Mixing mdadm versions

Mixing mdadm versions

[hansbkk]

I've created and manage sets of arrays with mdadm v3.1.4 - I've been
using System Rescue CD and Grml for my sysadmin tasks, as they are
based on fairly up-to-date gentoo and debian and have a lot of
convenient tools not available on the production OS, a "stable" (read:
old packages) flavor of RHEL, which turns out is running mdadm v2.6.4.
I spec'd v1.2 metadata for the big raid6 storage arrays, but kept to
0.90 for the smaller raid1's as some of those are my boot devices.

As per a previous thread, I've noticed on the production OS the output
of mdadm -E on a member returns a long string of "failed, failed". The
more modern mdadm reports everything's OK.

- Also mixed in are some "fled"s - whazzup with that?

Unfortunately the server is designed to run as a packaged appliance
and uses the rpath/conary package manager, so I'm hesitant to fiddle
around upgrading some bits, afraid that other bits will break - the
sysadmin tools are run from a web interface to a bunch of PHP scripts.

So, here are my questions:

As long as the more recent versions of mdadm report that everything's
OK, can I ignore the mishmosh output of the older mdadm -E report?

And am I correct in thinking that from now on I should create
everything with the older native packages that are actually going to
serve the arrays in production?

Re: Mixing mdadm versions

[Phil Turmel]

On 02/17/2011 05:21 AM, hansbkk@gmail.com wrote:
> I've created and manage sets of arrays with mdadm v3.1.4 - I've been
> using System Rescue CD and Grml for my sysadmin tasks, as they are
> based on fairly up-to-date gentoo and debian and have a lot of
> convenient tools not available on the production OS, a "stable" (read:
> old packages) flavor of RHEL, which turns out is running mdadm v2.6.4.
> I spec'd v1.2 metadata for the big raid6 storage arrays, but kept to
> 0.90 for the smaller raid1's as some of those are my boot devices.

The default data offset for for v1.1 and v1.2 meta-data changed in mdadm v3.1.2.  If you ever need to use the running system to "mdadm --create --assume-clean" in a recovery effort, the data segments will *NOT* line up if the original array was created with a current version of mdadm.

(git commit a380e2751efea7df "super1: encourage data alignment on 1Meg boundary")

> As per a previous thread, I've noticed on the production OS the output
> of mdadm -E on a member returns a long string of "failed, failed". The
> more modern mdadm reports everything's OK.
> 
> - Also mixed in are some "fled"s - whazzup with that?
> 
> Unfortunately the server is designed to run as a packaged appliance
> and uses the rpath/conary package manager, so I'm hesitant to fiddle
> around upgrading some bits, afraid that other bits will break - the
> sysadmin tools are run from a web interface to a bunch of PHP scripts.
> 
> So, here are my questions:
> 
> As long as the more recent versions of mdadm report that everything's
> OK, can I ignore the mishmosh output of the older mdadm -E report?

Don't know.

> And am I correct in thinking that from now on I should create
> everything with the older native packages that are actually going to
> serve the arrays in production?

If there's a more modern Red Hat mdadm package that you can include in your appliance, that would be my first choice.  After testing with the web tools, though.

Otherwise, I would say "Yes", for the above reason.  However, the reverse problem can also occur.  You won't be able to use a modern mdadm to do a "--create --assume-clean" on an offline system.  That's what happened to Simon in another thread.  Avoiding that might be worth the effort qualifying a newer version of mdadm.

Phil

Re: Mixing mdadm versions

[hansbkk]

OK, thanks for those details.

I plan to use the running system for day-to-day serving of the data,
and to use the more modern versions (which originally created the
arrays) for any recovery/maintenance.

I believe the running system will be getting upgraded (RHEL5?) in the
next few months, so unless I have reason to think there's actually
something wrong, I think I'll leave it alone, I really don't feel like
learning another package management system at the moment - the Linux
learning curve has been making my brain ache lately 8-)

On Thu, Feb 17, 2011 at 8:25 PM, Phil Turmel <philip@turmel.org> wrote:
> On 02/17/2011 05:21 AM, hansbkk@gmail.com wrote:
>> I've created and manage sets of arrays with mdadm v3.1.4 - I've been
>> using System Rescue CD and Grml for my sysadmin tasks, as they are
>> based on fairly up-to-date gentoo and debian and have a lot of
>> convenient tools not available on the production OS, a "stable" (read:
>> old packages) flavor of RHEL, which turns out is running mdadm v2.6.4.
>> I spec'd v1.2 metadata for the big raid6 storage arrays, but kept to
>> 0.90 for the smaller raid1's as some of those are my boot devices.
>
> The default data offset for for v1.1 and v1.2 meta-data changed in mdadm v3.1.2.  If you ever need to use the running system to "mdadm --create --assume-clean" in a recovery effort, the data segments will *NOT* line up if the original array was created with a current version of mdadm.
>
> (git commit a380e2751efea7df "super1: encourage data alignment on 1Meg boundary")
>
>> As per a previous thread, I've noticed on the production OS the output
>> of mdadm -E on a member returns a long string of "failed, failed". The
>> more modern mdadm reports everything's OK.
>>
>> - Also mixed in are some "fled"s - whazzup with that?
>>
>> Unfortunately the server is designed to run as a packaged appliance
>> and uses the rpath/conary package manager, so I'm hesitant to fiddle
>> around upgrading some bits, afraid that other bits will break - the
>> sysadmin tools are run from a web interface to a bunch of PHP scripts.
>>
>> So, here are my questions:
>>
>> As long as the more recent versions of mdadm report that everything's
>> OK, can I ignore the mishmosh output of the older mdadm -E report?
>
> Don't know.
>
>> And am I correct in thinking that from now on I should create
>> everything with the older native packages that are actually going to
>> serve the arrays in production?
>
> If there's a more modern Red Hat mdadm package that you can include in your appliance, that would be my first choice.  After testing with the web tools, though.
>
> Otherwise, I would say "Yes", for the above reason.  However, the reverse problem can also occur.  You won't be able to use a modern mdadm to do a "--create --assume-clean" on an offline system.  That's what happened to Simon in another thread.  Avoiding that might be worth the effort qualifying a newer version of mdadm.
>
> Phil
>
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html