El martes, 3 de octubre de 2006 06:27, Warren Turkal escribió:
> David,
>
> I think that I may have a better solution for mounting aoe based devices.
> The init script should only handle making the block devices available, like
> the lvm init script. The init script should therefore come before the
> S35mountall script. It should ifconfig up the interfaces that are needed.
> This would make the _netdev hack unneeded and not introduce errors in the
> fstab mounting. It would also allow stacking lvm or software raid or evms
> on top of the aoe devices, which your implementation seems to lack. Here is
> the pseudocode for the init.d script do_start() function.

        It was done on purpose (more below).

> modprobe aoe
> sleep 5  # avoid race condition of /dev/etherd being created

        Here you made a point: if you use udev and have AoE modular, 
/dev/etherd is 
not created at all.

> #optionally make sure /dev/etherd exists like in your script
>
> if [ -z "$INTERFACES" ]; then
>   # add all ethernet interfaces to INTERFACES
> fi
>
> #bring up interfaces over which to do aoe
> for iface in $INTERFACES; do
>   ifconfig $iface up
> done
>
> aoe_interfaces $INTERFACES
> aoe_discover
> sleep 5 # avoid race condition of /dev/etherd/e*.* being created

> Then the mountall init script is able to properly mount the aoe devices.
> You would also need to modify the lsb headers to not require $local_fs or
> $network.

        But this is not right.  Network is configured at stage 40 in initlevel 
S.  
Because of a reason.  You cannot interfere with it, and make premature 
initialization of all the network interfaces just because you think this is 
much better in your opinion is fine for your system, but not for mainstream 
Debian.  Have you even thought of NFS?  What do you think AoE is far more 
important than NFS, that is initialized on stage 45? (in fact is should be on 
40 as well, and reviewed at 45).

        Last week I asked a release manager, Steve Langasek, about my point of 
view, 
and he fully agreed with me on this issue.

        Best regards,


                Ender.
-- 
Network engineer
Debian Developer

Attachment: pgpay3WCvM8n2.pgp
Description: PGP signature

Reply via email to