CentOS virt - Jan 2016 - CentOS 6 Virt SIG Xen 4.6 packages available in centos-virt-xen-testing

On Thu, Jan 21, 2016 at 12:01 PM, Peter <peter at pajamian.dhs.org>
wrote:
> On 15/01/16 05:57, George Dunlap wrote:
>> As mentioned yesterday, Xen 4.6 packages are now available for
>> testing.  These also include an update to libvirt 1.3.0, in line with
>> what's available for CentOS 7.  Please test, particularly the
upgrade
>> if you can, and report any problems here.
>
> Per conversation in IRC, Xen 4.6 no longer includes xend and therefore
> no longer has the "xm" command.  This is problematic for people
who may
> be using xm in various scripts on their host (such as home-brewed backup
> scripts).
>
> I think it's a bad idea to break this functionality without warning by
> allowing a simple "yum update" to remove it.  You will take a lot
of
> people by surprise and cause such scripts to stop working, if people are
> running yum cron the situation becomes even worse.
Thanks, PJ, for your input.

Just to be clear:

1. In the Xen 4.4 packages (first released October 2014), xend was
disabled by default; so anyone using xend at the moment has already
manually intervened to enable deprecated functionality

2. In 4.4, the first time xm was executed, it printed this warning:
---
xend is deprecated and scheduled for removal. Please migrate to another
toolstack ASAP.

See http://wiki.xen.org/wiki/Choice_of_Toolstacks for information on
other alternatives, including xl which is designed to be a drop in
replacement for xm (http://wiki.xen.org/wiki/XL).
---

3. ...and on every subsequent invocation, it printed this warning:
"WARNING: xend/xm is deprecated"

I think this constitutes "warning" that the functionality was going to
break at some point. :-)

Also, in most cases "s/xm/xl/g;" Just Works; most people have reported
that changing from xm -> xl was pretty painless.  So this isn't like
upgrading from Python 2 to Python 3 (or QT 4 to 5, or...).
> I think that due to this lack of backwards compatibility with Xen 4.4
> and earlier versions it would be a good idea to not force the upgrade on
> people who are not wary of it.  I propose that the new packages carry
> the name "xen46" and they purposefully conflict with the old
"xen"
> packages.  That will require people to take positive action to do the
> upgrade and hence avoid breaking systems unintentionally.
This would avoid breaking things for people still using xm, which
certainly has some value.  However it has some costs:

* The packages between C6 and C7 will now be slightly different,
increasing the maintenance burden.  This is not only in the spec file,
but also in all the associated scripting machinery for managing
packages in the CBS and smoke-testing packages before pushing them
publicly.

* Instructions for installing Xen are now differend between C6 and C7,
and slightly more complicated, as they have to explain about Xen 4.6
vs alternatives.

* Users who have heeded the warning and switched to xl will have to
make an extra effort to switch to Xen 4.6.  If they don't follow
centos-virt, they may not notice that there's a new package to upgrade
to.

I'm a developer, not a server admin, so I can't gauge how important
this issue is.  Before making such a change, I'd like to hear opinions
from other people in the community about how important (or not) it is
to avoid breaking xm, given the ample warning (>1 year) users have
had.

On the other hand, explicitly moving to a "xen${VER}" (both for C6 and
C7) would make it simpler for people to step up and maintain older
versions in parallel if anybody wanted to do so.

Thanks again, Peter, for bringing this up.

Peace,
 -George

On 01/21/2016 04:32 AM, George Dunlap wrote:
> I'm a developer, not a server admin, so I can't gauge how important
> this issue is.  Before making such a change, I'd like to hear opinions
> from other people in the community about how important (or not) it is
> to avoid breaking xm, given the ample warning (>1 year) users have
> had.
> 
> On the other hand, explicitly moving to a "xen${VER}" (both for
C6 and
> C7) would make it simpler for people to step up and maintain older
> versions in parallel if anybody wanted to do so.
My inclination is towards a naming scheme like xen46, xen48, etc + a meta
package that always depends on the latest. It should be more obvious when
there's a major upgrade, and those who can't afford a major upgrade can
uninstall the meta package.

For the record, we have no particular desire for xen 4.4 but haven't done
enough testing to say xen 4.6 is good yet.

On 22/01/16 01:32, George Dunlap wrote:
> 1. In the Xen 4.4 packages (first released October 2014), xend was
> disabled by default; so anyone using xend at the moment has already
> manually intervened to enable deprecated functionality
Xen4CentOS was first available in CentOS 6.4 with Xen 4.2, so we really
need to look back that far.  You may be right here, though, because Xen
4.2 was the first version of Xen that used xl by default.  xl was still
new at the time, though, and there were admins, especially ones who were
migrating from a CentOS 5 based Xen, who would have preferred XM and
were already using XM in other installs, so xm would have still seen
widespread use at the time.  That means that those boxes could very well
still be using xm now.
> 2. In 4.4, the first time xm was executed, it printed this warning:
The warning was not in Xen 4.2 or 4.3.  Considering that it is
reasonable for someone to expect enterprise-like behavior from an
enterprise distro, one would not assume that XEND would just disappear
overnight.
> Also, in most cases "s/xm/xl/g;" Just Works; most people have
reported
> that changing from xm -> xl was pretty painless.  So this isn't like
> upgrading from Python 2 to Python 3 (or QT 4 to 5, or...).
There are some notable differences that can break compatibility in many
cases:

1.  xl does not support the "xl create foo" syntax, one must now use,
"xl create /path/to/foo.cfg".  This I can actually see breaking
compatibility for a *lot* of people.

2.  xl requires those that use the old xend networking setup to switch
to a distro-based networking setup.  This can be a serious undertaking
for someone running a production machine with the easy possibility of
breaking networking and requiring either OOB or physical access to the
machine to fix.  Therefore this in itself means that someone should be
able to plan for the upgrade at a convenient time rather than having it
sprung on them with a "yum update".

3.  Config files no longer support embedded python.

...There are additional incompatibilities that could bite an unwary
admin in the ass.
> This would avoid breaking things for people still using xm, which
> certainly has some value.  However it has some costs:
> 
> * The packages between C6 and C7 will now be slightly different,
> increasing the maintenance burden.  This is not only in the spec file,
> but also in all the associated scripting machinery for managing
> packages in the CBS and smoke-testing packages before pushing them
> publicly.
> 
> * Instructions for installing Xen are now differend between C6 and C7,
> and slightly more complicated, as they have to explain about Xen 4.6
> vs alternatives.
> 
> * Users who have heeded the warning and switched to xl will have to
> make an extra effort to switch to Xen 4.6.  If they don't follow
> centos-virt, they may not notice that there's a new package to upgrade
> to.
These are certainly concerns, but there is precedence for doing it this way:

mysql (5.0), mysql51, mysql55 in EL5.
bind (9.3), bind97 in EL5.
...etc

These are all cases of Red Hat offering newer versions without wishing
to break older installs, and at least in the case of mysql, dropping
support for the older version alltogether, but still doing it in a way
that doesn't *force* the admin to upgrade and break their systems as a
result.

I do believe that looking to what Red Hat has done in the past is
appropriate when determining what action to take ourselves as people
expect the same level of stability from CentOS, and consequenty Xen4CentOS.
> On the other hand, explicitly moving to a "xen${VER}" (both for
C6 and
> C7) would make it simpler for people to step up and maintain older
> versions in parallel if anybody wanted to do so.
This is true, and I'd leave that choice up to you.


Peter

On Thu, Jan 21, 2016 at 7:17 PM, Sarah Newman <srn at prgmr.com>
wrote:
> On 01/21/2016 04:32 AM, George Dunlap wrote:
>
>> I'm a developer, not a server admin, so I can't gauge how
important
>> this issue is.  Before making such a change, I'd like to hear
opinions
>> from other people in the community about how important (or not) it is
>> to avoid breaking xm, given the ample warning (>1 year) users have
>> had.
>>
>> On the other hand, explicitly moving to a "xen${VER}" (both
for C6 and
>> C7) would make it simpler for people to step up and maintain older
>> versions in parallel if anybody wanted to do so.
>
> My inclination is towards a naming scheme like xen46, xen48, etc + a meta
package that always depends on the latest. It should be more obvious when
> there's a major upgrade, and those who can't afford a major upgrade
can uninstall the meta package.
BTW, we had a discussion about this particular idea at the Virt SIG
meeting, and KB said that different naming of packages like this can
cause dependency problems.  For example, a package depends on
"xen-version >= $N" will fail because as far as rpm is concerned,
you
don't have package 'xen' installed at all.

Another idea is to keep separate repos, and then have
"centos-release-xen-$VERSION", which would always point to $VERSION,
and "centos-release-xen", which would always be the newest version.
That might satisfy both types of people.

 -George