Lustre discuss - May 2006 - Support for unmodified enterprise kernel?

Hello!

On Wed, Jan 11, 2006 at 05:52:52AM -0500, kyoung@canadian.net
wrote:
> While evaluating lustre for production usage, I found:
> (1) all RHEL3/RHEL4 kernels require patches
Right.
> (2) the SuSE Enterprise Linux 9 kernels have integrated lustre patches to
> some degree, but they are closer to the patches required by lustre 1.2.6/7,
> than those required by lustre 1.4.x.
This is not quite right. Latest SLES9 kernel from SUSE updates includes most of
1.4.x patches already. All extra patches left are just bugfixes.
1.4.x should work on (latest) SLES9 kernel just fine (in most of the cases).
> This has three consequences for me:
> (a) patching the kernel will invalidate my enterprise support agreement,
Out of interest - do they allow to insert arbitrary kernel modules without
invalidating the support agreement? If so, this is pretty strange.
It is possible to change kernel with kernel module in a similar manner
as the patch allows. (yes, this is somewhat harder in some cases, but this
is still possible, as e.g. some rootkits and binary-only modules show us).
> (c) since the prepatched kernel touches the VFS, I do not have a
> foolproof argument that my partitions using plain ext3 filesystems
won''t
> be corrupted by the patched kernel; therefore if I mount any production
Same argument is possbile with extra modules. Kernel module has access to
kernel memory. Extra FS module can easily corrupt data on other filesystems.
(even non-FS module can do this, of course. This happened more than once
already with e.g. nvidia driver.)
> filesystem with the patched kernel I am not protected by my enterprise
> support either.
So do they support you when you have 3rd party modules installed, or would they
just tell you "please try again with 3rd-party modules not loaded, then
come again"? (and in the former case this is not much different
from patched kernel - just run unpatched kernel to verify the source
of the problem).

Not that I defend a necessity of having lustre patches (I would prefer those
to be part of the kernel of course), I just like to understand
how is that support agreement limits your patching of kernel, but does not
limit you from loading strange modules.

Bye,
    Oleg

Hi,

While evaluating lustre for production usage, I found:
(1) all RHEL3/RHEL4 kernels require patches
(2) the SuSE Enterprise Linux 9 kernels have integrated lustre patches to
some degree, but they are closer to the patches required by lustre 1.2.6/7,
than those required by lustre 1.4.x.
(3) but even deploying lustre-1.2.6/7 on SuSE 9 will require *some* patches
though not that many; but then I suspect CFS no longer offers
commercial support for 1.2.x series

Therefore, I have come to the conclusion I cannot install any version of
lustre on any version of linux with enterprise support without patching
the kernel (or booting CFS''s prepatched ones using LILO).

This has three consequences for me:
(a) patching the kernel will invalidate my enterprise support agreement,
(b) any binary-only device driver I have (like EMC) for enterprise linux
can no longer be supported,
(c) since the prepatched kernel touches the VFS, I do not have a
foolproof argument that my partitions using plain ext3 filesystems
won''t
be corrupted by the patched kernel; therefore if I mount any production
filesystem with the patched kernel I am not protected by my enterprise
support either.

Have I missed something?

Let me make it very clear this is *not* an evaluation of lustre''s
technical merits.  Just evaluating the risks of my enterprise ass.
Is getting CFS to support all kernel related issues currently a
(or the only) solution?

Ken