[Spacewalk-list] How to update running applications?
Michael DeHaan
mdehaan at redhat.com
Thu Oct 9 21:22:52 UTC 2008
Gerhardus Geldenhuis wrote:
> Hi Adrian,
> I still think that a differentiation is called for between application
> deployment and os level deployment. If you have well over a 1000
> machines then investing time in looking at application level
> deployment mechanisms would be time well spend in my view.
> rpm/yum/spacewalk works well to distribute os patches but I do not
> think that one should spend to much time writing involved update
> scripts embedded in rpm's if there is available frameworks to
> distribute applications, especially if it is java based.
This is a bad route to take. Packages are packages.
First of all, it does not matter what language the deployment tools are
written in, and there is no major difference from updating OS plumbing
as applications. Pretty much all LAMP applications update fine via
yum without problems. Most desktop applications do. The idea that
you have to bring a database application down to upgrade it's database
is not language specific, it's an organizational thing. This is about
downtime windows.
It's another story that Java developers are typically not good about
packaging their software via RPM, but when there are RPM's, you have all
of the advantages of the packaging system.
But this thread really isn't about what is pushing something out, so
that's irrelevant -- what this thread was more about was "how do I
decide when to update software?". That's an organizational policy kind
of thing that doesn't depend on language or application, or whether
something is a libc update or a new openoffice.
Linux has no problem replacing /usr/bin/myapp while it's running without
affecting the running process. If it's a database application, then
yes, you want to stage your updates. The need to do that with respect
to downtime windows does /not/ change whether you have a J2EE app or
some other mission critical database/web application. It's just an
organizational thing, and yes, it's perfectly fine to be using Spacewalk
(or just yum) to do this -- lots of folks do.
The original post asked if downloading updates and applying them later
was a good idea. It might be, though such an application on top of
something like repotrack does not IIRC exist. It's a bit like the
"preupgrade" idea in newer Fedora but not so upgrade centric.
You could also just do something simple such as decide to run yum update
only at a certain time, which is simpler.
Ultimately something like Func may be ideally suited, where you could
just do:
Func "webservers*.example.org" call yum update
at 3:00 AM (from cron)
And you could update other servers at other times.
I am less familar with spacewalk's actions tooling but I suspect it can
do similar, either way you are still going through yum_rhn_plugin for
the updates.
Either way, I'd seriously look at doing something like that and training
your folks to understand scheduled outage windows for updates, and then
everything becomes much simpler.
>
> Regards
>
> On Thu, Oct 9, 2008 at 9:01 PM, Adrián Márques <amarques at geocom.com.uy
> <mailto:amarques at geocom.com.uy>> wrote:
>
> Thank you Michael for all of your valuable feedback.
>
> Yes, ooffice is not the best example. I just wanted to illustrate
> that operators will be working against a GUI, which can't just
> restart apparently out of its own volition whenever there's an
> update.
>
>
> For starters, RPM is a package format, yum is the updater.
> Deciding when you want to do updates is a good thing to be
> thinking about. Configuring spacewalk to /not/ pull down
> updates automatically is I presume doable (spacewalk folks?),
> in which case you can just push them out when you want
> (Spacewalk also have a facility to do this as well and can
> probably explain better).
>
>
> Maybe I wasn't clear with my previous mail. I do understand the
> difference between RPM and YUM. I know that from the spacewalk POV
> I could push the updates at specific times, but to the best of my
> knowledge this gives me no guarantee the updates will be applied
> to the clients at a specific time. Furthermore, I'm thinking of
> administering as much as well over a thousand hosts this way, in
> an environment were mantainance windows could be quite impractical
> to enforce. Thus I though of having the yum-rhn plugin download
> all rpms in the backgound, while the app might be running, and
> have them installed when I know I can restart the app without
> inconveniences. I'd also have to ensure that the app can't be
> started while an update is taking place.
>
> I hope with this I took care of any preceding ambiguities in my
> intentions.
>
> So I restate my questions: Do you consider this a terrible
> solution? Am I trying to defeat RPM/YUM/Spacewalk design with
> this? Is there a better alternative?
>
> Thank you all.
>
> Adrián.
>
> Michael DeHaan escribió:
>
> Adrián Márques wrote:
>
> Thanks a lot Michael.
> I'd never heard of condrestart before and your other
> suggestions also seem to be more than worth checking out.
>
> If I understood you correctly though, the advice you are
> giving me is geared towards updating services and in the
> process starting/stopping/restarting them based on
> conditions (I'll be looking into the links you provided to
> verify whether this is right). The conundrum I'm facing is
> that this is a desktop app, picture ooffice as an example,
> and I can't restart it while the users are logged in it.
>
> office doesn't have a backend service, so that may not be the
> best example.
>
> In either case, you'd want to deploy your new software in a
> window where it made since to do it.
>
> The hosts were this app will be running will be dedicated,
> so I'm leaning towards the idea of hacking the spacewalk
> client as I mentioned before (so it won't install the
> updates until told to) and periodically install downloaded
> rpms either at boot time or whenever I can ensure that no
> users are logged into this app. Do you consider this a
> terrible solution? Am I trying to defeat RPM with this? Is
> there a better alternative?
>
> For starters, RPM is a package format, yum is the updater.
> Deciding when you want to do updates is a good thing to be
> thinking about. Configuring spacewalk to /not/ pull down
> updates automatically is I presume doable (spacewalk folks?),
> in which case you can just push them out when you want
> (Spacewalk also have a facility to do this as well and can
> probably explain better).
>
>
>
> Thanks.
>
> Adrián.
>
> Michael DeHaan escribió:
>
> Gerhardus.Geldenhuis at gta-travel.com
> <mailto:Gerhardus.Geldenhuis at gta-travel.com> wrote:
>
> Hi Adrián,
> The question is probably more suitable on a rpm list.
> We don't distribute our jar's with rpm but I
> appreciate your dillema.
>
> I don't personnally think that for such complex
> applications, rpm is the right way to upgrade
> with. I see rpm has a way to distribute "core"
> stuff, and not suitable for distributing weird and
> wonderfull applications that requires restarts and
> database upgrades etc. There is application
> servers and frameworks that is better suited for
> this type of thing and that sits on a logically
> higher level. JBoss, Wesphere and I think even
> tomcat has some basic stuff build in.
>
> Regards
>
>
>
>
> This is generally true only for applications that are
> packaged /poorly/. Properly packaged RPMs can be
> configured to do "condrestart" as needed.
>
> If you need something more advanced, this is typically
> where you would want to look at something like
> puppet/cfengine/bcfg2 where you can describe the state
> you want your services to be (and remain) in.
>
> Also you may be interested in a simple tool such as
> Func (http://fedorahosted.org/func) for restarting
> services remotely. It is in many ways easier to
> deploy than SSH and also provides you a way of
> defining arbitrary groups of systems to send commands to.
>
> --Michael
>
> -----Original Message-----
> From: spacewalk-list-bounces at redhat.com
> <mailto:spacewalk-list-bounces at redhat.com>
> [mailto:spacewalk-list-bounces at redhat.com
> <mailto:spacewalk-list-bounces at redhat.com>] On
> Behalf Of Adrián Márques
> Sent: 06 October 2008 18:45
> To: spacewalk-list at redhat.com
> <mailto:spacewalk-list at redhat.com>
> Subject: [Spacewalk-list] How to update
> running applications?
>
> Dear all,
>
> I'm having a difficult time in finding the
> correct forum to pose this question since I
> guess it's part spacewalk, part RPM, part
> basic aplication design.
>
> Being an RPM end-user for years I had always
> assumed that updating running apps was a piece
> of cake with RPM. After all, I always did it
> and the most I got were occasional messages
> such as "Firefox restart required", never
> encountering other side-effects.
>
> However, after reading through the RPM guide,
> getting my hands dirty trying to package an
> app, scouring the internet for info and asking
> around in the RPM mailing list, I'm still not
> sure how to write my RPM so updating won't be
> an issue.
>
> The scenario is the following: I'd like to use
> spacewalk to distribute a custom desktop app.
> Hence, it's unavoidable that some users will
> be working at the time of an update. So far
> I've identified two problems with this, 1) If
> I have to modify the app's database schema as
> part of the update, the running app might
> break down if it's not aware of the change 2)
> This happens to be a java app, so in case a
> jar gets updated a class might get loaded that
> is not compatible with those already loaded by
> the running app.
>
> I understand that what I've explained so far
> might be outside of the scope of the subjects
> treated in this mailing list, if so, please
> just treat it as a detailed background problem
> description.
>
> What I would like to know from you guys is
> whether you've had any problems when updating
> apps through Satellite/Spacewalk due to them
> being running at the time or what precautions
> (if any) you take when doing so.
>
> Finally, I suppose that as a last resort I
> could hack the spacewalk client to only
> download updates and not install them until it
> is told to do so from the command line. Is
> this already possible? If not, where's the
> place to start with hacking this?
>
> I thank you all for your time and attention.
> It'd be amazing if you had any comments
> regarding my background issue, but I'll be
> more than satisfied and grateful if you could
> just provide any feedback on those questions
> strictly spacewalk-related.
>
> Thanks.
>
> Adrián.
>
>
>
>
> _______________________________________________
> Spacewalk-list mailing list
> Spacewalk-list at redhat.com <mailto:Spacewalk-list at redhat.com>
> https://www.redhat.com/mailman/listinfo/spacewalk-list
>
>
>
>
> --
> Gerhardus Geldenhuis
> Registered Linux User #193352
More information about the Spacewalk-list
mailing list