Since it's getting stuck on restart I'm pretty sure I can reproduce it.. when a write causes a crash we try to skip it on the next restart. Which is why it could be working on the second restart