State of GFS, OCFS2, Clustering, iSCSI, etc. in Fedora (recap a few bits)

Wed Jan 14 20:03:03 UTC 2009

On Wed, 2009-01-14 at 20:15 +0100, Dennis J. wrote:
> On 01/14/2009 01:01 PM, Fabio M. Di Nitto wrote:
> > Hi all,
> >
> > sorry if i can't reply to each email properly as I don't have a local
> > history of the mailing list.
> >
> > To reply to
> > http://lists.fedoraproject.org/pipermail/fedora-server-list/2009-January/000059.html
> >
> >> My concern right now isn't so much about the configuration of these
> >> technologies but with the fact that they don't seem to be maintained at all
> >> right now and in various states of disarray.
> >
> > Can you please be more specific? I have been updating redhat-cluster on a regular base in Fedora 9/10/rawhide.
> > What makes you think they are unmaintained?
> >
> 
> mounting ocfs2 partition fails due to missing load_module command in 
> startup script:
> https://bugzilla.redhat.com/show_bug.cgi?id=476469
> (A added the needed fix and the keyword "EasyFix" to the bug)

I forwarded this to upstream now. It's probably already fixed in more
recent versions of ocfs2-tools upstream. We were discussing to propagate
them into fedora 2 days ago.

> 
> Missing dependency: rgmanager
> https://bugzilla.redhat.com/show_bug.cgi?id=477496
> 
> system-config-cluster is looking for cman_tool in the wrong place
> https://bugzilla.redhat.com/show_bug.cgi?id=477499

Jim can you please look at those bugs and decide if
system-config-cluster should either be removed or fixed for Fedora in
favour of conga? cman already requires conga components.

> The reason I was bringing this up on the mailing-list is that bugs like the 
> ocfs2 issue and the problem with the cman_tool location are highly visible 
> so I was wondering why such bugs didn't seem to pop up on anyones radar.

Because clearly there is a very low user base for them. ocfs2 is "new"
in fedora as we (ocfs2 guys and me) revived the package not too long
ago.

system-config-cluster.. same comment as above.. it's just plain
obsolete.

> 
> After having the problems with system-config-cluster I tried Conga as the 
> alternative route for configuration but "luci" seems to have gone AWOL. 
> According to http://sourceware.org/cluster/conga/ Conga wasn't part of F7 
> and F8 because of python compat. issues. Is this the reason that luci isn't 
> part of F10?

conga/luci and modules are there from F9. I have updates some of those
builds myself so I am sure they are there.

> 
> >>> That's how I ran into the problems I mentioned. In the near future I will
> >>> have to setup approx. 30TB of storage that needs to be available for 5-10
> >>> servers. The problem is that OCFS2 seems to be just plain broken,
> >
> > Did you report those issues to upstream? We work with OCFS2 guys all the
> > time and they are very responsive.
> 
> The question for me is if the problem is fixed upstream how long does it 
> take for them to get into fedora?

Not long. If the problem is only package related it's usually less than
24 hours. If the fix need to go upstream, probably one or two weeks
(time it takes to do some builds, testing and collect enough fixes to
make it worth issuing an update for everybody).
For rawhide it's a lot faster of course.

>  The fix for the ocfs2 problem was posted 
> in Ubuntus launchpad in October and given the fact that this bug prevents 
> ocfs2 from working completely I'm wondering why it's still present in 
> Fedora. 

Because nobody reported it to fedora so far? As upstream we cannot track
all possible sources of bugs. That is why every distribution/packager
has the golden rule to be the gateway for distro bugs to upstream.

> The only conclusion I can draw from that is that virtually nobody 
> uses ocfs2 in fedora which is why it didn't get reported until now.

This is entirely possible. Except that upstream does some development
directly in Fedora and most of it on Debian (there are tons of bugs
upstream has been trying to get fixed in Debian and that are not flowing
into Ubuntu).

> That's not necessarily a big problem per se but if this technology is 
> considered low priority in Fedora then that factors into my decision to 
> rely on it.

See.. you reported me a problem approx 10 minutes ago. Piped through the
proper lines, it's being addressed as we speak. Clearly signs of user
base will gather attention from the right people.

> 
> >>   the web
> >>> configuration tool described in the RHEL 5 manual for GFS is no longer
> >>> present in Fedora
> >
> > conga is part of Fedora 9/10/rawhide. Look for ricci, modcluster,
> > cluster-snmp and cluster-cim packages.
> 
> What about luci?

It's in the package list I posted before.

> 
> >>   and the cluster configuration tool doesn't work because
> >>> it tries to call command line tools in the wrong locations.
> >
> > system-config-cluster has been obsoleted. Unless you mean some other
> > tools, in that case please specify which.
> >
> >>> What I'm wondering is that if fedora is the basis for RHEL then what are
> >>> the plans for RHEL 6 regarding these technologies. AFAIK fedora 11 will
> >>> form the basis for that but right now GFS and/or clustering in general
> >>> seems to suffer from bit-rot.
> >
> > The plans for F11 are very well set for upstream and been communicated
> > across the board through public mailing lists.
> >
> > Anyway, so far I have read a lot of "bit-rot" this or "broken" that, but
> > you didn't provide any real info on the problems.
> > I'd be very happy to coordinate the fixes upstream and propagate them
> > into Fedora ASAP given enough info.
> 
> I think the basic problem is that I'm trying to feel my way into the 
> clustering field but run into a lot of conflicting information.
>  For example 
> I am supposed to look upstream for info (which I generally agree with) but 
> when I look at the Conga homepage the user manual tells me to install 
> "luci". Apparently that information is outdated. Since I have to start 
> somewhere I chose the latest "Red Hat Cluster Suite for Red Hat Enterprise 
> Linux" which talks about system-config-cluster which as I learned above is 
> basically no longer relevant.

Jim, Ryan: can we please update the conga web site to reflect a newer
status of things?

> 
> Is there some kind of "clustering lab" site out there that contains more 
> recent information about these technologies?

Not yet but this has been discussed at Cluster Summit 2008. All
information will be collected (from all different projects) in one
single umbrella that will work as reference point.

> 
> >>> I'm just surprised that GFS is a cornerstone of RHEL 5 but now seems to be
> >>> basically unmaintained or is the RHEL team maintaining forked versions of
> >>> these tools?
> >
> > RHEL doesn't have any forked tool. You can check upstream wiki and git
> > repositories yourself. All our development is done in the open.
> 
> That's what I thought and that's why I was so confused when running into 
> these problems.

the problem you are hitting with system-config-cluster is packaging
related. cman_tool in RHEL4/5 has a different location than F10/rawhide.
This is a known issue that will be addressed in rawhide. F10 still holds
some legacy bits to be backward compatible with F9.

>  The trouble I ran into didn't seem to make sense given the 
> success Red Hat is having with this stuff in the field so I decided to take 
> this to the brand new fedora-server list so I could perhaps get some more 
> insight into what's going on with this stuff.

Well some of them are valid issues, others are old/not updated
documentation that is still at RHEL5 level. Things have been evolving a
lot in the meantime and generally new docs are generated when a new
stable upstream release happens. So far (since RHEL5/STABLE2) there have
been no new major stable releases (with 3.0.0 upcoming in a couple of
months just for F11). So you are looking into a transition phase of the
software and expect to be everything already settled.

Fabio