Reducing size of the weekly build

Reducing size of the weekly build

[Flanagan, Elizabeth]

Folks,

In an effort to speed up release times, I'm going to remove the deb
and rpm publishing from the weekly builds. We'll still have ipk
publishing and we will still generate debs and rpms, but as far as I
can tell, no one is actually utilizing the debs/rpms.

One thing I'm thinking of and I want some feedback here is setting the
package type on the main builds to just ipk and running a special
deb/rpm test being to ensure those package types are functional. This
should give us a bit of a speed increase for build times, without
introducing too much risk. Thoughts?

-b

-- 
Elizabeth Flanagan
Yocto Project
Build and Release

Re: Reducing size of the weekly build

[Burton, Ross]

Hi Beth,

On 21 May 2014 17:22, Flanagan, Elizabeth <elizabeth.flanagan@intel.com> wrote:
> In an effort to speed up release times, I'm going to remove the deb
> and rpm publishing from the weekly builds. We'll still have ipk
> publishing and we will still generate debs and rpms, but as far as I
> can tell, no one is actually utilizing the debs/rpms.

The QA scripts perform install tests, do they use the published feed,
or another local copy?  Breaking QA's ability to validate the the
package manager actually works would be a problem.

> One thing I'm thinking of and I want some feedback here is setting the
> package type on the main builds to just ipk and running a special
> deb/rpm test being to ensure those package types are functional. This
> should give us a bit of a speed increase for build times, without
> introducing too much risk. Thoughts?

The reduced coverage is something to consider, but if the other builds
were world and also built images then the reduction shouldn't be too
bad.  I wonder what the speed difference between ipk and ipkg+rpm is:
this is probably worth measuring.

Ross

Re: Reducing size of the weekly build

[Richard Purdie]

On Wed, 2014-05-21 at 19:58 +0100, Burton, Ross wrote:
> On 21 May 2014 17:22, Flanagan, Elizabeth <elizabeth.flanagan@intel.com> wrote:
> > One thing I'm thinking of and I want some feedback here is setting the
> > package type on the main builds to just ipk and running a special
> > deb/rpm test being to ensure those package types are functional. This
> > should give us a bit of a speed increase for build times, without
> > introducing too much risk. Thoughts?
> 
> The reduced coverage is something to consider, but if the other builds
> were world and also built images then the reduction shouldn't be too
> bad.  I wonder what the speed difference between ipk and ipkg+rpm is:
> this is probably worth measuring.

We do need to be careful about coverage here. We have seen bugs where
things failed in deb or rpm but not ipk and maybe only on some
architectures.

I'm fine with being more selective about what we publish however in the
builds it may be worth leaving this enabled.

Cheers,

Richard

Re: Reducing size of the weekly build

[Flanagan, Elizabeth]

On Wed, May 21, 2014 at 1:30 PM, Richard Purdie
<richard.purdie@linuxfoundation.org> wrote:
> On Wed, 2014-05-21 at 19:58 +0100, Burton, Ross wrote:
>> On 21 May 2014 17:22, Flanagan, Elizabeth <elizabeth.flanagan@intel.com> wrote:
>> > One thing I'm thinking of and I want some feedback here is setting the
>> > package type on the main builds to just ipk and running a special
>> > deb/rpm test being to ensure those package types are functional. This
>> > should give us a bit of a speed increase for build times, without
>> > introducing too much risk. Thoughts?
>>
>> The reduced coverage is something to consider, but if the other builds
>> were world and also built images then the reduction shouldn't be too
>> bad.  I wonder what the speed difference between ipk and ipkg+rpm is:
>> this is probably worth measuring.
>
> We do need to be careful about coverage here. We have seen bugs where
> things failed in deb or rpm but not ipk and maybe only on some
> architectures.
>
> I'm fine with being more selective about what we publish however in the
> builds it may be worth leaving this enabled.

Yeah, I can see that. If we had a world rpm/deb as Ross suggests, I'm
not sure that would actually give us a speed increase, so for now, I'm
ok with leaving what we build alone. But yeah, let's not publish the
rpm/debs.

>
> Cheers,
>
> Richard
>



-- 
Elizabeth Flanagan
Yocto Project
Build and Release