Creating this ticket to investigate this case further. This appears to have happened with a customer on Windows machine and might hint towards a bug.

Steps apparent from mongod logs:

1. A drop of a collection was issued causing a schema change. A schema change involves a metadata-checkpoint and writing of a new turtle file. For some reason (possibly file locked by antivirus) the turtle file rename failed, failing the checkpoint and the schema change operation.
2. Metatracking would have triggered a metadata unroll, which failed as well (possibly due to WT_PANIC set by the above operation ?)
3. Next reboot of mongod did not find the collection on which drop had failed. This hints towards MongoDB's catalog and WiredTiger's metadata to have gone out of sync.

To investigate:

1. What is the expected behavior in case metadata-checkpoint fails during a schema drop?
2. Why did metadata unroll fail?
3. How did MongoDB's catalog and WiredTiger's metadata go out of sync?