The most complicated part of this change is the addition of a corruption check that must be run the first time a file is opened after the server comes up, since we're buying these perf gains by playing a bit more fast and loose with the disk.
To check that this is behaving correctly we'll want to make sure that corrupting the most-recent transaction in a storage file rolls that transaction back.
This could be accomplished by updating an item that will land in a known vbucket, shutting down the server, and flipping some bits around end of the file. The update should be rolled back when the server comes back up, and nothing should freak out
A position guaranteed to affect an item body from the recentmost transaction is 4095 bytes behind the last position in the file that is a multiple of 4096, or: floor(file_length / 4096) * 4096 - 4095