July 23, 201213 yr I'd like to have the ability, when doing either a parity check/reconstruction or rebuilding a data drive, to do multiple verification reads before assuming that the data read from the various drives is correct. For instance, each drive must return the same data for the previous N={2 or 3 or X} reads before it is trusted and used the recompute the missing data on the failed drive / parity drive. This covers a fairly infrequent error case where the underlying drives/controller cannot be relied on to consistently return the same data when reading, but it does happen, and has happened to me (still trying to work through / discover / fix the problem). Previous occurrences: here, here, and here (me) Understandably, this will make the parity check take at least N times longer if enabled, but there are times when such a feature is needed to ensure data integrity. If, for instance, one of my other drives were to fail before I'm able to be sure this error were fixed...
July 23, 201213 yr Author Actually -- I have something of a need for this now. I have real parity errors (from an unclean shutdown) mixed in with intermittent parity errors (from my flaky device). I know which are which (the ones that persist are real, the ones that don't ... aren't). I either need something like I described, or the ability to selectively check / update the parity only for a given block. In my current semi-broken state, it's unlikely that I'll be able to run through a full parity update without any of the transient errors from my flaky drive popping up.
Archived
This topic is now archived and is closed to further replies.