On Wed, 23 Nov 2011, Alexandre Oliva wrote:
> On Nov 22, 2011, Sage Weil <[email protected]> wrote:
>
> > On Tue, 22 Nov 2011, Christian Brunner wrote:
>
> >> - compression: I'm using lzo compression right now, as my CPUs in the
> >> OSD nodes where idle most of the time and it is improving throughput
> >> quite a bit.
>
> > The caveat here is that there is a corner case with inline extents and the
> > clone ioctl that isn't implemented.
>
> What happens when you trigger that case? Crash or corruption?
>
> I ask because I've been running with zlib compression, and I haven't
> noticed any problems along these lines, though I haven't got to a point
> in which I'm doing more than rsyncing whole trees into the filesystem
> yet.
Took another look at this. The ioctl will return EINVAL from
fs/btrfs/ioctl.c:2416:
if (comp && (skip || trim)) {
ret = -EINVAL;
btrfs_end_transaction(trans, root);
goto out;
}
If os/FileStore.cc::_do_clone_range() sees EINVAL, it falls back to doing
a copy. That may not be ideal in some cases, but at least it's
"correct". For example,
0-8192 compressed inline
8192-100MB normal extent
and we clone 4k - 100MB, and get EINVAL, we ideally wouldn't fall back to
naively copying 100MB. For RBD, those blocks are 4MB by default, so it's
not a big deal.
So you should be okay here.
sage
> >> - nocluster (requires an additional btrfs patch from Alexandre): I was
> >> reading that he was able to reduce the excessive metadata overhead,
> >> but it is slower. Does anyone know more about it?
>
> > We haven't tried this.
>
> I've used it occasionally and temporarily, to rebalance metadata and
> bring btrfs (even with clustered allocation) back to a faster pace
> without umount.
>
> --
> Alexandre Oliva, freedom fighter http://FSFLA.org/~lxoliva/
> You must be the change you wish to see in the world. -- Gandhi
> Be Free! -- http://FSFLA.org/ FSF Latin America board member
> Free Software Evangelist Red Hat Brazil Compiler Engineer
> --
> To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
>
>
--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
