From 1e0d64c04883a8d7b09677e9721f9e189743e2f3 Mon Sep 17 00:00:00 2001
From: chrislu <chris.lu@gmail.com>
Date: Wed, 26 Oct 2022 12:32:24 -0700
Subject: [PATCH] master: delete partially created replicated volumes during
 volume growth

fix https://github.com/seaweedfs/seaweedfs/discussions/3792#discussioncomment-3973120
---
 weed/topology/allocate_volume.go | 12 +++++++++++
 weed/topology/volume_growth.go   | 34 ++++++++++++++++++++++++--------
 2 files changed, 38 insertions(+), 8 deletions(-)

diff --git a/weed/topology/allocate_volume.go b/weed/topology/allocate_volume.go
index 41c071f9f..0254eb7e6 100644
--- a/weed/topology/allocate_volume.go
+++ b/weed/topology/allocate_volume.go
@@ -30,3 +30,15 @@ func AllocateVolume(dn *DataNode, grpcDialOption grpc.DialOption, vid needle.Vol
 	})
 
 }
+
+func DeleteVolume(dn *DataNode, grpcDialOption grpc.DialOption, vid needle.VolumeId) error {
+
+	return operation.WithVolumeServerClient(false, dn.ServerAddress(), grpcDialOption, func(client volume_server_pb.VolumeServerClient) error {
+
+		_, allocateErr := client.VolumeDelete(context.Background(), &volume_server_pb.VolumeDeleteRequest{
+			VolumeId: uint32(vid),
+		})
+		return allocateErr
+	})
+
+}
diff --git a/weed/topology/volume_growth.go b/weed/topology/volume_growth.go
index 0f9ed2812..71610730f 100644
--- a/weed/topology/volume_growth.go
+++ b/weed/topology/volume_growth.go
@@ -228,10 +228,11 @@ func (vg *VolumeGrowth) findEmptySlotsForOneVolume(topo *Topology, option *Volum
 	return
 }
 
-func (vg *VolumeGrowth) grow(grpcDialOption grpc.DialOption, topo *Topology, vid needle.VolumeId, option *VolumeGrowOption, servers ...*DataNode) error {
+func (vg *VolumeGrowth) grow(grpcDialOption grpc.DialOption, topo *Topology, vid needle.VolumeId, option *VolumeGrowOption, servers ...*DataNode) (growErr error) {
+	var createdVolumes []storage.VolumeInfo
 	for _, server := range servers {
 		if err := AllocateVolume(server, grpcDialOption, vid, option); err == nil {
-			vi := storage.VolumeInfo{
+			createdVolumes = append(createdVolumes, storage.VolumeInfo{
 				Id:               vid,
 				Size:             0,
 				Collection:       option.Collection,
@@ -239,14 +240,31 @@ func (vg *VolumeGrowth) grow(grpcDialOption grpc.DialOption, topo *Topology, vid
 				Ttl:              option.Ttl,
 				Version:          needle.CurrentVersion,
 				DiskType:         option.DiskType.String(),
-			}
+			})
+			glog.V(0).Infof("Created Volume %d on %s", vid, server.NodeImpl.String())
+		} else {
+			glog.Warningf("Failed to assign volume %d on %s: %v", vid, server.NodeImpl.String(), err)
+			growErr = fmt.Errorf("failed to assign volume %d on %s: %v", vid, server.NodeImpl.String(), err)
+			break
+		}
+	}
+
+	if growErr == nil {
+		for i, vi := range createdVolumes {
+			server := servers[i]
 			server.AddOrUpdateVolume(vi)
 			topo.RegisterVolumeLayout(vi, server)
-			glog.V(0).Infoln("Created Volume", vid, "on", server.NodeImpl.String())
-		} else {
-			glog.V(0).Infoln("Failed to assign volume", vid, "to", servers, "error", err)
-			return fmt.Errorf("Failed to assign %d: %v", vid, err)
+			glog.V(0).Infof("Registered Volume %d on %s", vid, server.NodeImpl.String())
+		}
+	} else {
+		// cleaning up created volume replicas
+		for i, vi := range createdVolumes {
+			server := servers[i]
+			if err := DeleteVolume(server, grpcDialOption, vi.Id); err != nil {
+				glog.Warningf("Failed to clean up volume %d on %s", vid, server.NodeImpl.String())
+			}
 		}
 	}
-	return nil
+
+	return growErr
 }