summaryrefslogtreecommitdiffstats
path: root/health/health.d/boinc.conf
diff options
context:
space:
mode:
Diffstat (limited to 'health/health.d/boinc.conf')
-rw-r--r--health/health.d/boinc.conf62
1 files changed, 62 insertions, 0 deletions
diff --git a/health/health.d/boinc.conf b/health/health.d/boinc.conf
new file mode 100644
index 000000000..43c588db6
--- /dev/null
+++ b/health/health.d/boinc.conf
@@ -0,0 +1,62 @@
+# Alarms for various BOINC issues.
+
+# Warn on any compute errors encountered.
+template: boinc_compute_errors
+ on: boinc.states
+ os: *
+ hosts: *
+families: *
+ lookup: average -10m unaligned of comperror
+ units: tasks
+ every: 1m
+ warn: $this > 0
+ crit: $this > 1
+ delay: up 1m down 5m multiplier 1.5 max 1h
+ info: the total number of compute errors over the past 10 minutes
+ to: sysadmin
+
+# Warn on lots of upload errors
+template: boinc_upload_errors
+ on: boinc.states
+ os: *
+ hosts: *
+families: *
+ lookup: average -10m unaligned of upload_failed
+ units: tasks
+ every: 1m
+ warn: $this > 0
+ crit: $this > 1
+ delay: up 1m down 5m multiplier 1.5 max 1h
+ info: the average number of failed uploads over the past 10 minutes
+ to: sysadmin
+
+# Warn on the task queue being empty
+template: boinc_total_tasks
+ on: boinc.tasks
+ os: *
+ hosts: *
+families: *
+ lookup: average -10m unaligned of total
+ units: tasks
+ every: 1m
+ warn: $this < 1
+ crit: $this < 0.1
+ delay: up 5m down 10m multiplier 1.5 max 1h
+ info: the total number of locally available tasks
+ to: sysadmin
+
+# Warn on no active tasks with a non-empty queue
+template: boinc_active_tasks
+ on: boinc.tasks
+ os: *
+ hosts: *
+families: *
+ lookup: average -10m unaligned of active
+ calc: ($boinc_total_tasks >= 1) ? ($this) : (inf)
+ units: tasks
+ every: 1m
+ warn: $this < 1
+ crit: $this < 0.1
+ delay: up 5m down 10m multiplier 1.5 max 1h
+ info: the total number of active tasks
+ to: sysadmin