Edd Barrett wrote: > Hi, > > I just had a read of the man page for the new softraid features in > OpenBSD. Sounds like it's well on the way to replace raidframe, but > some questions remain unanswered. > > a) Can disks be dynamically added/removed to a mirror? > b) What happens when a disk fails?
Almost anything. Which is true of any RAID system. Had a hardware RAID system with a drive that failed. Ended up with a short across a set of power supply pins. System went down. Disk tray was damaged. RAID controller was damaged. Power supply on the computer was damaged. Power supply on a computer I used to verify the failure of the disk was damaged (it was an interesting day...) Data was fine. Until I botched the replacement of the RAID controller, at which point, 100G or so of data was restored from backup. :) All RAID is sensitive to the mode of failure of the drive. Software RAID has the problem of relying on a non-RAID-focused interface. Not all interfaces are happy with many common modes of failure. Many won't reboot on the alternate drive after a failure of the primary, most won't boot on alternate drive under many circumstances (drive doesn't completely fall off the bus). The above is a really bad answer. The proper answer is "FIND OUT FOR YOURSELF!". For ANY RAID system, YOU must know how to recover. I'm not aware of any software RAID system which does all the thinking for you -- that is actually one part fo the magic in SW RAID is that you control it completely. You must also practice. (And even then, you might botch things up. Or at least, I did. :) > c) Does softraid do the same as raidframe on fsck at boot? Not yet. It actually doesn't do any recovery itself yet. Hence, the note under "CAVEATS". > d) Is there a way to query the status of a volume? (Is this what > 'enumerate' means?) Same way all new OpenBSD RAID systems are monitored, bioctl(8) (which is darned cool). > Also softraid is not yet listed in the raid options section of teh FAQ. It also isn't in GENERIC yet. It isn't quite ready for production. The lack of a rebuild feature is a limiting factor. It isn't quite fatal, you can manually rebuild with dd(1). While doing this for a disk failure isn't all bad, doing it every time the power goes out unexpectedly (or it crashes for any reason) is a pain, and ignoring it after an event is potentially risky. Nick.
