I'd mentioned SMART -- the drive electronics standard -- and smartmontools. On all my systems, I set up daily "short offline" tests to have the electronics do self-tests and test the data bus, and weekly "long offline" (also called "extended offline") tests to have the electronics read every sector.
On my RAID systems, it's easy enough to take a problem drive out of service and run badblocks on it, then put it back in service if it is still useable.
On my single-drive systems, what I do will depend on what sectors have failed. And that's a manual process. I would map the drive sectors to partition sectors, then map those to block numbers, then determine if the blocks are unassigned, or if assigned, to which inodes. Not easy, but dumpfs(8) can help. I haven't done this in several years, because my single-drive OBSD systems are now down to a grand total of one, and and so far .... no tests have reported any errors.
|