Re: [gentoo-user] Re: Safe way to test a new kernel?

[Robert David]

V Sun, 26 Feb 2012 02:50:46 +0700
Pandu Poluan <pandu@××××××.info> napsáno:

> On Feb 26, 2012 2:05 AM, "Robert David"
> <robert.david.public@×××××.com> wrote:
> >
> > V Sat, 25 Feb 2012 10:32:20 -0800
> > Grant <emailgrant@×××××.com> napsáno:
> >
> > > >> >> I need to test a kernel config change on a remote system.
> > > >> >> Is there a safe way to do this?  The fallback thing in grub
> > > >> >> has never worked for me.  When does that ever work?
> > > >> >
> > > >> >
> > > >> > You can press ESC in the Grub screen and it will take you to
> > > >> > text-only mode. There, you select an entry, press "e" and
> > > >> > edit it. Press ENTER when you're finished, and then press
> > > >> > "b" to boot your modified entry.
> > > >> >
> > > >> > That way, you can boot whatever kernel you want if the
> > > >> > current one doesn't work.
> > > >>
> > > >> I can't do that remotely though.  I'm probably asking for
> > > >> something that doesn't exist.
> > > >>
> > > >> - Grant
> > > >>
> > > >
> > > > Don't do that if you don't have some tool like KVM, or other
> > > > remote management of the server. Or if it is available in the
> > > > data center, just call them and order this service for the time
> > > > you need to do updates.
> > > >
> > > > This is why I don't use gentoo on servers any more, just because
> > > > I rather stay safe than sorry.
> > >
> > > How is another distro different in this situation?
> > >
> > > - Grant
> >
> > Just because when using distros like Centos/RHEL or Debian stable,
> > you have very little chance that the kernel released will fail. Due
> > to extensive testing, user base and update policy. And major kernel
> > update you done only once in few years and the transition is tested
> > before release done (though you are supposed to test yourself to be
> > safe).
> >
> > This is not saying that gentoo is bad, I'm very big fan of gentoo.
> > But you have to concern where it use and where not.
> >
> > Robert.
> >
>
> Anecdotal, but...
>
> I once had an Ubuntu VM that can't shutdown after a kernel update.
> First boot after update went well, but when I rebooted it again, it
> pegged its vCPUs at 100% before I ordered the Xen hypervisor to put
> it out of its misery.

I don't want to start flame, but ubuntu was never a system for server
for me. It may be good for desktop, but not server. For me ubuntu is
too up-to-date to be a good server distro, even the LTS is not
something well done, maybe in version x.x.3-4 of LTS it s reasonable to
put on server, but it will soon end with support. So nothing for me.

>
> The bug was apparently in the portion of the kernel running in the
> primary CPU that's responsible for shutting down the other CPUs
> before cutting the power. And IIRC, this bug affects all
> multi-processor configuration.
>
> So, as you can see, binary distros can still fuck up royal time. Not
> to mention that if you have an exotic configuration, support for your
> configuration might not be built into the kernel by the distro.
>

This is true, but you cannot compare countless options that gentoo can
be configured with few options that binary distro is capable and thus
more probably tested before.

> Somehow I believe people deploying Gentoo servers will be much more
> careful...

Agree. But real word is not so ideal. I got situations where I need
upgrade basic gentoo server to more special virtualized environment.
Just a customer wanted that. This consist about changing net to bridge,
add/update kernel modules, etc. I ended with bricked server after a
long checks etc, so I have to check the server physicaly. I did
another thing in debian remotely within a half hour.

This also implies that I'm not so good admin, but things happen not
so good as you expect every time.

I still use gentoo on my notebook and work computer, or specialized
project where it benefits. But all the servers are migrated to debian
or centos. I just don't have so much time to play.  

Robert.

>
> Rgds,