On 2/12/22 8:47 PM, L A Walsh wrote: > > > On 2022/02/11 12:59, Eric Sandeen wrote: >> Hi folks, > >> because I don't want to stall on getting the fix for the root inode >> problems out there any longer. It's been ... quite long enough. >> >> Next release we can think about the dmapi junk, and possibly the >> workaround for the broken dumps with the root inode problem. >> >> This should at least be generating valid dumps on those filesystems >> with "interesting" geometry, again. > ---- > > Thanks for the Cc. > > As near as I understand your comments, this should generate new dumps that won't > have the problem I'm seeing, but will still leave previous dumps in the old format > that may show problems upon restoration. Correct, I want to spend more time on the workaround to be sure we have confidence in it, sorry. > FWIW, xfs/restore-dump aren't the only util to have problems with little-used geometries. > > The BSD-DB libs of 10-15 years ago used in in perl have problems with a RAID50 setup. AFAIK, the problem was never fixed, because the perl punted the problem to the bsdlib. Unfortunately, no one was supporting that lib anymore. I think a new db-lib was added > (mariadb?, I forget) that replaced the older format), so newer perls may not > have that bug. > > It only appeared when the "POSIX-Optimal-I/O" size was NOT a *power*-of-2. > The lib confused *factor*-of-2 alignment with Pow-of-2 for use in db-internal > size-allocations provided resulting in some internal allocations being of size-0, which doesn't work very well. > > Of course perl doesn't check for this error case, since the problem isn't considered > to be a problem in the perl-code. Funky ;) -Eric
