mirror of
https://github.com/alerta/alerta.git
synced 2025-01-24 17:29:39 +00:00
48 lines
1.4 KiB
YAML
48 lines
1.4 KiB
YAML
# Server reboots within last 2 hours
|
|
- event: ServerReboot
|
|
filter: # all
|
|
value: $now - $boottime
|
|
thresholdInfo: [ 'INFORM:<:7200' ]
|
|
text: [ 'Server rebooted less than 2 hours ago' ]
|
|
resource: $host
|
|
group: Hardware
|
|
graphs: [ ]
|
|
tags: [ 'owner:websys' ]
|
|
|
|
# Puppet Last Run alert
|
|
- event: PuppetLastRun
|
|
filter: # all
|
|
value: $now - $pup_last_run
|
|
thresholdInfo: [ 'MAJOR:>:7200','NORMAL:<:7200' ]
|
|
text: ['Puppet agent has not run for at least 2 hours (last puppet run at $pup_last_run)','Puppet agent is OK (last puppet run at $pup_last_run).']
|
|
resource: $host
|
|
group: Puppet
|
|
count: 2
|
|
repeat: 10
|
|
graphs: [ ]
|
|
tags: [ 'owner:websys' ]
|
|
|
|
# Puppet Resource alert
|
|
- event: PuppetResFailed
|
|
filter: # all
|
|
value: $pup_res_failed
|
|
thresholdInfo: [ 'WARNING:>:0','NORMAL:==:0' ]
|
|
text: [ 'Puppet agent failed to update $pup_res_failed out of $pup_res_total resources','Puppet is updating all $pup_res_total resources' ]
|
|
resource: $host
|
|
group: Puppet
|
|
count: 2
|
|
repeat: 10
|
|
graphs: [ 'pup_res_failed' ]
|
|
tags: [ 'owner:websys' ]
|
|
|
|
# Ganglia Heartbeat alert
|
|
- event: GangliaHeartbeat
|
|
filter: # all
|
|
value: $heartbeat
|
|
thresholdInfo: [ 'CRITICAL:>:90','NORMAL:<:90' ]
|
|
text: [ 'No heartbeat from Ganglia agent for at least 90 seconds','Heartbeat from Ganglia agent OK' ]
|
|
resource: $host
|
|
group: Ganglia
|
|
repeat: 10
|
|
graphs: [ ]
|
|
tags: [ 'owner:websys' ]
|