FYI: main-amd64 type poudriere(-devel) bulk -Ca times on same system: 157 Hrs vs. 65 Hrs

The difference is based on using differently configured
poudriere-jail world builds:


) poudriere-jail world based on WITHOUT_MALLOC_PRODUCTION and
WITH_LLVM_ASSERTIONS for building for main-amd64: 157 Hrs, 
building 35456 packages. (An official PkgBase based jail.)

vs.

) poudriere-jail world based on WITH_MALLOC_PRODUCTION and
WITHOUT_LLVM_ASSERTIONS for building for main-amd64: 65Hrs,
building 35479 packages. (Personal build of the jail content.)


For both:

ALLOW_MAKE_JOBS=yes and MAKE_JOBS_NUMBER=3 were in use. Like
beefy18 (the official main-amd64 builder system), 12 builders
were in use, but there are only 32 FreeBSD cpus, not the 48
beefy18 has. The system used takes less time than beefy18 does.

The kernel in use was the same official PkgBase build of main's
GENERIC-NODEBUG . The boot world was from the same PkgBase
build as well.

The actual build times are both a fraction under the figures
listed.


For reference:

# uname -apKU
you have mail
FreeBSD 7950X3D-ZFS 15.0-CURRENT FreeBSD 15.0-CURRENT main-n278320-3a33e39edd48 GENERIC-NODEBUG amd64 amd64 1500048 1500048

See:

https://discord.com/channels/727023752348434432/1250652091702050826/1401211463728627765

and its following message for plots indicating how the
package-building rate changed over time during each
bulk -Ca build.


Notes:

Pkgs-Built per unit elapsed time is not a great measure
of progress: it is not uniform across over the mix of
tens of thousands of small-to-build port-packages vs.
the comparatively few of the large-build port-packages.

Still, comparing and contrasting the shapes of the of
the curves indicates a significant difference based on
the poudriere-jail world differences: it is more than
just a total time difference.

Too bad the likes of main-amd64 uses the same build
server for both production package builds and for
testing main-amd64. (main-aarch64 and main-armv7 would
be similar.) Running testing with the debug context
and production package builds the debug context would
increase the test coverage. (I've had examples of
debug builds working and non-debug builds panicking
before.) The more timely package distributions would
help with having more timely security updates available
for main-amd64 packages as well.


Odd idea relative to the just above?:

Something for beefy17 to be involved in when it stops
being the port-package builder for main-i386?


===
Mark Millard
marklmi at yahoo.com