Yeah, it happened again after two days, at 2:04 AM. At least we caught it with nagios immediately. This time it was not on the leader, so the other two nodes continued to serve data, and the app is still up and running. I am wondering if we should downgrade from dgraph 21.03.0 to some stable version. Anything, just stable. 
PS. I noticed that our dgraph nodes run run constantly at ~600% out 1600% (16 CPU VMs). Maybe it happens during high load, the little badger can’t compact well and loses the tables? Little badger 
It’s hard to be a badger in dgraph! 
W0917 02:36:56.361629 3018868 log.go:36] While running doCompact: MANIFEST removes non-existing table
W0917 02:36:56.416721 3018868 log.go:36] [Compactor: 1] LOG Compact FAILED with error: MANIFEST removes non-existing table...