Re: woody: Remaining RC bugs in base system
In article <[🔎] 20020125084231.GB18822@A-Eskwadraat.nl>,
Bas Zoetekouw <bas@debian.org> wrote:
>Hi Matt!
>
>You wrote:
>
>> #127635: sysvinit: halt called with -h is fatal for root-raid systems
>[critical]
>>
>> NEEDS ATTENTION
>> The final consensus seems to be that this is more likely a kernel bug than
>> a sysvinit bug. Someone who understands the MD subsystem needs to look
>> into this and reassign it with some details.
>
>No, I think it is a bug in raidtools. Raidtools should provide a init.d
>script that flushes the raid superblocks before the halt.
It does provide such a script, but apparently it isn't enough or
doesn't work.
The whole situation is a mess and should be solved in the kernel.
What we're doing in userspace is just a kludge which works for
most people.
As I understand, with the new Hedrick IDE code it will be solved
and this will show up in newer 2.4 kernels (perhaps 2.4.19) and
it has already been integrated in 2.5. So for now I'm leaving the
userspace kludge around until Debian 3.2 or so ;)
Since this is all a workaround anyway I've added another one. Simply
do not put the IDE drives into standby mode if RAID is active.
Yes, that could mean that if you have a RAID array on an ATX machine
and the system powers off that the IDE write cache might not have been
completely flushed. However the final write (flushing the RAID
metadata) and the following poweroff are all done in kernel mode,
there's nothing you can do about it in userspace anyway.
Mike.
Reply to: