[rhos-list] Unable to access Swift using Keystone on reboot

[David Hernandez]

Pete.

I consulted one of our developers and showed him the error we were getting 
when trying to write objects to HPSS.  He told me that the xattr data 
contained in the object it was trying to write did contain some binary 
data.: 

[80][02]}q[01](U[0e]Content-Lengthq[02]U[04]1024q[03]U[04]nameq[04]U8/AUTH_91186334603648d78602bf5481e9393f/c3/tmp/data4.fileq[05]U[13]X-Object-Meta-Mtimeq[06]U\r1352219008.97q[07]U[04]ETagq[08]U 
1868a60a20c5ed4a02ecb257f7019bebq\tU[0b]X-Timestampq\nU[10]1352219888.77388q[0b]U[0c]Content-Typeq[0c]U[18]application/octet 


The two digit hex values in the square brackets are the non-printable 
characters that were part of the xattr value.


David Hernandez

Contractor / HPSS
IBM Global Business Services - US Federal
12301 Kurland Dr Suite 300
Houston, TX 77034-4812 
Mobile 713-444-5755



From:   Pete Zaitcev <zaitcev at redhat.com>
To:     David Hernandez/Houston/Contr/IBM at IBMUS
Cc:     Derek Higgins <derekh at redhat.com>, rhos-list at redhat.com, 
zaitcev at redhat.com
Date:   11/05/2012 08:35 PM
Subject:        Re: Unable to access Swift using Keystone on reboot



On Mon, 5 Nov 2012 17:22:48 -0600
David Hernandez <dhern at us.ibm.com> wrote:

> Does Swift try to write binary extended attributes?

It may if user asks it to, but I think not for itself normally.

> Does Swift use setfattr to write extended atrributes?

Swift uses the syscall fsetxattr(2), while setfattr(1) is a shell command.

With the defails aside, I thought about the problem of HPSS a little bit
in general.

I continue to think that your attempts of running Swift on top of
HPSS VFS are ill-advised. If I may, you really need a different thing:
a module for so-called "Swift LFS" (Local File System) which supports
HPSS natively. Unfortunately, LFS does not yet exist. The Red Hat Storage
team is working on an implementation of it, called "Gluster UFO" (Unified
Storage Objects). If someone wrote a plugin that worked as "HPSS UFO",
the issues of extended attributes, or any other kind of impedance
matching, would be contained within such plugin. It could even bypass
HPSS VFS altogether. I think it would be ideal for your needs --
as much as I understand them, of course.

So, if you want to deploy Swift today, please use tried and tested
storage back-ends. But by all means, feel free to help OpenStack community
to develop LFS/UFO for tomorrow.

Yours,
-- Pete


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/rhos-list/attachments/20121106/1c506a0f/attachment.htm>