We know of one expected case for this, where the machine shuts down during migration (homedir is unmounted in the middle of migration, causing the files to disappear)

The only other reason I can think of that files would disappear is due to low storage cleanup, but that should happen on the same thread as the migration, so it shouldn't be an issue.

why does the unmount happen in middle of migration ? Does it happen on a different cryptohome thread?

Yes, Unmount is called on the main thread, whereas the migration is run on the mount thread.  Unmount should probably be called on the mount thread as well.

summary of current understanding 

 - power shutdown is currently allowed and that causes an error to be sent out, we should disallow that b/35900434
 - power shutdown and retry probably results in resumed successful migration (to be confirmed by dan)
 - cryptohome should do something about error stats or gracefully handle normal shutdown if it really gracefully handles normal shutdown.

Ok, the culprit seems to be the powerd policy.  In login screen lid close policy is to shutdown, and we never change that since we stop the signin flow early.  This should be changed to suspend.  Tracked at
https://bugs.chromium.org/p/chromium/issues/detail?id=726229

As for successful migrations:
I'm analyzing the stats, but there's definitely discrepancies.  In particular we've got a lot of client IDs for which we seem to have partial stats (more end statuses than start statuses, more successes than initial migration attempts, etc).  However the vast majority of these failures do seem to eventually lead to successful migrations, and the outstanding ones could easily be explained by people that shut the lid and just haven't gotten around to logging back in yet.

There's a report from tester that might be different case than what we know
https://bugs.chromium.org/p/chromium/issues/detail?id=728892

sub-bugs are closed, mark this as fixed?