From 25baa38762fd19be2edf1cbfbfcfa6cd0fc5e074 Mon Sep 17 00:00:00 2001
From: Scott Seago <sseago@redhat.com>
Date: Tue, 10 Mar 2026 15:41:17 -0400
Subject: [PATCH] refactor: Optimize VSC handle readiness polling for VSS
 backups

Co-authored-by: aider (gemini/gemini-2.5-pro) <aider@aider.chat>
Signed-off-by: Scott Seago <sseago@redhat.com>
---
 pkg/util/csi/volume_snapshot.go | 141 +++++++++++++++++++-------------
 1 file changed, 83 insertions(+), 58 deletions(-)

diff --git a/pkg/util/csi/volume_snapshot.go b/pkg/util/csi/volume_snapshot.go
index 4e4103efa8..58e54341f9 100644
--- a/pkg/util/csi/volume_snapshot.go
+++ b/pkg/util/csi/volume_snapshot.go
@@ -604,72 +604,97 @@ func WaitUntilVSCHandleIsReady(
 		return vsc, nil
 	}
 
-	// We'll wait 10m for the VSC to be reconciled polling
-	// every 5s unless backup's csiSnapshotTimeout is set
-	interval := 5 * time.Second
+	// We'll wait for the VSC to be reconciled, trying a fast poll interval first
+	// before falling back to a slower poll interval for the full csiSnapshotTimeout.
 	vsc := new(snapshotv1api.VolumeSnapshotContent)
+	var interval time.Duration
+
+	pollFunc := func(ctx context.Context) (bool, error) {
+		vs := new(snapshotv1api.VolumeSnapshot)
+		if err := crClient.Get(
+			ctx,
+			crclient.ObjectKeyFromObject(volSnap),
+			vs,
+		); err != nil {
+			return false,
+				errors.Wrapf(
+					err,
+					"failed to get volumesnapshot %s/%s",
+					volSnap.Namespace, volSnap.Name,
+				)
+		}
+
+		if vs.Status == nil || vs.Status.BoundVolumeSnapshotContentName == nil {
+			log.Infof("Waiting for CSI driver to reconcile volumesnapshot %s/%s. Retrying in %ds",
+				volSnap.Namespace, volSnap.Name, interval/time.Second)
+			return false, nil
+		}
+
+		if err := crClient.Get(
+			ctx,
+			crclient.ObjectKey{
+				Name: *vs.Status.BoundVolumeSnapshotContentName,
+			},
+			vsc,
+		); err != nil {
+			return false,
+				errors.Wrapf(
+					err,
+					"failed to get VolumeSnapshotContent %s for VolumeSnapshot %s/%s",
+					*vs.Status.BoundVolumeSnapshotContentName, vs.Namespace, vs.Name,
+				)
+		}
 
+		// we need to wait for the VolumeSnapshotContent
+		// to have a snapshot handle because during restore,
+		// we'll use that snapshot handle as the source for
+		// the VolumeSnapshotContent so it's statically
+		// bound to the existing snapshot.
+		if vsc.Status == nil ||
+			vsc.Status.SnapshotHandle == nil {
+			log.Infof(
+				"Waiting for VolumeSnapshotContents %s to have snapshot handle. Retrying in %ds",
+				vsc.Name, interval/time.Second)
+			if vsc.Status != nil &&
+				vsc.Status.Error != nil {
+				log.Warnf("VolumeSnapshotContent %s has error: %v",
+					vsc.Name, *vsc.Status.Error.Message)
+			}
+			return false, nil
+		}
+
+		return true, nil
+	}
+
+	// The short interval for the first ten seconds is due to the fact that
+	// Microsoft VSS backups have a hard-coded unfreeze call after 10 seconds,
+	// so we need to minimize waiting time during the first 10 seconds.
+	// First poll with a short interval and timeout.
+	interval = 1 * time.Second
+	timeout := 10 * time.Second
 	err := wait.PollUntilContextTimeout(
 		context.Background(),
 		interval,
-		csiSnapshotTimeout,
+		timeout,
 		true,
-		func(ctx context.Context) (bool, error) {
-			vs := new(snapshotv1api.VolumeSnapshot)
-			if err := crClient.Get(
-				ctx,
-				crclient.ObjectKeyFromObject(volSnap),
-				vs,
-			); err != nil {
-				return false,
-					errors.Wrapf(
-						err,
-						"failed to get volumesnapshot %s/%s",
-						volSnap.Namespace, volSnap.Name,
-					)
-			}
-
-			if vs.Status == nil || vs.Status.BoundVolumeSnapshotContentName == nil {
-				log.Infof("Waiting for CSI driver to reconcile volumesnapshot %s/%s. Retrying in %ds",
-					volSnap.Namespace, volSnap.Name, interval/time.Second)
-				return false, nil
-			}
-
-			if err := crClient.Get(
-				ctx,
-				crclient.ObjectKey{
-					Name: *vs.Status.BoundVolumeSnapshotContentName,
-				},
-				vsc,
-			); err != nil {
-				return false,
-					errors.Wrapf(
-						err,
-						"failed to get VolumeSnapshotContent %s for VolumeSnapshot %s/%s",
-						*vs.Status.BoundVolumeSnapshotContentName, vs.Namespace, vs.Name,
-					)
-			}
+		pollFunc,
+	)
 
-			// we need to wait for the VolumeSnapshotContent
-			// to have a snapshot handle because during restore,
-			// we'll use that snapshot handle as the source for
-			// the VolumeSnapshotContent so it's statically
-			// bound to the existing snapshot.
-			if vsc.Status == nil ||
-				vsc.Status.SnapshotHandle == nil {
-				log.Infof(
-					"Waiting for VolumeSnapshotContents %s to have snapshot handle. Retrying in %ds",
-					vsc.Name, interval/time.Second)
-				if vsc.Status != nil &&
-					vsc.Status.Error != nil {
-					log.Warnf("VolumeSnapshotContent %s has error: %v",
-						vsc.Name, *vsc.Status.Error.Message)
-				}
-				return false, nil
-			}
+	if err == nil {
+		return vsc, nil
+	}
+	if !wait.Interrupted(err) {
+		return nil, err
+	}
 
-			return true, nil
-		},
+	// If the first poll timed out, poll with a longer interval and the full timeout.
+	interval = 5 * time.Second
+	err = wait.PollUntilContextTimeout(
+		context.Background(),
+		interval,
+		csiSnapshotTimeout,
+		true,
+		pollFunc,
 	)
 
 	if err != nil {