103 |
|
|
104 |
|
# The services checks to run on a host |
105 |
|
Host.serviceChecksPackage=uk.ac.ukc.iscream.filter.plugins |
106 |
< |
Host.serviceChecks=HTTP;SMTP |
106 |
> |
Host.serviceChecks=HTTP;SMTP;POP3;SSH;POP3;IMAP |
107 |
|
|
108 |
|
|
109 |
|
|
113 |
|
|
114 |
|
# the monitors to run |
115 |
|
Monitor.PluginsPackage=uk.ac.ukc.iscream.client.monitors |
116 |
< |
Monitor.Plugins=CPU;Heartbeat |
116 |
> |
Monitor.Plugins=CPU;Memory;Swap;Heartbeat; |
117 |
|
|
118 |
+ |
|
119 |
|
## CPU monitor configuration |
120 |
|
# threshold values |
121 |
|
# |
131 |
|
Monitor.CPU.alertTimeout.CAUTION=1800 |
132 |
|
Monitor.CPU.alertTimeout.CRITICAL=3600 |
133 |
|
|
134 |
+ |
|
135 |
+ |
## Memory monitor configuration |
136 |
+ |
# threshold values |
137 |
+ |
# |
138 |
+ |
# note: if the UPPER threshold is broken, |
139 |
+ |
# the alertTimeout values are halved. |
140 |
+ |
# This is to escalate the alert quicker. |
141 |
+ |
Monitor.Memory.threshold.LOWER=80 |
142 |
+ |
Monitor.Memory.threshold.UPPER=90 |
143 |
+ |
|
144 |
+ |
# timings of the alerts (in seconds) |
145 |
+ |
Monitor.Memory.alertTimeout.NOTICE=60 |
146 |
+ |
Monitor.Memory.alertTimeout.WARNING=900 |
147 |
+ |
Monitor.Memory.alertTimeout.CAUTION=1800 |
148 |
+ |
Monitor.Memory.alertTimeout.CRITICAL=3600 |
149 |
+ |
|
150 |
+ |
|
151 |
+ |
## Swap monitor configuration |
152 |
+ |
# threshold values |
153 |
+ |
# |
154 |
+ |
# note: if the UPPER threshold is broken, |
155 |
+ |
# the alertTimeout values are halved. |
156 |
+ |
# This is to escalate the alert quicker. |
157 |
+ |
Monitor.Swap.threshold.LOWER=80 |
158 |
+ |
Monitor.Swap.threshold.UPPER=90 |
159 |
+ |
|
160 |
+ |
# timings of the alerts (in seconds) |
161 |
+ |
Monitor.Swap.alertTimeout.NOTICE=60 |
162 |
+ |
Monitor.Swap.alertTimeout.WARNING=900 |
163 |
+ |
Monitor.Swap.alertTimeout.CAUTION=1800 |
164 |
+ |
Monitor.Swap.alertTimeout.CRITICAL=3600 |
165 |
+ |
|
166 |
+ |
|
167 |
|
## Heartbeat monitor configuration |
168 |
|
# threshold values |
169 |
|
# |
171 |
|
# the alertTimeout values are halved. |
172 |
|
# This is to escalate the alert quicker. |
173 |
|
Monitor.Heartbeat.checkPeriod=50 |
174 |
+ |
# these are values in seconds from when the heartbeat |
175 |
+ |
# was *expected* to arrive (ie. last + hearbeat period) |
176 |
|
Monitor.Heartbeat.threshold.LOWER=120 |
177 |
|
Monitor.Heartbeat.threshold.LOWER=300 |
178 |
|
|