Containers moved to a different host lose connection to their base image

My situation is a 3 node LXD 3.19 cluster that each have a zfs pool ‘data’ for local storage.

I can create several containers on node2 and they will each only use up 20MB worth of data because of their connection to the base ubuntu:18.04 image like such:

root@node2# zfs list
NAME                                                                           USED  AVAIL  REFER  MOUNTPOINT
data                                                                          2.33G   120G    96K  none
data/containers                                                               1.61G   120G    96K  none
data/containers/accepted-mullet                                               21.8M   120G   747M  /var/snap/lxd/common/lxd/storage-pools/zfs/containers/accepted-mullet
data/containers/careful-longhorn                                              21.8M   120G   747M  /var/snap/lxd/common/lxd/storage-pools/zfs/containers/careful-longhorn
data/containers/ceph-mgr3                                                     21.8M   120G   747M  /var/snap/lxd/common/lxd/storage-pools/zfs/containers/ceph-mgr3
data/containers/ceph-mon3                                                     21.8M   120G   746M  /var/snap/lxd/common/lxd/storage-pools/zfs/containers/ceph-mon3
data/containers/closing-wolf                                                  21.8M   120G   747M  /var/snap/lxd/common/lxd/storage-pools/zfs/containers/closing-wolf
data/containers/knowing-dinosaur                                              21.8M   120G   747M  /var/snap/lxd/common/lxd/storage-pools/zfs/containers/knowing-dinosaur
data/containers/normal-kingfish                                               21.8M   120G   747M  /var/snap/lxd/common/lxd/storage-pools/zfs/containers/normal-kingfish
data/images                                                                    736M   120G    96K  none
data/images/979ff60086ca9fb1c93d3131f8eca291820a524f0bd0b7d621a3c2f5f41ef185   736M   120G   736M  /var/snap/lxd/common/lxd/storage-pools/zfs/images/979ff60086ca9fb1c93d3131f8eca291820a524f0bd0b7d621a3c2f5f41ef185

But once I move my 5 containers to node 1 so that I can do maintenance/reboot node2 all of a sudden these 5 containers take up an extra 3.5GB worth of space.

root@node1# zfs list
NAME                                                                           USED  AVAIL  REFER  MOUNTPOINT
data                                                                          3.97G   918G    24K  none
data/containers                                                               3.32G   918G    24K  none
data/containers/accepted-mullet                                                673M   918G   673M  /var/snap/lxd/common/lxd/storage-pools/zfs/containers/accepted-mullet
data/containers/careful-longhorn                                               673M   918G   673M  /var/snap/lxd/common/lxd/storage-pools/zfs/containers/careful-longhorn
data/containers/closing-wolf                                                   673M   918G   673M  /var/snap/lxd/common/lxd/storage-pools/zfs/containers/closing-wolf
data/containers/knowing-dinosaur                                               673M   918G   673M  /var/snap/lxd/common/lxd/storage-pools/zfs/containers/knowing-dinosaur
data/containers/normal-kingfish                                                673M   918G   673M  /var/snap/lxd/common/lxd/storage-pools/zfs/containers/normal-kingfish
data/images                                                                    663M   918G    24K  none
data/images/979ff60086ca9fb1c93d3131f8eca291820a524f0bd0b7d621a3c2f5f41ef185   663M   918G   663M  /var/snap/lxd/common/lxd/storage-pools/zfs/images/979ff60086ca9fb1c93d3131f8eca291820a524f0bd0b7d621a3c2f5f41ef185

Their connection to the base image is now forever lost.

Is this expected behavior, and if so, what can I do to prevent this? Do I need to setup deduplication on my zfs pool to win back the lost storage?

That’s normal, no way you can prevent it.

For the relationship to be present, the source must be absolutely identical, as in, even have the same internal zfs identifier. This isn’t possible when both systems have independently created that dataset.