With the current release there is one issue with the scary title that we are aware of. It seems to be happening every now and then on ARM64 during initial data sync, but is not limited to the RaspberryPi. As far as we can tell, this needs to get investigated a little more at the cardano-node level. There is no indication of what is going on neither on the console nor in the log file.
However, it sounds more serious than it is, because a simple docker restart has so far fixed this every time. We have not seen this behavior once the node is fully synced. If you already have block data from some other node, you can copy it to this node's data mount point. Initial sync will be much faster and you'll likely not run into this issue. When running on x86_64 you would not be seeing this.
Due to some non-trivial bug in the Haskell Nix layer, it is currently not possible to use the upstream build to produce an arm64 cardano-node binary. In this project we have a convoluted workaround that builds the node in Docker, instead of using Nix. Eventually we would of course like to see all of that happening upstream, so that this project would not need to build the node at all, but instead could focus on added functionality (if even necessary). The end user would not be affected by this issue - our arm64 image runs just fine when built like this.
We will keep a close eye on these issues.