tested with CLI; todo test this on actual cluster & write docs

determined-ai · Sep 18, 2024 · a64649e · a64649e
1 parent cc561b8
commit a64649e
Show file tree

Hide file tree

Showing 6 changed files with 19 additions and 17 deletions.
diff --git a/helm/charts/determined/Chart.yaml b/helm/charts/determined/Chart.yaml
@@ -9,4 +9,4 @@ home: https://github.com/determined-ai/determined.git
 # If using a non-release version (e.g., X.Y.Z.dev0) you will have to specify an
 # existing official release version (e.g., X.Y.Z) or specify a commit has
 # that has been publicly published (all commits from main).
-appVersion: "0.36.0"
+appVersion: "0.36.1-dev0"
diff --git a/helm/charts/determined/templates/db-persistent-volume-claim.yaml b/helm/charts/determined/templates/db-persistent-volume-claim.yaml
@@ -4,8 +4,8 @@ apiVersion: v1
 kind: PersistentVolumeClaim
 metadata:
   name: 
-    {{ if and .Release.IsUpgrade .Values.db.claimName }}
-    determined-db-pvc-{{ .Values.db.claimName }}
+    {{ if and .Release.IsUpgrade .Values.db.restoreSnapshotName }}
+    determined-db-pvc-{{ .Values.db.restoreSnapshotName }}
     {{- else -}}
     determined-db-pvc-{{ .Release.Name }}
     {{ end }}

diff --git a/helm/charts/determined/templates/db-volume-snapshot-class.yaml b/helm/charts/determined/templates/db-volume-snapshot-class.yaml
@@ -5,4 +5,4 @@ metadata:
   name: determined-db-snapshot-class
 driver: pd.csi.storage.gke.io
 deletionPolicy: Delete
-{{ end }}
+{{ end }}
diff --git a/helm/charts/determined/templates/db-volume-snapshot.yaml b/helm/charts/determined/templates/db-volume-snapshot.yaml
@@ -11,9 +11,9 @@ spec:
   volumeSnapshotClassName: determined-db-snapshot-class
   source:
     persistentVolumeClaimName: 
-    {{ if .Values.db.claimName }}
-    determined-db-pvc-{{ .Values.db.claimName }}
-    {{- else -}}
-    determined-db-pvc-{{ .Release.Name }}
-    {{ end }}
-{{ end }}
+      {{ if .Values.db.claimName }}
+      determined-db-pvc-{{ .Values.db.claimName }}
+      {{- else -}}
+      determined-db-pvc-{{ .Release.Name }}
+      {{ end }}
+{{ end }}
diff --git a/helm/charts/determined/values.yaml b/helm/charts/determined/values.yaml
@@ -203,12 +203,14 @@ db:
 
   # snapshotName refers to the volume snapshot name, a backup of the database's persistent volume.
   # If defined, Helm will "snapshot" the DB upon its next upgrade, creating a volumeSnapshot called
-  # determined-db-snapsnot-{snapshotName}.
-  # snapshotName:
+  # determined-db-snapsnot-{snapshotName}. This CANNOT be the name of a PV that
+  # already exists, like the {{Release.Name}}.
+  # snapshotName: determined1
 
   # restoreSnapshotName refers to the volume snapshot name which you wish to restore the DB to.
-  # This only applies during a Helm upgrade, and if defined.
-  # restoreSnapshotName: 
+  # This only applies during a Helm upgrade, and if defined. This CANNOT be the name of a PVC that
+  # already exists, like the {{Release.Name}}.
+  # restoreSnapshotName:
 
 # checkpointStorage controls where checkpoints are stored. Supported types include `shared_fs`,
 # `gcs`, and `s3`.
@@ -375,7 +377,7 @@ resourcePools:
 # defaultComputeResourcePool: default
 
 ## Configure the initial user password for the cluster
-initialUserPassword: Abcd1234
+# initialUserPassword:
 
 # additional_resource_managers:
 #   - resource_manager:

diff --git a/master/internal/logretention/logretention.go b/master/internal/logretention/logretention.go
@@ -92,7 +92,7 @@ func DeleteExpiredTaskLogs(ctx context.Context, days *int16) (int64, error) {
 	if days != nil {
 		defaultLogRetentionDays = *days
 	}
-	log.WithField("default-retention-days", defaultLogRetentionDays).Info("deleting expired task logs")
+	log.WithField("default-retention-days", defaultLogRetentionDays).Trace("deleting expired task logs")
 	r, err := db.Bun().NewRaw(fmt.Sprintf(`
 		WITH log_retention_tasks AS (
 			SELECT COALESCE(r.log_retention_days, %d) as log_retention_days, t.task_id, t.end_time
@@ -112,6 +112,6 @@ func DeleteExpiredTaskLogs(ctx context.Context, days *int16) (int64, error) {
 		return 0, errors.Wrap(err, "error deleting expired task logs")
 	}
 	rows, err := r.RowsAffected()
-	log.WithFields(logrus.Fields{"rows": rows, "err": err}).Info("deleted expired task logs")
+	log.WithFields(logrus.Fields{"rows": rows, "err": err}).Trace("deleted expired task logs")
 	return rows, err
 }