If the "change repository blob store" task fails to complete running it can't be resumed. The blob store of the repository will already have been changed, and not all data will have been moved over to the new blob store. We've seen this happen on multiple occasions now. The only workaround is to try and move the repository back to its original blob store and hope there are no failures during the second move. If a repository is left in a partially moved state it results in a major outage that is very difficult to recover from.
Expected: It must be possible to resume the task in the case of a failure to move all content.
One possible way to do that would be to copy, rather than move the blobs, marking each one as moved in the original after successful copy. After all have been moved successfully the task could then remove the originals from the source blob store.