|
|
|
|
|
143
|
|
|
def flush(self, metrics): |
|
144
|
|
|
"""Flush the metrics string to graphite, return immediately if not connected. |
|
145
|
|
|
|
|
146
|
|
|
This is used on shutdown. |
|
147
|
|
|
|
|
148
|
|
|
:param metrics: the metrics string to send. |
|
149
|
|
|
:type metrics: str |
|
150
|
|
|
:rtype: bool |
|
151
|
|
|
""" |
|
152
|
|
|
nbytes = len(metrics) |
|
153
|
|
|
msg = "Graphite: {0}:{1} flushing {2} bytes" |
|
154
|
|
|
self.log.debug(msg.format(self.addr[0], self.addr[1], nbytes)) |
|
155
|
|
|
if not self.socket: |
|
156
|
|
|
msg = "Graphite: {0}:{1} not connected, dropping metrics" |
|
157
|
|
|
self.log.debug(msg.format(self.addr[0], self.addr[1])) |
|
158
|
|
|
return |
|
159
|
|
|
try: |
|
160
|
|
|
if PY3: |
|
161
|
|
|
self.socket.sendall(bytes(metrics, 'utf8')) |
|
162
|
|
|
else: |
|
163
|
|
|
self.socket.sendall(metrics) |
|
164
|
|
|
msg = "Graphite: {0}:{1} {2} bytes flushed" |
|
165
|
|
|
self.log.debug(msg.format(self.addr[0], self.addr[1], nbytes)) |
|
166
|
|
|
except Exception as e: |
|
|
|
|
|
|
167
|
|
|
msg = "Graphite: flush: {0}:{1}: write: exception: dropping metrics: {2}" |
|
168
|
|
|
self.log.error(msg.format(self.addr[0], self.addr[1], e)) |
|
169
|
|
|
|
|
170
|
|
|
|