On Tue, May 14, 2013 at 12:16:57PM +0200, Federico Giannici wrote: > On 05/14/13 11:07, Stefan Sperling wrote: > >On Sun, May 12, 2013 at 05:27:31PM +0200, giann...@neomedia.it wrote: > >>I had to go to "/dev" and execute "sh /dev/MAKEDEV" with all the sd0, sd1, > >>sd2, sd3, sd4 (maybe only sd2 was required), and then restart the upgrade > >>script. > > > >That's somewhat expected. The installer doesn't support softraid yet. > > > >Strictly speaking, setups where system partitions (other than /home > >or other user data partitions) are hosted on softraid should currently > >be considered experimental. Manual intervention is required for such > >setups, and that includes creating disk device files that are missing > >from the ramdisk. > > > >Perhaps the install script could easily be fixed to create the necessary > >device files. But it needs a non-trivial amount of work to support > >installing to a softraid volume and also upgrading them (I've tried to > >write support for that a while back, it's not very simple). Also, not > >many architectures can currently boot from softraid, and I'm not sure > >whether all architectures can already host the root filesystem on it. > > > >BTW, I found that on sparc64, the ramdisk is so starved for inodes that > >creating device nodes for one or two additional disks will cause problems > >while downloading the sets. Fix is to remove unneeded devices from the /dev > >directory as well as creating the needed ones. > > Apart from the MAKEDEV of the missing disk devices, the upgrade > script apparently worked correctly in two amd64 PCs with softraid 1 > boot. What other things I have to pay attention to, and what else > could go wrong? > > I'm asking this because I was thinking to do other amd64 machines > with softraid 1 boot. But maybe it's still better to wait before > doing this? > > Thanks. > > -- > ___________________________________________________ > __ > |- giann...@neomedia.it > |ederico Giannici http://www.neomedia.it > ___________________________________________________ >
The upgrade process scans fstab and creates the devices needed to mount any ffs filesystems. So I imagine sd4 would have been created but constituent devices that don't appear in fstab or have no ffs filesystems specified will not be created. In the new RAID world it needs another check. Probably the easiest is to create all devices up to the highest disk encountered. This does, as pointed out, have some implications for inode starved install media on some architectures. Doing disklabel on all devices and looking for RAID partitions might be possible too. Which would minimize inode starvation. Deep thought. Hmmm. .... Ken
