@joaquin Just got a reply from Ben from Kind here: dgraph errors on restart · Issue #1909 · kubernetes-sigs/kind · GitHub and after seeing that, I also did a test on my end to dig into this further and from the looks of it, I guess the data is still available in the PVC even after Docker/VM restarts but still there is something else which is causing the issue.
If you notice the date, those are the folders from the PVC mounted to Dgraph pod which were created 1 day ago (before the pre-emptible VM restart). So, the data does exist in the PVC but Dgraph fails.
My guess is that Dgraph might be maintaining some kind of state or info in memory which can get lost across restarts (but strange to see this not happening when doing a rollout restart) maybe because all the nodes are not down all at once as you said.