[PC-BSD Testing] Rolling release criticism

Kris Moore kris at pcbsd.org
Wed Dec 18 08:07:29 PST 2013


On 12/18/2013 07:44, Claudio L. wrote:
>
> On 12/17/2013 14:55, Joe Maloney wrote:
>> This also seems like a good idea to me if it doesn't cause too much
>> overhead or slow the project down a bunch I suppose. 
>
> Good point, we don't want to increase overhead. But if it's done in an
> organized way it can actually be done without putting too much extra
> work.
>
> For example:
> * Let's say the 1st of each month they do a "package freeze", where
> they select the packages that will be pushed in the next cycle.
> * The 15th of the month they finish preparing the set of packages and
> it gets pushed out to one branch.
> * There's 1 and a half months to fix any problems with this set of
> packages. And if a package can't be fixed within this time, it is
> removed from the set.
> * The last day of the second month, they push the package set out to
> the more stable branch.
> * ... and it loops ad infinitum...
>
> So they reduce the frequency of packages to a 2-month cycle, that
> should help Kris et al. to pace themselves (vs. the frenzy they have
> today with a monthly cycle or even faster), and we get a second branch
> with relatively low effort and a more polished set of packages.
> The 2-month cycle would introduce a delay of up to 2 months for the
> bleeding edge packages and up to 4 months for the stable one (this is
> a worst case scenario, for a package that comes out the day after they
> do the package freeze). I think it's reasonable.
>
>
>> That's why I was suggesting security and critical updates for RELEASE
>> only.
>
> But that would essentially freeze that branch for its life cycle,
> which goes against the idea of a rolling release. In my opinion it
> should lag behind the other one, but still roll. The static nature of
> FreeBSD gets a lot of criticism and it's one of the differentiating
> reasons to use PCBSD (or TrueOS) versus plain FreeBSD.
>
>
>> That way things like VirtualBox don't get borked when that package
>> get's updated. Like it is now. :( I know what you mean about spending
>> extra time fixing little things on what should be a production
>> system. Perhaps this every 2 months idea could work. I'd still say
>> apply the idea you just mentioned about the every 2 months to STABLE,
>> and roll a CURRENT release with the more current packages instead
>> that can trickle down into STABLE after they have recieved testing.
>
> That's slightly different from what I proposed above, it could work as
> well.
>
> Well, the ideas are now out there. I think we did our part, now it's
> up to the team to decide if and how they want to implement any of
> this. I'm positive that in 2014 we'll get back the stability we need.
>
> Claudio
>

I think these ideas are great. I also agree 100% with the renaming,
since RELEASE/STABLE is confusing, and I think those terms are best left
to the FreeBSD world/kernel portions. So here's what I think we will do:

1. As mentioned, create new PRODUCTION / EDGE set of packages.
  - Update EDGE every 2~ weeks, depending upon breakage of ports upstream.

  - Update PRODUCTION every quarter
    * One month before release, freeze the most recent EDGE set of ports
    * Begin testing / fixing major ports  - Desktops/VirtualBox, etc
    * Once testing is finished, push out new PRODUCTION set and unfreeze
EDGE again
   
  - Add selection mechanism to system manager GUI, letting you switch
between these sets
 

Does this sound like a reasonable timeframe for us to work on?

-- 
Kris Moore
PC-BSD Software
iXsystems



More information about the Testing mailing list