I'd also version the DB in the URL, else you could end up changing the file out from under someone who's already got the page loaded, with who-knows-what results depending on how different the file is. You could just prefix a head to every range request to check for changes, but that adds overhead and doesn't actually completely close the gap, so it'd still be possible to read a file different from the one you intended. Cost is more disk usage, depending on how many copies you keep around, but at least keeping the most recent "old" version seems reasonable unless you're skating really close to the quota on whatever system you're using.
