| 14 |
|
|
| 15 |
|
# for computing machines |
| 16 |
|
group.computing=Host.129.12.4.*; |
| 17 |
+ |
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 |
+ |
group.compsoc=Host.compsoc1.ukc.ac.uk |
| 22 |
+ |
|
| 23 |
|
config.computing=computing.conf |
| 24 |
+ |
config.rocks=rocks.conf |
| 25 |
+ |
config.compsoc1=rocks.conf |
| 26 |
|
|
| 27 |
|
# for library machines |
| 28 |
|
group.library=Host.129.12.58.*; |
| 45 |
|
|
| 46 |
|
# The time interval at which the Queue's should be monitored |
| 47 |
|
Queue.MonitorInterval=15 |
| 48 |
+ |
# The upper limit for a Queue |
| 49 |
+ |
Queue.SizeLimit=1000 |
| 50 |
+ |
# The removal algorithm |
| 51 |
+ |
Queue.RemoveAlgorithm=FIRST |
| 52 |
|
|
| 53 |
|
|
| 42 |
– |
|
| 54 |
|
############################################################ |
| 55 |
|
#### FilterManager configuration |
| 56 |
|
############################################################ |
| 58 |
|
# The default port to listen for new host connections |
| 59 |
|
FilterManager.listenPort=4567 |
| 60 |
|
|
| 61 |
+ |
# Access control list |
| 62 |
+ |
FilterManager.ACL=DEFAULT:DENY;*.ukc.ac.uk:ALLOW;129.12.*:ALLOW |
| 63 |
|
|
| 64 |
|
|
| 65 |
+ |
|
| 66 |
|
############################################################ |
| 67 |
|
#### Filter configuration |
| 68 |
|
############################################################ |
| 72 |
|
Filter.TCPListenPort=4589 |
| 73 |
|
|
| 74 |
|
# Plugin Information |
| 75 |
< |
Filter.PluginsPackage=uk.ac.ukc.iscream.filter.plugins |
| 76 |
< |
Filter.Plugins=TypeChecker;EnforceEssentialData |
| 75 |
> |
Filter.PluginsPackage=uk.org.iscream.cms.server.filter.plugins |
| 76 |
> |
Filter.Plugins=TypeChecker;EnforceEssentialData;SourceChecker;KeyChecker |
| 77 |
|
|
| 78 |
|
# The name for the root filter |
| 79 |
|
RootFilter.name=root |
| 81 |
|
# The default parent filter (should almost always be the same as Filter.rootFilter) |
| 82 |
|
Filter.parentFilter=root |
| 83 |
|
|
| 84 |
+ |
# The input methods to activate |
| 85 |
+ |
# (comment or set to 0 to disable) |
| 86 |
+ |
Filter.ActivateTCPReader=1 |
| 87 |
+ |
Filter.ActivateUDPReader=1 |
| 88 |
+ |
Filter.ActivateCORBAReader=1 |
| 89 |
|
|
| 90 |
+ |
# Access control list for the TCP and UDP inputs |
| 91 |
+ |
Filter.TCPACL=DEFAULT:DENY;*.ukc.ac.uk:ALLOW;129.12.*:ALLOW |
| 92 |
+ |
Filter.UDPACL=DEFAULT:DENY;*.ukc.ac.uk:ALLOW;129.12.*:ALLOW |
| 93 |
|
|
| 94 |
+ |
# Access control list for the SourceChecker plugin |
| 95 |
+ |
# - packets not permitted by this ACL will be filtered |
| 96 |
+ |
Filter.SourceCheckerPluginACL=DEFAULT:DENY;*.ukc.ac.uk:ALLOW;129.12.*:ALLOW |
| 97 |
+ |
|
| 98 |
+ |
# Enforce host authentication? |
| 99 |
+ |
# (comment or set to 0 to disable) |
| 100 |
+ |
Filter.EnforceHostAuth=1 |
| 101 |
+ |
|
| 102 |
+ |
# Length of keys to use in authentication |
| 103 |
+ |
Filter.KeyLength=15 |
| 104 |
+ |
|
| 105 |
+ |
|
| 106 |
+ |
|
| 107 |
|
############################################################ |
| 108 |
|
#### Root filter interfaces configuration |
| 109 |
|
############################################################ |
| 121 |
|
|
| 122 |
|
# The port for the client interface to listen on |
| 123 |
|
ClientInterface.listenPort=4510 |
| 124 |
+ |
# The name for the client interface |
| 125 |
+ |
ClientInterface.name=realtimeclients |
| 126 |
|
|
| 127 |
+ |
# Access Control List for TCP Clients |
| 128 |
+ |
ClientInterface.TCPControlChannelACL=DEFAULT:DENY;*.ukc.ac.uk:ALLOW;129.12.*:ALLOW |
| 129 |
+ |
ClientInterface.TCPDataChannelACL=DEFAULT:DENY;*.ukc.ac.uk:ALLOW;129.12.*:ALLOW |
| 130 |
|
|
| 131 |
|
|
| 132 |
+ |
|
| 133 |
|
############################################################ |
| 134 |
+ |
#### Database interface configuration |
| 135 |
+ |
############################################################ |
| 136 |
+ |
|
| 137 |
+ |
# The name for the Database interface |
| 138 |
+ |
DBInterface.name=database |
| 139 |
+ |
|
| 140 |
+ |
|
| 141 |
+ |
|
| 142 |
+ |
############################################################ |
| 143 |
|
#### Host Configuration |
| 144 |
|
############################################################ |
| 145 |
|
|
| 146 |
|
# The default filter for a new host |
| 147 |
< |
Host.filter=filter1 |
| 147 |
> |
Host.filter=computingFilter; |
| 148 |
|
|
| 149 |
|
# The default update times for a new host |
| 150 |
|
Host.UDPUpdateTime=10 |
| 151 |
|
Host.TCPUpdateTime=60 |
| 102 |
– |
Host.AveragerUpdateTime=5 |
| 152 |
|
|
| 153 |
|
# The services checks to run on a host |
| 154 |
< |
Host.serviceChecksPackage=uk.ac.ukc.iscream.filter.plugins |
| 106 |
< |
Host.serviceChecks=HTTP;SMTP |
| 154 |
> |
Host.serviceChecksPackage=uk.org.iscream.cms.server.filter.plugins |
| 155 |
|
|
| 156 |
|
|
| 157 |
|
|
| 159 |
|
#### Monitor Configuration |
| 160 |
|
############################################################ |
| 161 |
|
|
| 162 |
< |
# the monitors to run |
| 163 |
< |
Monitor.PluginsPackage=uk.ac.ukc.iscream.client.monitors |
| 164 |
< |
Monitor.Plugins=CPU; |
| 162 |
> |
## The monitors to run |
| 163 |
> |
Monitor.PluginsPackage=uk.org.iscream.cms.server.client.monitors |
| 164 |
> |
Monitor.Plugins=CPU;Load;Process;Disk;Memory;Swap;Services;\ |
| 165 |
> |
Heartbeat;Queue;UserCount;WebFeeder; |
| 166 |
|
|
| 167 |
< |
# CPU monitor configuration |
| 168 |
< |
# threshold values |
| 167 |
> |
## Default alert timeouts |
| 168 |
> |
# these can be override on a per-monitor basis |
| 169 |
> |
# timings of the alerts (in seconds) |
| 170 |
> |
Monitor.alertTimeout.NOTICE=60 |
| 171 |
> |
Monitor.alertTimeout.WARNING=900 |
| 172 |
> |
Monitor.alertTimeout.CAUTION=1800 |
| 173 |
> |
Monitor.alertTimeout.CRITICAL=3600 |
| 174 |
> |
|
| 175 |
> |
## Threshold values |
| 176 |
|
# |
| 177 |
|
# note: if the UPPER threshold is broken, |
| 178 |
|
# the alertTimeout values are halved. |
| 179 |
|
# This is to escalate the alert quicker. |
| 180 |
+ |
# |
| 181 |
+ |
|
| 182 |
+ |
# CPU monitor threshold values |
| 183 |
|
Monitor.CPU.threshold.LOWER=90 |
| 184 |
|
Monitor.CPU.threshold.UPPER=99 |
| 185 |
|
|
| 127 |
– |
# timings of the alerts (in seconds) |
| 128 |
– |
Monitor.CPU.alertTimeout.NOTICE=60 |
| 129 |
– |
Monitor.CPU.alertTimeout.WARNING=900 |
| 130 |
– |
Monitor.CPU.alertTimeout.CAUTION=1800 |
| 131 |
– |
Monitor.CPU.alertTimeout.CRITICAL=3600 |
| 186 |
|
|
| 187 |
+ |
# Load monitor threshold values |
| 188 |
+ |
Monitor.Load.threshold.LOWER=10.0 |
| 189 |
+ |
Monitor.Load.threshold.UPPER=20.0 |
| 190 |
|
|
| 191 |
|
|
| 192 |
+ |
# Process Count threshold values |
| 193 |
+ |
Monitor.Process.threshold.LOWER=1000 |
| 194 |
+ |
Monitor.Process.threshold.UPPER=1500 |
| 195 |
+ |
|
| 196 |
+ |
|
| 197 |
+ |
# Memory monitor threshold values |
| 198 |
+ |
Monitor.Memory.threshold.LOWER=80 |
| 199 |
+ |
Monitor.Memory.threshold.UPPER=90 |
| 200 |
+ |
# Whether to include 'cache' in the 'free' value |
| 201 |
+ |
# (1 is true, anything else is false) |
| 202 |
+ |
Monitor.Memory.useCacheAsFree=1 |
| 203 |
+ |
|
| 204 |
+ |
|
| 205 |
+ |
# Swap monitor threshold values |
| 206 |
+ |
Monitor.Swap.threshold.LOWER=80 |
| 207 |
+ |
Monitor.Swap.threshold.UPPER=90 |
| 208 |
+ |
|
| 209 |
+ |
|
| 210 |
+ |
# UserCount monitor threshold values |
| 211 |
+ |
Monitor.UserCount.threshold.LOWER=100 |
| 212 |
+ |
Monitor.UserCount.threshold.UPPER=200 |
| 213 |
+ |
|
| 214 |
+ |
|
| 215 |
+ |
# Disk monitor threshold values |
| 216 |
+ |
Monitor.Disk.threshold.LOWER=90 |
| 217 |
+ |
Monitor.Disk.threshold.UPPER=95 |
| 218 |
+ |
# Thresholds given are PERCENTAGE or VALUE |
| 219 |
+ |
# where PERCENTAGE is a percentage of disk in use |
| 220 |
+ |
# and VALUE is an absolute value of space free |
| 221 |
+ |
Monitor.Disk.thresholdMeasure=PERCENTAGE |
| 222 |
+ |
|
| 223 |
+ |
|
| 224 |
+ |
# Queue monitor threshold values |
| 225 |
+ |
Monitor.Queue.threshold.LOWER=50 |
| 226 |
+ |
Monitor.Queue.threshold.UPPER=80 |
| 227 |
+ |
# Thresholds given are PERCENTAGE or VALUE |
| 228 |
+ |
# where PERCENTAGE is a percentage of disk in use |
| 229 |
+ |
# and VALUE is an absolute value of space free |
| 230 |
+ |
Monitor.Queue.thresholdMeasure=PERCENTAGE |
| 231 |
+ |
|
| 232 |
+ |
|
| 233 |
+ |
# Services monitor threshold values |
| 234 |
+ |
# |
| 235 |
+ |
# THE SERVICES MONITOR USES A HACK |
| 236 |
+ |
# WHEREBY IT USES THE SERVICE STATUS |
| 237 |
+ |
# AS THE THRESHOLD LEVEL. THIS SHOULD |
| 238 |
+ |
# BE LOOKED AT! |
| 239 |
+ |
# |
| 240 |
+ |
# how many times we should be at the highest level before |
| 241 |
+ |
# we go to a FINAL alert |
| 242 |
+ |
Monitor.Services.reachFINALcount=5 |
| 243 |
+ |
|
| 244 |
+ |
|
| 245 |
+ |
# Heartbeat threshold values |
| 246 |
+ |
# these are values in seconds from when the heartbeat |
| 247 |
+ |
# was *expected* to arrive (ie. last + hearbeat period) |
| 248 |
+ |
Monitor.Heartbeat.threshold.LOWER=180 |
| 249 |
+ |
Monitor.Heartbeat.threshold.UPPER=300 |
| 250 |
+ |
# this is how often we will run a "check" of heartbeats |
| 251 |
+ |
Monitor.Heartbeat.checkPeriod=50 |
| 252 |
+ |
# how many times we should be at the highest level before |
| 253 |
+ |
# we go to a FINAL alert |
| 254 |
+ |
Monitor.Heartbeat.reachFINALcount=5 |
| 255 |
+ |
# list of hosts we expect to have at startup |
| 256 |
+ |
# this ensures we notice hosts that aren't running, not |
| 257 |
+ |
# just those that have gone down recently. |
| 258 |
+ |
Monitor.Heartbeat.initialHosts=raptor.ukc.ac.uk;myrtle.ukc.ac.uk;kestrel.ukc.ac.uk |
| 259 |
+ |
|
| 260 |
+ |
|
| 261 |
+ |
|
| 262 |
|
############################################################ |
| 263 |
|
#### Alerter Configuration |
| 264 |
|
############################################################ |
| 265 |
+ |
# The following var's are understood by ALL alerters for messages |
| 266 |
+ |
# |
| 267 |
+ |
# %level% - the alert level (eg, WARNING) |
| 268 |
+ |
# %threshold% - the threshold broken (eg, LOWER) |
| 269 |
+ |
# %source% - the source of the alert (eg, raptor.ukc.ac.uk) |
| 270 |
+ |
# %value% - the value reached (eg, 95) |
| 271 |
+ |
# %thresholdValue% - the value of the threshold broken (eg, 90) |
| 272 |
+ |
# %attributeName% - the attribute that has caused the alert (eg, CPU User) |
| 273 |
+ |
# %timeTillNextAlert% - the time the next alert will be sent out |
| 274 |
+ |
# %timeSinceFirstAlert% - the time elapsed since the first alert for this problem |
| 275 |
+ |
# %timeOfFirstAlert% - the time the first alert was sent |
| 276 |
+ |
# |
| 277 |
|
|
| 278 |
|
# alerters to run |
| 279 |
< |
Alerter.PluginsPackage=uk.ac.ukc.iscream.client.alerters |
| 280 |
< |
Alerter.Plugins=EMail;IRC; |
| 279 |
> |
Alerter.PluginsPackage=uk.org.iscream.cms.server.client.alerters |
| 280 |
> |
Alerter.Plugins=EMail;IRC;WebFeeder;Logging; |
| 281 |
|
|
| 282 |
+ |
|
| 283 |
|
# EMail alerter configuration |
| 284 |
< |
Alerter.EMail.level = CAUTION |
| 284 |
> |
Alerter.EMail.level = WARNING |
| 285 |
|
Alerter.EMail.destList = dev@i-scream.org.uk |
| 286 |
|
Alerter.EMail.sender = dev@i-scream.org.uk |
| 287 |
|
Alerter.EMail.smtpServer = mercury.ukc.ac.uk |
| 288 |
|
Alerter.EMail.subject = i-scream alert: %level% alert on %source% for %attributeName% |
| 289 |
< |
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\nThe next alert (should one occur) will be sent in %timeTillNextAlert%. |
| 289 |
> |
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%. |
| 290 |
|
|
| 291 |
+ |
|
| 292 |
|
# IRC alerter configuration |
| 293 |
|
Alerter.IRC.level = OK |
| 294 |
|
Alerter.IRC.IRCServer = killigrew.ukc.ac.uk |
| 295 |
|
Alerter.IRC.IRCPort = 6667 |
| 296 |
|
Alerter.IRC.nickList = iscreamBot;_iscreamBot;i-screamBot |
| 297 |
|
Alerter.IRC.user = i-scream |
| 298 |
< |
Alerter.IRC.comment = i-scream alerting bot |
| 298 |
> |
Alerter.IRC.comment = i-scream alerting bot (based on PircBot) |
| 299 |
> |
Alerter.IRC.finger = i-scream alerting bot (based on PircBot) |
| 300 |
|
Alerter.IRC.channel = #i-scream |
| 301 |
|
Alerter.IRC.message = %level%: %attributeName% on %source% has passed %threshold%(%thresholdValue%) threshold with %value% - time till next alert (should one occur), %timeTillNextAlert% |
| 302 |
|
Alerter.IRC.reconnectDelay = 30 |
| 313 |
|
Alerter.IRC.helpCommand = help |
| 314 |
|
Alerter.IRC.rejectMessage = sorry, I don't understand your request |
| 315 |
|
|
| 316 |
+ |
|
| 317 |
+ |
# Logging alerter configuration |
| 318 |
+ |
Alerter.Logging.level = WARNING |
| 319 |
+ |
Alerter.Logging.message = %level%: %attributeName% on %source% has passed %threshold%(%thresholdValue%) threshold with %value% |
| 320 |
+ |
|
| 321 |
+ |
|
| 322 |
+ |
|
| 323 |
|
############################################################ |
| 324 |
+ |
#### WebFeeder Configuration |
| 325 |
+ |
############################################################ |
| 326 |
|
|
| 327 |
+ |
# Define these (one or both_ to turn on the WebFeeder's, |
| 328 |
+ |
# comment it to turn them off |
| 329 |
+ |
WebFeeder.latestActive = true |
| 330 |
+ |
WebFeeder.alertActive = true |
| 331 |
+ |
|
| 332 |
+ |
# Only pass on alerts past this level |
| 333 |
+ |
Alerter.WebFeeder.level = OK |
| 334 |
+ |
|
| 335 |
+ |
# Check period in seconds (for cleaning stale/old alerts) |
| 336 |
+ |
WebFeeder.checkPeriod = 120 |
| 337 |
+ |
|
| 338 |
+ |
# Delete alerts older than this, in seconds. |
| 339 |
+ |
# Files are only check every WebFeeder.checkPeriod seconds. |
| 340 |
+ |
WebFeeder.alertDeleteOlderThan = 300 |
| 341 |
+ |
|
| 342 |
+ |
# Root Path |
| 343 |
+ |
WebFeeder.rootPath = /usr/local/proj/co600_10/webroot |
| 344 |
+ |
|
| 345 |
+ |
# Latest Data |
| 346 |
+ |
WebFeeder.latestSubDir = latest |
| 347 |
+ |
WebFeeder.latestFileName = latest_data |
| 348 |
+ |
|
| 349 |
+ |
# Alert Data |
| 350 |
+ |
WebFeeder.alertSubDir = alert |
| 351 |
+ |
WebFeeder.alertFileName = alert_data |
| 352 |
+ |
|
| 353 |
+ |
|
| 354 |
+ |
############################################################ |
