[Ocfs2-users] Unstable Cluster Node

[Mark Fasheh]

On Tue, Dec 04, 2007 at 09:56:23AM -0500, Randy Ramsdell wrote:
> I thought about sending this directly to you, but it is related to the 
> ocfs2-users group, because I regularly have to compile the ocfs2 drivers 
> and we do not upgrade our kernels for this. Each time I only have to patch 
> a very few lines and it compiles as expected. Are you saying that to 
> include driver source into the kernel source, it stops the driver from 
> compiling separately  as a module?

I'm saying that "swimming against the stream" with respect to kernel/module
ABI (which breaks extremely often) is prohibitvely expensive for Oracle and
therefore we can't devote engineering resources to it. Our commitment to
mainline kernels is to the code you download from kernel.org. Personally, I
hope that the regular backporting of patches and bugfixes for folks would
indicate that we're pretty serious about helping folks on mainline kernels.

The fact is though, that Ocfs2 1.2 is only tested and coded to work against
a limited number of kernels (which *are* providing a much stronger ABI
guarantee) and that is very unlikely to ever change.


So, it's great that you manage to get Ocfs2 1.2 compiling against a mainline
kernel, but please understand - that's a far cry from us officially saying
"use Ocfs2 1.2 for any kernel you like" and then purporting to support the
result.


> I really know the answer to this, but thought maybe you could elaborate as
> to why you say that it is the procedure for adding driver source to the
> kernel source as the problem.

Randy, I get the sense that you're trying to draw me into a protracted
argument here.

You made a statement about "why vendors and developers force a complete
kernel OS upgrade for a simple driver", and I simply pointed out that it's
not always such a clear cut case.

To repeat, if you want to continue this conversation, feel free to do it
off-list. I actually think that you could help us formulate a nice FAQ
entry, IF you can cool off and drop the adversarial tone...
	--Mark

--
Mark Fasheh
Senior Software Developer, Oracle
mark.fasheh at oracle.com