Re: Making pbulk build just one new package

["J. Lewis Muir" <jlmuir%imca-cat.org@localhost>]

On 10/15/15 4:19 AM, Jonathan Perkin wrote:
> * On 2015-10-15 at 03:49 BST, J. Lewis Muir wrote:
> 
>> I like having pbulk build the package because that ensures it's built
>> in exactly the same way as the other packages I've built.  To get pbulk
>> to build just one package, I move pkglist to pkglist.orig and create
>> pkglist with just one line: the package I want pbulk to build (e.g.
>> "local/chkver").  I then run bulkbuild in the chroot.  Unfortunately, it
>> spends about 5 minutes scanning and resolving before it starts to build
>> that one package.  Is there a better way?
> 
> It's doing what it needs to do, but there are ways to optimise it:

Hi, Jonathan!

Thanks for the suggestions!

>  - Ensure you have reuse_scan_results=yes set in pbulk.conf

Yep, I've got that set.

>  - When running a limited build, I usually limit the number of scan
>    chroots a lot.  Unlike a full build, limited builds require
>    multiple scan phases, each of which will run through chroot
>    setup and teardown, which can quickly become the bottleneck.
> 
> For only one or two packages in pkglist I'd use just one scan chroot.
> 
> In our SmartOS 'tools' builds, which have a pkglist of around 40
> packages, I use 4 (compared to the full runs of 12+), though if this
> was OSX I'd reduce it even further as OSX sandbox creation is a lot
> more intensive.

OK, good to know.  I was using build_chroots=4 and scan_chroots=4.  I
tried with scan_chroots=1, but unfortunately that didn't really provide
a noticeable speed-up.

> Hope that helps,

Thanks, it does; I now know it's about as fast as can be expected.

Regards,

Lewis