Re: Package conflicts, breaks, and problematic upgrades... with irqbalance

Thomas,

That works perfectly, thanks!

Technically, we do provide the functionality that irqbalance is supposed to provide for GCE so this makes sense to me. And there is literally no reason to install our software outside of GCE so I don't feel bad about saying we provide irqbalance.

We can discuss more about irqbalance if you want- but suffice to say, it does not work and has never worked for GCE and likely other hypervisors as well. We have not spent any significant effort trying to figure out why because we already had our own solution and didn't spend time debugging it.

On Tue, Feb 27, 2018 at 12:48 PM Zach Marano <zmarano@google.com> wrote:

Thanks Thomas,
So you are saying I should make our package:

Priority: optional
Provides: irqbalance
Conflicts: irqbalance

Would the Provides and Conflicts statement... conflict?

-----
Zach Marano
zmarano@google.com

On Tue, Feb 27, 2018 at 12:43 PM Thomas Goirand <zigo@debian.org> wrote:
On 02/27/2018 02:46 AM, Zach Marano wrote:
> According to the policy manual [3], using 'breaks' should work here
> because our package is installed already and the kernel is trying to
> install irqbalance on upgrade.

Usually, IIRC, Breaks: is for when 2 packages hold the same file. Here,
it likely would be Conflicts:.

> However, this doesn't seem to work
> correctly in all cases. Using just `apt upgrade` it correctly says it
> cannot install the kernel because it has conflicting dependencies. Using
> `apt-get upgrade` however still removes our package and installs
> irqbalance (same with apt-get dist-upgrade).

This wont happen if you add a Provides:. I believe Bastian thought
irqbalance had a Priority: required, but that's not the case, so it
should be fine to with Conflicts: + Provides:.

> What I did find is that
> changing the priority of our package to 'required' instead of 'optional'
> seems to work correctly with 'breaks'.

Do not do that! Your package certainly isn't to be required on all systems.

> From what I gather about
> 'required', it is reserved for things like the kernel package.

Exactly.

> So, what is the right way to fix this? And yes, I do believe the having
> irqbalance as a recommends for the kernel package is just wrong- however
> lets deal with that separately.

I do believe it's a nice thing that the kernel recommends it, as it is
useful on most system running the kernel.

Also, I agree with Bastian that we should get to understand why
irqbalance is so broken that you can't use it within GCE. I don't
understand this bit. Could you expand on this before taking drastic
measures like I just described? It'd be best to fix irqbalance to fit
your environment, IMO.

I hope this helps.

Cheers,

Thomas Goirand (zigo)

P.S: The HTML version of your message has tiny chars, and is therefore
almost unreadable. Please disable it, which is also the sensitive thing
to do for Debian lists.