On Fri, 2002-02-22 at 11:24, Greg Freemyer wrote:
>
> I've just gotten interested in XFS because of its ACL support in the metadata
> and via xfsdump/xfsrestore.
>
> I'm trying to figure out what its current status is, and what is likely to
> happen over the next few months.
>
> Does the below sound correct:
>
> Current release (XFS v1.0.2):
> fairly stable, appears to be production quality
> Supports ACLs on disk and with xfsdump/xfsrestore.
> With xfsrestore a single file can be restored complete with its
> associated ACL metadata.
> Uses a XFS specific interface to the kernel from userland to manipulate
> ACL.
> Has trouble with LVM snapshots (maybe this is CVS code only).
CVS and the kernel patches have come a long way since 1.02 was released,
a lot of bugs have been fixed in the kernel and user space code which are
not in the original rpms.
LVM snapshots do appear to have some issues still, although I am
not sure what the current status really is on this one.
>
> XFS v2.0:
> Will be released when the new Linus approved EA interface is added to the
> kernel.
> (I can't tell if this is 2.4.18, or 2.5.x)
> The filesystem layout itself will not change, but the userland tools from
> V2.0 will not be compatible with a V1.0 kernel and vice-versa.
>
They will be compatible in all areas except acls and extended
attributes.
> Questions:
> Is the new EA API coming out in 2.4.18 the same as the one in 2.5.x?
yes.
>
> If so, how long after 2.4.18 is released do you expect XFS V2.0 to be
> released? (days, weeks, months??)
Days - we have rc3 in a tree now, rc4 just came out and is really small.
>
> The first release of V2.0 will be what quality? (alpha, beta, production??)
Well, in theory things keep getting better as we go along, so cvs should
be 'better' than the 1.02 release rpms. Regressions do happen, but in
general I would say we should be at or higher than the reliability level
of the original 1.02 release. I do not think there will be much 'extra'
testing going into this version if that is what you are asking.
>
> Is the 2.4.18 kernel likely to be of production quality? I have read about a
> lot of problems in the whole 2.4.x series of kernels, but it sounds like it
> is all coming together in the most recent kernels.
>
Things do seem to be getting better. For general linux things you pretty
much have to look at linux-kernel and judge for yourself.
> Because the of the new standard EA interface, will V2.0 userland ACL tools
> will be able to control both XFS and ext3 with the acl.bestbits.at extensions.
That is the theory.
>
> Is star (http://acl.bestbits.at/backup.html) able to support XFS complete
> with ACLs by using the Posix ACL interface. If not now, what about with XFS
> V2.0?
>
I have no idea - it all depends on how star extracts data from the
filesystem, does it run on a live filesystem, or does it use the
block interface. If the former then it may work, if the latter then
no it will not work.
> What about xfsdump/xfsrestore and ext3 with ACL support?
xfsdump will not work on filesystems other than xfs - since it uses xfs
specific extensions to scan the filesystem. Restore should be able to
restore an xfs filesystem with acls to another acl enabled filesystem
type - I think.
>
> Does Amanda work well with xfsdump/xfsrestore? star?
>
Amanda works with xfsdump/restore, I cannot speak for star.
> In general, do you think XFS or ext3 with ACL extensions is a more production
> ready environment. (I know your biased, but I would still appreciate an
> answer.)
>
I will leave the answer to that to external people, all I can say is
that xfs has had more production time than ext3, but ext3 comes from
an existing linux filesystem which has had way more linux time
than xfs has.
I actually use both filesystems - since when testing radical xfs things
I want to be able to isolate things down to one filesystem and not have
to reinstall if things go bad.
Steve
--
Steve Lord voice: +1-651-683-3511
Principal Engineer, Filesystem Software email: lord@xxxxxxx
|