custom longhorn storage class

clusters/base/infra/longhorn/longhorn-system/helmrelease.yaml

@@ -61,10 +61,9 @@ spec:
       # чтобы не разъезжались реплики при 3 нодах
       replicaSoftAntiAffinity: false
     persistence:
-      defaultClass: true
-      defaultClassReplicaCount: 3
-      defaultFsType: ext4
-      reclaimPolicy: Retain
+      # the default "longhorn" StorageClass lives in ../storage-classes/longhorn.yaml -
+      # the chart hardcodes staleReplicaTimeout: "30" and offers no value to override it
+      createStorageClass: false
     csi:
       attacherReplicaCount: 3
       provisionerReplicaCount: 3

clusters/base/infra/longhorn/storage-classes/kustomization.yaml

@@ -1,5 +1,6 @@
 apiVersion: kustomize.config.k8s.io/v1beta1
 kind: Kustomization
 resources:
+  - longhorn.yaml
   - longhorn-no-retain.yaml
   - longhorn-single-replica.yaml

clusters/base/infra/longhorn/storage-classes/longhorn-no-retain.yaml

@@ -4,9 +4,11 @@ metadata:
   name: longhorn-no-retain
   annotations:
     storageclass.kubernetes.io/is-default-class: "false"
+    # StorageClass parameters are immutable - let Flux recreate the object on change
+    kustomize.toolkit.fluxcd.io/force: enabled
 provisioner: driver.longhorn.io
 reclaimPolicy: Delete
 allowVolumeExpansion: true
 parameters:
   numberOfReplicas: "3"
-  staleReplicaTimeout: "30"
+  staleReplicaTimeout: "2880"

clusters/base/infra/longhorn/storage-classes/longhorn-single-replica.yaml

@@ -4,9 +4,11 @@ metadata:
   name: longhorn-single-replica
   annotations:
     storageclass.kubernetes.io/is-default-class: "false"
+    # StorageClass parameters are immutable - let Flux recreate the object on change
+    kustomize.toolkit.fluxcd.io/force: enabled
 provisioner: driver.longhorn.io
 reclaimPolicy: Retain
 allowVolumeExpansion: true
 parameters:
   numberOfReplicas: "1"
-  staleReplicaTimeout: "30"
+  staleReplicaTimeout: "2880"

clusters/base/infra/longhorn/storage-classes/longhorn.yaml

@@ -0,0 +1,27 @@
+# Default Longhorn class. Managed here instead of by the chart (persistence.createStorageClass: false
+# in the HelmRelease) because the chart hardcodes staleReplicaTimeout: "30".
+# Parameters otherwise mirror what chart 1.12.1 rendered.
+apiVersion: storage.k8s.io/v1
+kind: StorageClass
+metadata:
+  name: longhorn
+  annotations:
+    storageclass.kubernetes.io/is-default-class: "true"
+    # StorageClass parameters are immutable - let Flux recreate the object on change
+    kustomize.toolkit.fluxcd.io/force: enabled
+provisioner: driver.longhorn.io
+reclaimPolicy: Retain
+allowVolumeExpansion: true
+volumeBindingMode: Immediate
+parameters:
+  numberOfReplicas: "3"
+  # minutes a failed replica is kept for reuse before it is discarded and fully rebuilt;
+  # 30 was shorter than a planned node shutdown and caused a ~1.2TB full rebuild of kube-3
+  staleReplicaTimeout: "1440"
+  fromBackup: ""
+  fsType: ext4
+  dataLocality: disabled
+  unmapMarkSnapChainRemoved: ignored
+  disableRevisionCounter: "true"
+  dataEngine: v1
+  backupTargetName: default