Prevent Ceph PgsUnclean alert because of backfill

priteau · priteau · commit b3a01c141c89 · 2024-05-07T22:22:05.000+02:00
A Ceph cluster is often busy backfilling, such as when using the Ceph
balancer or adding new OSDs. It is not helpful to raise the PgsUnclean
alert because of this activity.
diff --git a/etc/kayobe/kolla/config/prometheus/ceph.rules b/etc/kayobe/kolla/config/prometheus/ceph.rules
@@ -117,7 +117,7 @@ groups:
         requests.
 
   - alert: PgsUnclean
-    expr: ceph_pool_metadata * on(pool_id,instance) group_left() (ceph_pg_total - ceph_pg_clean) > 0
+    expr: ceph_pool_metadata * on(pool_id,instance) group_left() (ceph_pg_total - ceph_pg_clean - ceph_pg_backfilling - ceph_pg_backfill_wait) > 0
     for: 15m
     labels:
       severity: warning
diff --git a/releasenotes/notes/improve-ceph-pgs-unclean-alert-98306d397344b572.yaml b/releasenotes/notes/improve-ceph-pgs-unclean-alert-98306d397344b572.yaml
@@ -0,0 +1,6 @@
+---
+fixes:
+  - |
+    Prevents raising a Ceph ``PgsUnclean`` alert because of backfilling which
+    can frequently happen because of normal rebalancing activities, such as use
+    of the Ceph balancer or OSD addition.
