root / tools / munin-node-from-hell / muninnode-from-hell @ 8589c6df
Historique | Voir | Annoter | Télécharger (13,1 ko)
| 1 |
#!/usr/bin/python |
|---|---|
| 2 |
# .- coding: utf-8 -. |
| 3 |
# |
| 4 |
# Artificial munin node that behaves in all the ways you would like |
| 5 |
# ordinary nodes _not_ to behave. |
| 6 |
# |
| 7 |
# Intended use is for designing and debugging munin-server poller to handle |
| 8 |
# such problems. |
| 9 |
# |
| 10 |
# See the file MIT-LICENSE for licensing information. |
| 11 |
# |
| 12 |
# Copyright (C) 2011 Karstensen IT |
| 13 |
# Written by Lasse Karstensen <lasse.karstensen@gmail.com>, Dec 2011. |
| 14 |
|
| 15 |
import os, sys, time, random |
| 16 |
import socket |
| 17 |
import threading |
| 18 |
import SocketServer |
| 19 |
import ConfigParser |
| 20 |
|
| 21 |
VERSION = "muninnode-from-hell v0.1" |
| 22 |
modules = {}
|
| 23 |
|
| 24 |
class MuninPlugin: |
| 25 |
def __init__(self): |
| 26 |
self.current_load = None |
| 27 |
self.current_locks = None |
| 28 |
|
| 29 |
def sleep_fetch(self, conf): |
| 30 |
period = None |
| 31 |
if conf.get("mode") == "sleepy" and conf.get("sleepyness"):
|
| 32 |
period = float(conf.get("sleepyness"))
|
| 33 |
if conf.get("mode") == "exp" and conf.get("lambd"):
|
| 34 |
period = random.expovariate(1 / float(conf.get("lambd")))
|
| 35 |
|
| 36 |
if period: |
| 37 |
#print "will sleep %.3f seconds" % period |
| 38 |
time.sleep(period) |
| 39 |
|
| 40 |
def sleep_config(self, conf): |
| 41 |
return self.sleep_fetch(conf) |
| 42 |
|
| 43 |
def find_load(self): |
| 44 |
# At about a thousand node instances you get this: |
| 45 |
#IOError: [Errno 24] Too many open files: '/proc/loadavg' |
| 46 |
# cache it for a bit.. |
| 47 |
if (not self.current_load) or random.randint(0,100) == 1: |
| 48 |
load = open("/proc/loadavg", "r").read()
|
| 49 |
load, rest = load.split(" ", 1)
|
| 50 |
self.current_load = float(load) |
| 51 |
return self.current_load |
| 52 |
|
| 53 |
def find_locks(self): |
| 54 |
if (not self.current_locks) or random.randint(0,100) == 1: |
| 55 |
fp = open("/proc/locks", "r")
|
| 56 |
self.current_locks = len(fp.readlines()) |
| 57 |
return self.current_locks |
| 58 |
|
| 59 |
class load(MuninPlugin): |
| 60 |
def fetch(self, conf): |
| 61 |
self.sleep_fetch(conf) |
| 62 |
return "load.value %.2f" % self.find_load() |
| 63 |
|
| 64 |
def config(self, conf): |
| 65 |
self.sleep_config(conf) |
| 66 |
return """graph_title Load average |
| 67 |
graph_args --base 1000 -l 0 |
| 68 |
graph_vlabel load |
| 69 |
graph_scale no |
| 70 |
graph_category system |
| 71 |
load.label load |
| 72 |
graph_info The load average of the machine describes how many processes are in the run-queue (scheduled to run "immediately"). |
| 73 |
load.info 5 minute load average """ |
| 74 |
modules["load"] = load() |
| 75 |
|
| 76 |
class locks(MuninPlugin): |
| 77 |
def fetch(self, conf): |
| 78 |
self.sleep_fetch(conf) |
| 79 |
return "locks.value %i" % self.find_locks() |
| 80 |
|
| 81 |
def config(self, conf): |
| 82 |
self.sleep_config(conf) |
| 83 |
return """graph_title Filesystem locks |
| 84 |
graph_vlabel number of locks |
| 85 |
graph_scale no |
| 86 |
graph_info This graph shows file system lock info |
| 87 |
graph_category system |
| 88 |
locks.label number of locks |
| 89 |
locks.info Number of active locks""" |
| 90 |
modules["locks"] = locks() |
| 91 |
|
| 92 |
class tarpit(MuninPlugin): |
| 93 |
"Nasty plugin that never responds" |
| 94 |
def fetch(self, conf): |
| 95 |
time.sleep(1000) |
| 96 |
|
| 97 |
def config(self, conf): |
| 98 |
time.sleep(1000) |
| 99 |
modules["tarpit"] = tarpit() |
| 100 |
|
| 101 |
class always_warning(MuninPlugin): |
| 102 |
conftext = """graph_title Always in LEVEL |
| 103 |
graph_vlabel Level |
| 104 |
graph_scale no |
| 105 |
graph_info A simple graph that is always in LEVEL |
| 106 |
graph_category always_LEVEL |
| 107 |
generic.label Level |
| 108 |
generic.info Level usually above warning level |
| 109 |
generic.warning 5 |
| 110 |
generic.critical 10""" |
| 111 |
|
| 112 |
def fetch(self, conf): |
| 113 |
return "generic.value 10" |
| 114 |
|
| 115 |
def config(self, conf): |
| 116 |
return self.conftext.replace("LEVEL","warning")
|
| 117 |
modules["always_warning"] = always_warning() |
| 118 |
|
| 119 |
class always_critical(always_warning): |
| 120 |
def fetch(self, conf): |
| 121 |
return "generic.value 20" |
| 122 |
|
| 123 |
def config(self, conf): |
| 124 |
return self.conftext.replace("LEVEL","critical")
|
| 125 |
modules["always_critical"] = always_critical() |
| 126 |
|
| 127 |
class failing_plugin(MuninPlugin): |
| 128 |
"A really broken plugin" |
| 129 |
def fetch(self, conf): |
| 130 |
return "# Bad exit" |
| 131 |
|
| 132 |
def config(self, conf): |
| 133 |
return "# Bad exit" |
| 134 |
modules["failing_plugin"] = failing_plugin() |
| 135 |
|
| 136 |
class failing_plugin2(MuninPlugin): |
| 137 |
def fetch(self, conf): |
| 138 |
return "# Bad exit" |
| 139 |
|
| 140 |
def config(self, conf): |
| 141 |
return """graph_title Config works, fetch fails |
| 142 |
graph_vlabel Level |
| 143 |
graph_category failing |
| 144 |
generic.label generic_label_here |
| 145 |
generic.info never_really_used""" |
| 146 |
modules["failing_plugin2"] = failing_plugin2() |
| 147 |
|
| 148 |
class failing_plugin3(MuninPlugin): |
| 149 |
def config(self, conf): |
| 150 |
return """graph_title A plugin with two dses but only fetch value for one |
| 151 |
graph_args --base 1000 -l 0 |
| 152 |
fivemin.label 1 minute load |
| 153 |
onemin.label 5 minute load""" |
| 154 |
def fetch(self, conf): |
| 155 |
return "onemin.value 1" |
| 156 |
modules["failing_plugin3"] = failing_plugin3() |
| 157 |
|
| 158 |
|
| 159 |
class graph_area(MuninPlugin): |
| 160 |
"A plugin that uses STACK and AREA. From proc_pri. Use: testing the grapher" |
| 161 |
def fetch(self, conf): |
| 162 |
return """high.value 3 |
| 163 |
low.value 2 |
| 164 |
locked.value 1""" |
| 165 |
|
| 166 |
def config(self, conf): |
| 167 |
return """graph_title AREA and STACK |
| 168 |
graph_order low high locked |
| 169 |
graph_category graphtest |
| 170 |
graph_info This graph shows nuber of processes at each priority |
| 171 |
graph_args --base 1000 -l 0 |
| 172 |
graph_vlabel Number of processes |
| 173 |
high.label high priority |
| 174 |
high.draw STACK |
| 175 |
high.info The number of high-priority processes (tasks) |
| 176 |
low.label low priority |
| 177 |
low.draw AREA |
| 178 |
low.info The number of low-priority processes (tasks) |
| 179 |
locked.label locked in memory |
| 180 |
locked.draw STACK |
| 181 |
locked.info The number of processes that have pages locked into memory (for real-time and custom IO) |
| 182 |
""" |
| 183 |
modules["graph_area"] = graph_area() |
| 184 |
|
| 185 |
class utf8_graphcat(MuninPlugin): |
| 186 |
"A plugin with a graph category which has UTF-8 in it" |
| 187 |
def fetch(self, conf): |
| 188 |
return "apples.value %.2f" % self.find_load() |
| 189 |
|
| 190 |
def config(self, conf): |
| 191 |
return """graph_title Example UTF-8 graph |
| 192 |
graph_vlabel apples |
| 193 |
graph_category foo™ |
| 194 |
apples.label apples |
| 195 |
graph_info Apples eaten |
| 196 |
apples.info Apples eaten""" |
| 197 |
modules["utf8_graphcat"] = utf8_graphcat() |
| 198 |
|
| 199 |
class utf8_graphname(MuninPlugin): |
| 200 |
"A plugin with a UTF-8 name" |
| 201 |
def fetch(self, conf): |
| 202 |
return "apples.value %.2f" % self.find_load() |
| 203 |
|
| 204 |
def config(self, conf): |
| 205 |
return """graph_title Example UTF-8 graph |
| 206 |
graph_vlabel apples |
| 207 |
graph_category system |
| 208 |
apples.label apples |
| 209 |
graph_info Apples eaten |
| 210 |
apples.info Apples eaten""" |
| 211 |
modules["utf8_™graphname"] = utf8_graphname() |
| 212 |
|
| 213 |
|
| 214 |
class ArgumentTCPserver(SocketServer.ThreadingTCPServer): |
| 215 |
def __init__(self, server_address, RequestHandlerClass, args): |
| 216 |
SocketServer.ThreadingTCPServer.__init__(self,server_address, RequestHandlerClass) |
| 217 |
self.args = args |
| 218 |
|
| 219 |
|
| 220 |
class MuninHandler(SocketServer.StreamRequestHandler): |
| 221 |
""" |
| 222 |
Munin server implementation. |
| 223 |
|
| 224 |
This is based on munin_node.py by Chris Holcombe / http://sourceforge.net/projects/pythonmuninnode/ |
| 225 |
|
| 226 |
Possible commands: |
| 227 |
list, nodes, config, fetch, version or quit |
| 228 |
""" |
| 229 |
|
| 230 |
def handle(self): |
| 231 |
if self.server.args.get("verbose"): print "%s: Connection from %s:%s. server args is %s" \
|
| 232 |
% (self.server.args["name"], self.client_address[0], self.client_address[1], self.server.args) |
| 233 |
# slow path |
| 234 |
hostname = self.server.args["name"] |
| 235 |
full_hostname = hostname |
| 236 |
|
| 237 |
moduleprofile = self.server.args["pluginprofile"] |
| 238 |
modulenames = set(moduleprofile) |
| 239 |
|
| 240 |
self.wfile.write("# munin node at %s\n" % hostname)
|
| 241 |
|
| 242 |
while True: |
| 243 |
line = self.rfile.readline().strip() |
| 244 |
try: |
| 245 |
cmd, args = line.split(" ", 1)
|
| 246 |
except ValueError: |
| 247 |
cmd = line |
| 248 |
args = "" |
| 249 |
|
| 250 |
if not cmd or cmd == "quit": |
| 251 |
break |
| 252 |
|
| 253 |
if cmd == "list": |
| 254 |
# List all plugins that are available |
| 255 |
self.wfile.write(" ".join(self.server.args["plugins"].keys()) + "\n")
|
| 256 |
elif cmd == "nodes": |
| 257 |
# We just support this host |
| 258 |
self.wfile.write("%s\n.\n" % full_hostname)
|
| 259 |
elif cmd == "config": |
| 260 |
# display the config information of the plugin |
| 261 |
if not self.server.args["plugins"].has_key(args): |
| 262 |
self.wfile.write("# Unknown service\n.\n" )
|
| 263 |
else: |
| 264 |
config = self.server.args["plugins"][args].config(self.server.args) |
| 265 |
if config is None: |
| 266 |
self.wfile.write("# Unknown service\n.\n")
|
| 267 |
else: |
| 268 |
self.wfile.write(config + "\n.\n") |
| 269 |
elif cmd == "fetch": |
| 270 |
# display the data information as returned by the plugin |
| 271 |
if not self.server.args["plugins"].has_key(args): |
| 272 |
self.wfile.write("# Unknown service\n.\n")
|
| 273 |
else: |
| 274 |
data = self.server.args["plugins"][args].fetch(self.server.args) |
| 275 |
if data is None: |
| 276 |
self.wfile.write("# Unknown service\n.\n")
|
| 277 |
else: |
| 278 |
self.wfile.write(data + "\n.\n") |
| 279 |
elif cmd == "version": |
| 280 |
# display the server version |
| 281 |
self.wfile.write("munin node on %s version: %s\n" %
|
| 282 |
(full_hostname, VERSION)) |
| 283 |
else: |
| 284 |
self.wfile.write("# Unknown command. Try list, nodes, " \
|
| 285 |
"config, fetch, version or quit\n") |
| 286 |
|
| 287 |
|
| 288 |
def start_servers(instances): |
| 289 |
# TODO: Listen to IPv6 |
| 290 |
HOST = "0.0.0.0" |
| 291 |
servers = {}
|
| 292 |
for iconf in instances: |
| 293 |
print "Setting up instance %s at port %s" \ |
| 294 |
% (iconf["name"], iconf["expanded_port"]) |
| 295 |
|
| 296 |
server = ArgumentTCPserver((HOST, iconf["expanded_port"]), MuninHandler, iconf) |
| 297 |
server_thread = threading.Thread(target=server.serve_forever) |
| 298 |
server_thread.daemon = True |
| 299 |
server_thread.start() |
| 300 |
|
| 301 |
servers[iconf["name"]] = server |
| 302 |
return servers |
| 303 |
|
| 304 |
|
| 305 |
|
| 306 |
def usage(): |
| 307 |
print "Usage: %s [--run] [--verbose] [--muninconf] <configfile> <configfileN>" % sys.argv[0] |
| 308 |
|
| 309 |
def main(): |
| 310 |
if len(sys.argv) <= 2: |
| 311 |
usage() |
| 312 |
sys.exit(1) |
| 313 |
|
| 314 |
verbose = False |
| 315 |
if "--verbose" in sys.argv: |
| 316 |
verbose = True |
| 317 |
|
| 318 |
config = ConfigParser.RawConfigParser() |
| 319 |
for configfile in sys.argv[1:]: |
| 320 |
if not configfile.endswith(".conf"):
|
| 321 |
continue |
| 322 |
if verbose: |
| 323 |
print "Reading config file %s" % configfile |
| 324 |
config.read(configfile) |
| 325 |
|
| 326 |
instancekeys = [ key for key in config.sections() if key.startswith("instance:") ]
|
| 327 |
servers = {}
|
| 328 |
|
| 329 |
instances = [] |
| 330 |
|
| 331 |
for key in instancekeys: |
| 332 |
instancename = key.split(":", 2)[1]
|
| 333 |
portrange = [] |
| 334 |
if config.has_option(key, "port"): |
| 335 |
portrange = [ config.getint(key, "port") ] |
| 336 |
if config.has_option(key, "portrange"): |
| 337 |
rangestr = config.get(key, "portrange") |
| 338 |
ranges = rangestr.split("-")
|
| 339 |
range_expanded = range(int(ranges[0]), int(ranges[1])+1, 1) |
| 340 |
portrange += range_expanded |
| 341 |
|
| 342 |
if len(portrange) == 0: |
| 343 |
print "WARN: No port or portrange defined for instance %s" \ |
| 344 |
% instancename |
| 345 |
|
| 346 |
pluginprofile = "pluginprofile:%s" % config.get(key, "pluginprofile") |
| 347 |
if not config.has_section(pluginprofile): |
| 348 |
print "WARN: Definition for pluginprofile %s not found, skipping" \ |
| 349 |
% config.get(key, "pluginprofile") |
| 350 |
continue |
| 351 |
|
| 352 |
plugins = {}
|
| 353 |
tentative_pluginlist = config.get(pluginprofile, "plugins").split(",")
|
| 354 |
assert(len(tentative_pluginlist) > 0) |
| 355 |
for tentative_plugin in tentative_pluginlist: |
| 356 |
tentative_plugin = tentative_plugin.strip() |
| 357 |
if not modules.has_key(tentative_plugin): |
| 358 |
print "WARN: Pluginprofile %s specifies unknown plugin %s" \ |
| 359 |
% (pluginprofile, tentative_plugin) |
| 360 |
continue |
| 361 |
|
| 362 |
# support more than one instantiation of the same plugin. |
| 363 |
plugininstancename = tentative_plugin |
| 364 |
i=2 |
| 365 |
while (plugins.has_key(plugininstancename)): |
| 366 |
plugininstancename = tentative_plugin + str(i) |
| 367 |
i += 1 |
| 368 |
|
| 369 |
plugins[plugininstancename] = modules[tentative_plugin] |
| 370 |
|
| 371 |
for portinstance in portrange: |
| 372 |
instanceconfig = dict() |
| 373 |
|
| 374 |
for k,v in config.items(key): |
| 375 |
instanceconfig[k] = v |
| 376 |
|
| 377 |
instanceconfig["plugins"] = plugins |
| 378 |
instanceconfig["verbose"] = verbose |
| 379 |
|
| 380 |
instanceconfig["name"] = "%s-%s" % (instancename, portinstance) |
| 381 |
instanceconfig["expanded_port"] = portinstance |
| 382 |
|
| 383 |
instances.append(instanceconfig) |
| 384 |
# XXX: need to store what handlers we should have. |
| 385 |
print instances |
| 386 |
|
| 387 |
# output sample munin config for the poller |
| 388 |
if "--muninconf" in sys.argv: |
| 389 |
for i in instances: |
| 390 |
print "[%s;%s]\n\taddress %s\n\tuse_node_name yes\n\tport %s\n" \ |
| 391 |
% ( "fromhell", i["name"], config.get("base","hostname"), i["port"])
|
| 392 |
|
| 393 |
|
| 394 |
if "--run" in sys.argv: |
| 395 |
if verbose: print "Starting up.." |
| 396 |
servers = start_servers(instances) |
| 397 |
|
| 398 |
try: |
| 399 |
while True: |
| 400 |
time.sleep(0.5) |
| 401 |
except KeyboardInterrupt: |
| 402 |
print "Caught Ctrl-c, shutting down.." |
| 403 |
for port, server in servers.items(): |
| 404 |
server.shutdown() |
| 405 |
sys.exit(0) |
| 406 |
|
| 407 |
if __name__ == "__main__": |
| 408 |
main() |
