1 |
tdb |
1.2 |
# This is the default system configuration file |
2 |
|
|
# used by all components of the i-scream system |
3 |
|
|
# |
4 |
|
|
# ALL configuration that uses the centralised |
5 |
|
|
# configuration system has this file as its |
6 |
|
|
# root of configuration |
7 |
|
|
# |
8 |
tdb |
1.59 |
# $Author: tdb $ |
9 |
tdb |
1.69 |
# $Id: system.conf,v 1.68 2002/10/12 21:56:37 tdb Exp $ |
10 |
ajm |
1.14 |
# |
11 |
|
|
############################################################ |
12 |
|
|
#### Configuration Files & Groups |
13 |
|
|
############################################################ |
14 |
|
|
|
15 |
|
|
# for computing machines |
16 |
tdb |
1.15 |
group.computing=Host.129.12.4.*; |
17 |
tdb |
1.41 |
group.rocks=Host.agate.ukc.ac.uk;Host.arkose.ukc.ac.uk;Host.basalt.ukc.ac.uk;Host.chalk.ukc.ac.uk;\ |
18 |
|
|
Host.chert.ukc.ac.uk;Host.granite.ukc.ac.uk;Host.jade.ukc.ac.uk;Host.jasper.ukc.ac.uk;\ |
19 |
|
|
Host.magnetite.ukc.ac.uk;Host.obsidian.ukc.ac.uk;Host.pumice.ukc.ac.uk;\ |
20 |
|
|
Host.pyrite.ukc.ac.uk;Host.slate.ukc.ac.uk;Host.topaz.ukc.ac.uk; |
21 |
tdb |
1.42 |
group.compsoc=Host.compsoc1.ukc.ac.uk |
22 |
tdb |
1.41 |
|
23 |
tdb |
1.15 |
config.computing=computing.conf |
24 |
tdb |
1.41 |
config.rocks=rocks.conf |
25 |
tdb |
1.42 |
config.compsoc1=rocks.conf |
26 |
ajm |
1.14 |
|
27 |
|
|
# for library machines |
28 |
tdb |
1.15 |
group.library=Host.129.12.58.*; |
29 |
ajm |
1.14 |
config.library=library.conf |
30 |
|
|
|
31 |
tdb |
1.15 |
# for SSB machines |
32 |
|
|
group.ssb=Host.stue*.ukc.ac.uk; |
33 |
|
|
config.ssb=ssb.conf |
34 |
|
|
|
35 |
ajm |
1.14 |
# mySQL configuration |
36 |
|
|
config.mySQL=mySQL.conf |
37 |
|
|
|
38 |
ajm |
1.16 |
############################################################ |
39 |
|
|
#### Misc system-wide configuration |
40 |
|
|
############################################################ |
41 |
|
|
|
42 |
|
|
# the time (in seconds) that checks should be made for |
43 |
|
|
# changes in configuration |
44 |
|
|
ConfigurationProxy.updateTime=60 |
45 |
|
|
|
46 |
|
|
# The time interval at which the Queue's should be monitored |
47 |
|
|
Queue.MonitorInterval=15 |
48 |
tdb |
1.40 |
# The upper limit for a Queue |
49 |
|
|
Queue.SizeLimit=1000 |
50 |
|
|
# The removal algorithm |
51 |
|
|
Queue.RemoveAlgorithm=FIRST |
52 |
tdb |
1.2 |
|
53 |
tdb |
1.68 |
# the time (in seconds) at which the XMLCache's contents |
54 |
|
|
# will be rotated. in real terms, this means the minimum |
55 |
|
|
# age at which an XMLPacket can be before it is deleted |
56 |
|
|
# from the cache. |
57 |
|
|
XMLCache.cleanupPeriod=30 |
58 |
ajm |
1.14 |
|
59 |
|
|
############################################################ |
60 |
|
|
#### FilterManager configuration |
61 |
|
|
############################################################ |
62 |
tdb |
1.2 |
|
63 |
|
|
# The default port to listen for new host connections |
64 |
|
|
FilterManager.listenPort=4567 |
65 |
|
|
|
66 |
tdb |
1.63 |
# Access control list |
67 |
|
|
FilterManager.ACL=DEFAULT:DENY;*.ukc.ac.uk:ALLOW;129.12.*:ALLOW |
68 |
|
|
|
69 |
ajm |
1.14 |
|
70 |
|
|
|
71 |
|
|
############################################################ |
72 |
|
|
#### Filter configuration |
73 |
|
|
############################################################ |
74 |
|
|
|
75 |
tdb |
1.2 |
# The default port for Filter's to listen on |
76 |
|
|
Filter.UDPListenPort=4589 |
77 |
|
|
Filter.TCPListenPort=4589 |
78 |
|
|
|
79 |
ajm |
1.14 |
# Plugin Information |
80 |
tdb |
1.55 |
Filter.PluginsPackage=uk.org.iscream.cms.server.filter.plugins |
81 |
tdb |
1.65 |
Filter.Plugins=TypeChecker;EnforceEssentialData;SourceChecker;KeyChecker |
82 |
ajm |
1.14 |
|
83 |
tdb |
1.2 |
# The name for the root filter |
84 |
|
|
RootFilter.name=root |
85 |
ajm |
1.14 |
|
86 |
|
|
# The default parent filter (should almost always be the same as Filter.rootFilter) |
87 |
|
|
Filter.parentFilter=root |
88 |
|
|
|
89 |
tdb |
1.61 |
# The input methods to activate |
90 |
|
|
# (comment or set to 0 to disable) |
91 |
|
|
Filter.ActivateTCPReader=1 |
92 |
|
|
Filter.ActivateUDPReader=1 |
93 |
|
|
Filter.ActivateCORBAReader=1 |
94 |
tdb |
1.62 |
|
95 |
|
|
# Access control list for the TCP and UDP inputs |
96 |
|
|
Filter.TCPACL=DEFAULT:DENY;*.ukc.ac.uk:ALLOW;129.12.*:ALLOW |
97 |
|
|
Filter.UDPACL=DEFAULT:DENY;*.ukc.ac.uk:ALLOW;129.12.*:ALLOW |
98 |
tdb |
1.64 |
|
99 |
|
|
# Access control list for the SourceChecker plugin |
100 |
|
|
# - packets not permitted by this ACL will be filtered |
101 |
|
|
Filter.SourceCheckerPluginACL=DEFAULT:DENY;*.ukc.ac.uk:ALLOW;129.12.*:ALLOW |
102 |
tdb |
1.65 |
|
103 |
|
|
# Enforce host authentication? |
104 |
|
|
# (comment or set to 0 to disable) |
105 |
|
|
Filter.EnforceHostAuth=1 |
106 |
tdb |
1.66 |
|
107 |
|
|
# Length of keys to use in authentication |
108 |
|
|
Filter.KeyLength=15 |
109 |
tdb |
1.62 |
|
110 |
ajm |
1.14 |
|
111 |
|
|
|
112 |
|
|
############################################################ |
113 |
|
|
#### Root filter interfaces configuration |
114 |
|
|
############################################################ |
115 |
|
|
|
116 |
tdb |
1.2 |
# Comment either of these (or delete them) and |
117 |
|
|
# that interface won't be started |
118 |
|
|
RootFilter.realtimeInterfaceName=realtimeclients |
119 |
tdb |
1.69 |
#RootFilter.dbInterfaceName=database |
120 |
tdb |
1.2 |
|
121 |
ajm |
1.14 |
|
122 |
|
|
|
123 |
|
|
############################################################ |
124 |
|
|
#### Client interface configuration |
125 |
|
|
############################################################ |
126 |
|
|
|
127 |
tdb |
1.2 |
# The port for the client interface to listen on |
128 |
|
|
ClientInterface.listenPort=4510 |
129 |
tdb |
1.52 |
# The name for the client interface |
130 |
|
|
ClientInterface.name=realtimeclients |
131 |
tdb |
1.63 |
|
132 |
|
|
# Access Control List for TCP Clients |
133 |
|
|
ClientInterface.TCPControlChannelACL=DEFAULT:DENY;*.ukc.ac.uk:ALLOW;129.12.*:ALLOW |
134 |
|
|
ClientInterface.TCPDataChannelACL=DEFAULT:DENY;*.ukc.ac.uk:ALLOW;129.12.*:ALLOW |
135 |
tdb |
1.52 |
|
136 |
|
|
|
137 |
|
|
|
138 |
|
|
############################################################ |
139 |
|
|
#### Database interface configuration |
140 |
|
|
############################################################ |
141 |
|
|
|
142 |
|
|
# The name for the Database interface |
143 |
|
|
DBInterface.name=database |
144 |
tdb |
1.2 |
|
145 |
ajm |
1.14 |
|
146 |
|
|
|
147 |
|
|
############################################################ |
148 |
|
|
#### Host Configuration |
149 |
|
|
############################################################ |
150 |
tdb |
1.2 |
|
151 |
|
|
# The default filter for a new host |
152 |
tdb |
1.36 |
Host.filter=computingFilter; |
153 |
tdb |
1.13 |
|
154 |
tdb |
1.2 |
# The default update times for a new host |
155 |
tdb |
1.7 |
Host.UDPUpdateTime=10 |
156 |
tdb |
1.2 |
Host.TCPUpdateTime=60 |
157 |
|
|
|
158 |
ajm |
1.8 |
# The services checks to run on a host |
159 |
tdb |
1.55 |
Host.serviceChecksPackage=uk.org.iscream.cms.server.filter.plugins |
160 |
tdb |
1.2 |
|
161 |
ajm |
1.14 |
|
162 |
tdb |
1.12 |
|
163 |
ajm |
1.14 |
############################################################ |
164 |
|
|
#### Monitor Configuration |
165 |
|
|
############################################################ |
166 |
|
|
|
167 |
tdb |
1.47 |
## The monitors to run |
168 |
tdb |
1.55 |
Monitor.PluginsPackage=uk.org.iscream.cms.server.client.monitors |
169 |
tdb |
1.51 |
Monitor.Plugins=CPU;Load;Process;Disk;Memory;Swap;Services;\ |
170 |
|
|
Heartbeat;Queue;UserCount;WebFeeder; |
171 |
tdb |
1.24 |
|
172 |
tdb |
1.47 |
## Default alert timeouts |
173 |
|
|
# these can be override on a per-monitor basis |
174 |
|
|
# timings of the alerts (in seconds) |
175 |
|
|
Monitor.alertTimeout.NOTICE=60 |
176 |
|
|
Monitor.alertTimeout.WARNING=900 |
177 |
|
|
Monitor.alertTimeout.CAUTION=1800 |
178 |
|
|
Monitor.alertTimeout.CRITICAL=3600 |
179 |
ajm |
1.14 |
|
180 |
tdb |
1.47 |
## Threshold values |
181 |
ajm |
1.16 |
# |
182 |
|
|
# note: if the UPPER threshold is broken, |
183 |
|
|
# the alertTimeout values are halved. |
184 |
|
|
# This is to escalate the alert quicker. |
185 |
tdb |
1.47 |
# |
186 |
|
|
|
187 |
|
|
# CPU monitor threshold values |
188 |
ajm |
1.16 |
Monitor.CPU.threshold.LOWER=90 |
189 |
|
|
Monitor.CPU.threshold.UPPER=99 |
190 |
|
|
|
191 |
tdb |
1.44 |
|
192 |
tdb |
1.47 |
# Load monitor threshold values |
193 |
tdb |
1.44 |
Monitor.Load.threshold.LOWER=10.0 |
194 |
|
|
Monitor.Load.threshold.UPPER=20.0 |
195 |
|
|
|
196 |
|
|
|
197 |
tdb |
1.47 |
# Process Count threshold values |
198 |
tdb |
1.44 |
Monitor.Process.threshold.LOWER=1000 |
199 |
|
|
Monitor.Process.threshold.UPPER=1500 |
200 |
|
|
|
201 |
tdb |
1.24 |
|
202 |
tdb |
1.47 |
# Memory monitor threshold values |
203 |
tdb |
1.24 |
Monitor.Memory.threshold.LOWER=80 |
204 |
|
|
Monitor.Memory.threshold.UPPER=90 |
205 |
tdb |
1.67 |
# Whether to include 'cache' in the 'free' value |
206 |
|
|
# (1 is true, anything else is false) |
207 |
|
|
Monitor.Memory.useCacheAsFree=1 |
208 |
tdb |
1.24 |
|
209 |
|
|
|
210 |
tdb |
1.47 |
# Swap monitor threshold values |
211 |
tdb |
1.24 |
Monitor.Swap.threshold.LOWER=80 |
212 |
|
|
Monitor.Swap.threshold.UPPER=90 |
213 |
tdb |
1.50 |
|
214 |
|
|
|
215 |
|
|
# UserCount monitor threshold values |
216 |
|
|
Monitor.UserCount.threshold.LOWER=100 |
217 |
|
|
Monitor.UserCount.threshold.UPPER=200 |
218 |
tdb |
1.24 |
|
219 |
ajm |
1.33 |
|
220 |
tdb |
1.47 |
# Disk monitor threshold values |
221 |
tdb |
1.42 |
Monitor.Disk.threshold.LOWER=90 |
222 |
|
|
Monitor.Disk.threshold.UPPER=95 |
223 |
tdb |
1.47 |
# Thresholds given are PERCENTAGE or VALUE |
224 |
|
|
# where PERCENTAGE is a percentage of disk in use |
225 |
|
|
# and VALUE is an absolute value of space free |
226 |
tdb |
1.46 |
Monitor.Disk.thresholdMeasure=PERCENTAGE |
227 |
ajm |
1.33 |
|
228 |
tdb |
1.46 |
|
229 |
tdb |
1.56 |
# Queue monitor threshold values |
230 |
tdb |
1.49 |
Monitor.Queue.threshold.LOWER=50 |
231 |
|
|
Monitor.Queue.threshold.UPPER=80 |
232 |
tdb |
1.47 |
# Thresholds given are PERCENTAGE or VALUE |
233 |
|
|
# where PERCENTAGE is a percentage of disk in use |
234 |
|
|
# and VALUE is an absolute value of space free |
235 |
tdb |
1.46 |
Monitor.Queue.thresholdMeasure=PERCENTAGE |
236 |
|
|
|
237 |
ajm |
1.34 |
|
238 |
tdb |
1.47 |
# Services monitor threshold values |
239 |
ajm |
1.34 |
# |
240 |
|
|
# THE SERVICES MONITOR USES A HACK |
241 |
|
|
# WHEREBY IT USES THE SERVICE STATUS |
242 |
|
|
# AS THE THRESHOLD LEVEL. THIS SHOULD |
243 |
|
|
# BE LOOKED AT! |
244 |
|
|
# |
245 |
tdb |
1.47 |
# how many times we should be at the highest level before |
246 |
|
|
# we go to a FINAL alert |
247 |
tdb |
1.38 |
Monitor.Services.reachFINALcount=5 |
248 |
|
|
|
249 |
ajm |
1.14 |
|
250 |
tdb |
1.47 |
# Heartbeat threshold values |
251 |
tdb |
1.25 |
# these are values in seconds from when the heartbeat |
252 |
|
|
# was *expected* to arrive (ie. last + hearbeat period) |
253 |
tdb |
1.48 |
Monitor.Heartbeat.threshold.LOWER=180 |
254 |
tdb |
1.38 |
Monitor.Heartbeat.threshold.UPPER=300 |
255 |
tdb |
1.47 |
# this is how often we will run a "check" of heartbeats |
256 |
tdb |
1.48 |
Monitor.Heartbeat.checkPeriod=50 |
257 |
tdb |
1.47 |
# how many times we should be at the highest level before |
258 |
|
|
# we go to a FINAL alert |
259 |
|
|
Monitor.Heartbeat.reachFINALcount=5 |
260 |
tdb |
1.58 |
# list of hosts we expect to have at startup |
261 |
|
|
# this ensures we notice hosts that aren't running, not |
262 |
|
|
# just those that have gone down recently. |
263 |
|
|
Monitor.Heartbeat.initialHosts=raptor.ukc.ac.uk;myrtle.ukc.ac.uk;kestrel.ukc.ac.uk |
264 |
tdb |
1.47 |
|
265 |
ajm |
1.14 |
|
266 |
ajm |
1.23 |
|
267 |
ajm |
1.14 |
############################################################ |
268 |
|
|
#### Alerter Configuration |
269 |
|
|
############################################################ |
270 |
ajm |
1.23 |
# The following var's are understood by ALL alerters for messages |
271 |
|
|
# |
272 |
|
|
# %level% - the alert level (eg, WARNING) |
273 |
|
|
# %threshold% - the threshold broken (eg, LOWER) |
274 |
|
|
# %source% - the source of the alert (eg, raptor.ukc.ac.uk) |
275 |
|
|
# %value% - the value reached (eg, 95) |
276 |
|
|
# %thresholdValue% - the value of the threshold broken (eg, 90) |
277 |
|
|
# %attributeName% - the attribute that has caused the alert (eg, CPU User) |
278 |
|
|
# %timeTillNextAlert% - the time the next alert will be sent out |
279 |
|
|
# %timeSinceFirstAlert% - the time elapsed since the first alert for this problem |
280 |
|
|
# %timeOfFirstAlert% - the time the first alert was sent |
281 |
|
|
# |
282 |
ajm |
1.14 |
|
283 |
|
|
# alerters to run |
284 |
tdb |
1.55 |
Alerter.PluginsPackage=uk.org.iscream.cms.server.client.alerters |
285 |
tdb |
1.54 |
Alerter.Plugins=EMail;IRC;WebFeeder;Logging; |
286 |
tdb |
1.47 |
|
287 |
ajm |
1.14 |
|
288 |
|
|
# EMail alerter configuration |
289 |
tdb |
1.38 |
Alerter.EMail.level = WARNING |
290 |
|
|
Alerter.EMail.destList = dev@i-scream.org.uk |
291 |
ajm |
1.14 |
Alerter.EMail.sender = dev@i-scream.org.uk |
292 |
|
|
Alerter.EMail.smtpServer = mercury.ukc.ac.uk |
293 |
ajm |
1.16 |
Alerter.EMail.subject = i-scream alert: %level% alert on %source% for %attributeName% |
294 |
tdb |
1.38 |
Alerter.EMail.message = The i-scream distributed central monitoring system has\nraised a %level% alert for the host %source%.\n\nThe value for %attributeName% of %value% has exceeded the\n%threshold% threshold value of %thresholdValue%.\n\nThis alert was originally raised at %timeOfFirstAlert%,\nwhich was %timeSinceFirstAlert% ago.\n\nThe next alert (should one occur) will be sent in %timeTillNextAlert%. |
295 |
|
|
|
296 |
ajm |
1.14 |
|
297 |
|
|
# IRC alerter configuration |
298 |
ajm |
1.16 |
Alerter.IRC.level = OK |
299 |
ajm |
1.14 |
Alerter.IRC.IRCServer = killigrew.ukc.ac.uk |
300 |
|
|
Alerter.IRC.IRCPort = 6667 |
301 |
tdb |
1.19 |
Alerter.IRC.nickList = iscreamBot;_iscreamBot;i-screamBot |
302 |
tdb |
1.17 |
Alerter.IRC.user = i-scream |
303 |
tdb |
1.59 |
Alerter.IRC.comment = i-scream alerting bot (based on PircBot) |
304 |
|
|
Alerter.IRC.finger = i-scream alerting bot (based on PircBot) |
305 |
ajm |
1.14 |
Alerter.IRC.channel = #i-scream |
306 |
ajm |
1.16 |
Alerter.IRC.message = %level%: %attributeName% on %source% has passed %threshold%(%thresholdValue%) threshold with %value% - time till next alert (should one occur), %timeTillNextAlert% |
307 |
tdb |
1.17 |
Alerter.IRC.reconnectDelay = 30 |
308 |
|
|
Alerter.IRC.startupNotice = i-scream alerting bot activated |
309 |
|
|
Alerter.IRC.stopCommand = stop alerts |
310 |
|
|
Alerter.IRC.startCommand = start alerts |
311 |
|
|
Alerter.IRC.lastAlertCommand = last alert |
312 |
|
|
Alerter.IRC.joinCommand = join |
313 |
tdb |
1.18 |
Alerter.IRC.nickChangeCommand = nick |
314 |
|
|
Alerter.IRC.statCommand = statistics |
315 |
|
|
Alerter.IRC.uptimeCommand = uptime |
316 |
|
|
Alerter.IRC.timeSinceLastAlertCommand = time since last alert |
317 |
tdb |
1.19 |
Alerter.IRC.versionCommand = version |
318 |
tdb |
1.17 |
Alerter.IRC.helpCommand = help |
319 |
|
|
Alerter.IRC.rejectMessage = sorry, I don't understand your request |
320 |
tdb |
1.54 |
|
321 |
|
|
|
322 |
|
|
# Logging alerter configuration |
323 |
|
|
Alerter.Logging.level = WARNING |
324 |
tdb |
1.57 |
Alerter.Logging.message = %level%: %attributeName% on %source% has passed %threshold%(%thresholdValue%) threshold with %value% |
325 |
tdb |
1.29 |
|
326 |
tdb |
1.47 |
|
327 |
|
|
|
328 |
tdb |
1.29 |
############################################################ |
329 |
|
|
#### WebFeeder Configuration |
330 |
|
|
############################################################ |
331 |
|
|
|
332 |
tdb |
1.32 |
# Define these (one or both_ to turn on the WebFeeder's, |
333 |
|
|
# comment it to turn them off |
334 |
|
|
WebFeeder.latestActive = true |
335 |
|
|
WebFeeder.alertActive = true |
336 |
tdb |
1.43 |
|
337 |
|
|
# Only pass on alerts past this level |
338 |
tdb |
1.53 |
Alerter.WebFeeder.level = OK |
339 |
tdb |
1.37 |
|
340 |
|
|
# Check period in seconds (for cleaning stale/old alerts) |
341 |
tdb |
1.42 |
WebFeeder.checkPeriod = 120 |
342 |
tdb |
1.37 |
|
343 |
|
|
# Delete alerts older than this, in seconds. |
344 |
|
|
# Files are only check every WebFeeder.checkPeriod seconds. |
345 |
tdb |
1.42 |
WebFeeder.alertDeleteOlderThan = 300 |
346 |
tdb |
1.32 |
|
347 |
tdb |
1.29 |
# Root Path |
348 |
|
|
WebFeeder.rootPath = /usr/local/proj/co600_10/webroot |
349 |
|
|
|
350 |
|
|
# Latest Data |
351 |
|
|
WebFeeder.latestSubDir = latest |
352 |
tdb |
1.31 |
WebFeeder.latestFileName = latest_data |
353 |
tdb |
1.32 |
|
354 |
|
|
# Alert Data |
355 |
|
|
WebFeeder.alertSubDir = alert |
356 |
|
|
WebFeeder.alertFileName = alert_data |
357 |
ajm |
1.16 |
|
358 |
tdb |
1.47 |
|
359 |
tdb |
1.17 |
############################################################ |