Hi,
I posted this a while ago aswell :)
MD can't start anymore when /etc/mdadm.conf uses /dev/disk/by-id/* ,
it bluntly tells you there are no suiteable drives.
Since libata tends to mangle my drive order by every reboot since I
use 3 controllers and the drivers dont generate uevents always in the
same sequence the only thing left to use are UUIDs.
Now, this isnt a problem but is this the *only* way the mdadm devices
will startable now and is this the only way the MD module will
support?
And if yes, maybe this should be communicated somewhere, cause
something tells me I am not the only one who had this problem or will
have :)
Patrick
On Monday February 26, [email protected] wrote:
> Hi,
>
> I posted this a while ago aswell :)
>
> MD can't start anymore when /etc/mdadm.conf uses /dev/disk/by-id/* ,
> it bluntly tells you there are no suiteable drives.
What else does it say if you run with "-v"
>
> Since libata tends to mangle my drive order by every reboot since I
> use 3 controllers and the drivers dont generate uevents always in the
> same sequence the only thing left to use are UUIDs.
No. You can use the uuid in the md metadata. Don't mention devices
in mdadm.conf at all. Just
ARRAY /dev/md1 UUID=xxxxx:xxxxx:Xxxxxx:xxxxx
and it should find the right devices and assemble them without
problems.
>
> Now, this isnt a problem but is this the *only* way the mdadm devices
> will startable now and is this the only way the MD module will
> support?
The only supported way to assemble arrays when md is a module is to
use mdadm. Mdadm has several ways to identify an array. Check the
man page.
>
> And if yes, maybe this should be communicated somewhere, cause
> something tells me I am not the only one who had this problem or will
> have :)
>
Suggestions for specific improvements to the md and mdadm documentation
are always welcome. Read mdadm.8 again and then tell me what you
wanted it to tell you that wasn't clear.
Thanks,
NeilBrown
On 2/26/07, Neil Brown <[email protected]> wrote:
> ARRAY /dev/md1 UUID=xxxxx:xxxxx:Xxxxxx:xxxxx
>
> and it should find the right devices and assemble them without
> problems.
That's my point, this is the *only* way working at the moment when
using mdadm --assemble , since /dev/disk/by-id* tells me no suiteable
drives, /dev/sd? DOES work but isnt reliable since /dev/sdl is not
always /dev/sdl but can be /dev/sdc the next reboot :D
>
> The only supported way to assemble arrays when md is a module is to
> use mdadm. Mdadm has several ways to identify an array. Check the
> man page.
I know, and i use mdadm. I used to use /dev/sd? since I did a simple
migration from raidtools to mdadm and this was the easiest way, then I
started to use libata which mangled my drive order every reboot, so I
got adviced to use /dev/disk/by-id/* , and now I am using UUID=some
128 bits hash here
> Suggestions for specific improvements to the md and mdadm documentation
> are always welcome. Read mdadm.8 again and then tell me what you
> wanted it to tell you that wasn't clear.
Well, what I am kinda trying to say is: function that used to work
prior to 2.6.20 doesn't seem to be working anymore :D
mdadm -assemble doesn't work with /dev/disk/by-id/* is used.
And /dev/sd? is not useable due to the drive order not being the same
after every reboot, but this is more a libata/udev problem I guess.
I am in the train now, I will do the mdadm -v thingy as soon as I am
at a somewhat more stable connection than 2Mbit HDPSA
Patrick