[Cluster-devel] Cluster 3.0.7 stable release

Fabio M. Di Nitto fdinitto at redhat.com
Tue Jan 12 14:24:19 UTC 2010


Hi Marc,

On 1/12/2010 3:10 PM, Marc - A. Dahlhaus [ Administration | Westermann
GmbH ] wrote:

>> - - linux kernel 2.6.31
> 
> gfs-kernel doesn't build against 2.6.32 without the following (untested)
> change:
> 
> sed 's/generic_file_aio_write_nolock/blkdev_aio_write/' \
>    -i gfs-kernel/src/gfs/ops_file.c
> 
> But i don't know if this change is the right thing to do...

No, this fix is not correct. I spoke at length with the GFS team and the
porting to 2.6.32 is a lot more complex.

> --8<--
>> Fabio M. Di Nitto (15):
> --8<--
>>       gfs: disable gfs kernel and userland tool build by default
> 
> This one caught my attention...
> 
> Is porting of gfs-kernel to latest vanilla kernel still in progress or
> is gfs1 going to be abandoned anytime soon?

For now we had to put gfs on hold because of the complexity to port it
to 2.6.32 and allow the release of user land tools with many bug fixes.

We are also encouraging people to move away from gfs in favor of gfs2 so
that we can better focus our resources to maintain one file system
instead of two. I don´t think this was ever a secret since gfs2 is part
of kernel upstream and gfs1 is not and never will :)

So basically, the code is there and builds (with the restriction to the
kernel version), it will receive bug fixes, but we want to make it a
user conscious choice that they are building something that will
disappear in the future (hence the requirement to explicitly enable gfs
build).

> Some quick and dirty testing with gfs2 went fine so far.

GFS2 has become a lot more stable with recent kernel releases but please
file BZs for any problem you might find.

I did CC the GFS team in case you might have more specific questions.

Cheers
Fabio




More information about the Cluster-devel mailing list