Add downtime recording rules for moving time windows

Co-authored-by: Wesley Bermbach <[email protected]> Co-authored-by: Istvan Zoltan Ballok <[email protected]>
gardener-attic · May 11, 2022 · a5ddcb5 · a5ddcb5
1 parent f4bfb9a
commit a5ddcb5
Showing 1 changed file with 80 additions and 0 deletions.
diff --git a/charts/connectivity-exporter/rules/error-budget.yaml b/charts/connectivity-exporter/rules/error-budget.yaml
@@ -90,6 +90,7 @@ groups:
 
   - record: downtime
     labels:
+      kind: anchored
       in: hour
     expr: |
       sum without(hour) (
@@ -99,6 +100,7 @@ groups:
 
   - record: downtime
     labels:
+      kind: anchored
       in: day
     expr: |
       sum without(day) (
@@ -108,6 +110,7 @@ groups:
 
   - record: downtime
     labels:
+      kind: anchored
       in: week
     expr: |
       sum without(week) (
@@ -117,6 +120,7 @@ groups:
 
   - record: downtime
     labels:
+      kind: anchored
       in: calendarmonth
     expr: |
       sum without(year, month) (
@@ -128,6 +132,82 @@ groups:
           (service:sli_seconds:total{year!="", month!=""})
       ))
 
+  - record: service:sli_seconds:moving:total
+    expr: |
+      service:sli_seconds
+      +
+      (
+        max_over_time(
+          service:sli_seconds:moving:total[10m]
+        )
+        or
+        service:sli_seconds * 0
+      )
+
+  - record: downtime
+    labels:
+      kind: moving
+      in: hour
+    expr: |
+      (
+        service:sli_seconds:moving:total
+        -
+        (
+          service:sli_seconds:moving:total offset 1h
+          or
+          service:sli_seconds:moving:total * 0
+        )
+      )
+      / 60 / 60
+
+  - record: downtime
+    labels:
+      kind: moving
+      in: day
+    expr: |
+      (
+        service:sli_seconds:moving:total
+        -
+        (
+          service:sli_seconds:moving:total offset 1d
+          or
+          service:sli_seconds:moving:total * 0
+        )
+      )
+      / 60 / 60 / 24
+
+  - record: downtime
+    labels:
+      kind: moving
+      in: week
+    expr: |
+      (
+        service:sli_seconds:moving:total
+        -
+        (
+          service:sli_seconds:moving:total offset 1w
+          or
+          service:sli_seconds:moving:total * 0
+        )
+      )
+      / 60 / 60 / 24 / 7
+
+  - record: downtime
+    labels:
+      kind: moving
+      in: four_weeks
+    expr: |
+      (
+        service:sli_seconds:moving:total
+        -
+        (
+          service:sli_seconds:moving:total offset 4w
+          or
+          service:sli_seconds:moving:total * 0
+        )
+      )
+      / 60 / 60 / 24 / 7 / 4
+
   - record: downtime:in
     expr: count by (in) (downtime) * 0