lots of fixes, automation for bitfolk

[distro-setup] / filesystem / etc / prometheus / rules / iank.yml

diff --git a/filesystem/etc/prometheus/rules/iank.yml b/filesystem/etc/prometheus/rules/iank.yml
index 048edafa5f801b26b858e1c9bfd0016ee92a3a78..75b5cbcfc760925b79d51fd6c3262de7a6186a97 100644 (file)
--- a/filesystem/etc/prometheus/rules/iank.yml
+++ b/filesystem/etc/prometheus/rules/iank.yml
@@ -56,7 +56,23 @@ groups:
 # there is also the absent() function, but i didnt see a way to make that work
   - alert: mysers_units_missing
     expr: |-
-      count(up{job="node"}) by (instance) * 3 unless count(node_systemd_unit_state{name=~"(systemstatus|btrfsmaintstop|dynamicipupdate).service",state="active"}) by (instance)
+      count(up{job="node"} == 1) by (instance) * 3 unless
+      count(node_systemd_unit_state{name=~"(systemstatus|btrfsmaintstop|dynamicipupdate).service",state="active"}) by (instance)
+    for: 20m
+    labels:
+      severity: warn
+
+  - alert: epanicclean_not_active
+    expr: |-
+      node_systemd_unit_state{name="epanicclean.service",state="active"} != 1
+    for: 20m
+    labels:
+      severity: warn
+
+  - alert: epanicclean_missing
+    expr: |-
+      count(up{job=~"node|tlsnode"} == 1) by (instance) unless
+      count(node_systemd_unit_state{job=~"node|tlsnode",name="epanicclean.service",state="active"}) by (instance)
     for: 20m
     labels:
       severity: warn
@@ -74,23 +90,40 @@ groups:
     labels:
       severity: day
 
+  - alert: mailtest_check_vps
+    expr: |-
+      time() - mailtest_check_last_usec{job="tlsnode"} >= 60 * 12
+    labels:
+      severity: day
+    annotations:
+      summary: '12 minutes down'
+
+  # 42 mins: enough for a 30 min queue run plus 12
+  - alert: mailtest_check_vps
+    expr: |-
+      time() - mailtest_check_last_usec{job="tlsnode"} >= 60 * 42
+    labels:
+      severity: prod
+    annotations:
+      summary: '42 minutes down'
 
-  - alert: mailtest_check
+  - alert: mailtest_check_mailhost
     expr: |-
-      time() - mailtest_check_last_usec > 60 * 12
+      time() - max by (folder,from) (mailtest_check_last_usec{job="node"}) >= 60 * 12
     labels:
       severity: day
     annotations:
       summary: '12 minutes down'
 
   # 42 mins: enough for a 30 min queue run plus 12
-  - alert: mailtest_check
+  - alert: mailtest_check_mailhost
     expr: |-
-      time() - mailtest_check_last_usec > 60 * 42
+      time() - max by (folder,from) (mailtest_check_last_usec{job="node"}) >= 60 * 42
     labels:
       severity: prod
     annotations:
-      summary: '43 minutes down'
+      summary: '42 minutes down'
+
 
   - alert: 1pmtest
     expr: hour() == 17 and minute() < 5
@@ -101,22 +134,26 @@ groups:
       summary: Prometheus daily test alert
 
 
-
-# alternate expression, to calculate if the alert would have fired is:
+#### Inhibit notes ####
+## Example of expressions to detect if the target_down alert
+# fired in the last 24 hours. Initially, I thought his could
+# be an alert which inhibits up_resets, but eventually I figured
+# that doesn't make much sense, and the idea of using an alert
+# that is not an indication of something wrong, only inhibits another
+# alert, I think works better to integrate directly into the
+# alert it would inhibit, this may mean a recording rule. That avoids
+# an alert we have to ignore or filter out.
+#
+# Alternate expression, to calculate if the alert would have fired is:
 #  min_over_time(sum_over_time(up[30m])[1d:]) == 0
 #  where 30m matches the for: time in target_down
 #
-# sum_over_time is not needed, just convenience for graphing
-  - alert: target_down_inhibitor
-    expr: |-
-      sum_over_time(ALERTS{alertname="target_down"}[1d])
-    labels:
-      severity: ignore
-    annotations:
-      summary: alert that indicates target_down alert fired in the last day
-      description: "VALUE = {{ $value }}"
+# Note: for graphing, surround in the expression in sum_over_time()
+# ALERTS{alertname="target_down",alertstate="firing"}[1d]
+#### end Inhibit notes ####
 
-# For targets where we alert except for longer downtimes, we
+
+# For targets where we alert only on long downtimes, we
 # still want to know if it is going down many times for short times over
 # a long period of time. But ignore reboots.
 #
@@ -124,18 +161,16 @@ groups:
 # avg_over_time(node_systemd_unit_state{name="dynamicipupdate.service",state="active"}[1d]) < .95
   - alert: up_resets
     expr: |-
-      resets(up[3d]) - changes(node_boot_time_seconds[3d]) > 15
+      resets(up[2d]) - changes(node_boot_time_seconds[2d]) > 12
     labels:
       severity: warn
     annotations:
-      summary: "Target has gone down {{ $value }} times in 3 days, > 15"
-
+      summary: "Target has gone down {{ $value }} times in 2 days, > 12"
 
 
 
 # https://awesome-prometheus-alerts.grep.to/rules
 
-
 # todo, we should probably group the prometheus alerts that indicate a
 # host-local problem.
 # eg, set a label  alert-group: local-prom, then make a receiver that