If, during the same scheduler round, one host writes to a file and another one reads from it, it'll currently be nondeterministic which one happens first.
We could try to prevent this problem; maybe using overlay file systems so that hosts don't see writes from other hosts. This could be fairly complicated and heavy-weight though.
It might not be that difficult to detect such problems. e.g. we could record the sets of files that each host reads and writes and compare at the end of the simulation or between scheduler rounds.
If, during the same scheduler round, one host writes to a file and another one reads from it, it'll currently be nondeterministic which one happens first.
We could try to prevent this problem; maybe using overlay file systems so that hosts don't see writes from other hosts. This could be fairly complicated and heavy-weight though.
It might not be that difficult to detect such problems. e.g. we could record the sets of files that each host reads and writes and compare at the end of the simulation or between scheduler rounds.