On Fri, 19 Mar 1999, Piete Brooks wrote:
> > The 'old' mdadd+mdrun was always re-creating arrays as well.
>
> So that has to be used ... [ but cannot share a /etc/raidtab :-( ]
no, you should use mkraid at every bootup. You can alias it to raid0start
or whatever ... the old tools _were_ recreating the array for every
'startup'. The difference now is that you can have persistent
configuration data, which can be auto-started. (either at bootup or
manually) raidstart does not use anything from the raidtab, it just picks
one device to get the kernel a superblock, but otherwise the other
parameters are irrelevant for raidstart.
> > raidstart (and autostart) starts only 'persistent' arrays.
>
> I am learning ...
>
> raidstart appears to read the SB from the first "device" it finds (not device
> 0, but the first) in /etc/raidtab and use that.
it's the kernel that reads that device.
> This means that if the first disk of a RAID1 is bust/corrupt/..., raidstart
> won't work until the lines of /etc/raidtab are re-arranged.
> Is this intended ?
well, i've planned to 'probe' every device mentioned in raidtab, but the
preferred way is to use autostart anyway. (autostart can deal with such
partly-corrupt scenarios too, plus with lots of other scenarios like
device renumbering)
> Would it work if raidstart could be asked to ignore /etc/raidtab and instead
> start the RAID device whose SB can be found on /dev/XXXN
yep, nice idea, this could be a new option. (because the parameter to
raidstart has to be the MD device) Feel free to send me patches.
-- mingo
