On Tue, Nov 22, 2011 at 04:03:15PM -0800, Gregory Farnum wrote: > We're going to look into this in more detail very shortly. Right now > all I can tell you is that none of the drives ever actually lost > power, so unless something is explicitly telling them to clear their > caches I don't know how the drives could have lost their cache to > cause a problem like this. > But for now I'll just see what I can get by zeroing out the log, and > we'll get back to you again if we manage to reproduce this in a > situation where we can tell you more definitively about the caching > and barriers. I would be very interested in that. The situation you have is fairly easy to pinpoint down - we do log recovery and find and EFI record in the log - that is one that logs the intent to free extents. We haven't found the corresponing EFD record, which cancels it out as part of the transaction doing the actual freeing. I'd love to see a log dump of a fs iff this can be reproduced as that would shows us if we had some sort of bug in recovery. _______________________________________________ xfs mailing list xfs@xxxxxxxxxxx http://oss.sgi.com/mailman/listinfo/xfs
