Re: llvm10 build failure on Rpi3

From: Mark Millard via freebsd-ports <freebsd-ports_at_freebsd.org>
Date: Wed, 23 Jun 2021 17:58:56 -0700
Misc notes . . .

Looking at your logs, I expect trying to build both
llvm10 and rust in parallel is likely to run into
resource issues on teh RPi3B+. For builds in that
context, it may be better to do something like:

# poudriere buld -j main devel/llvm10
# poudriere buld -j main lang/rust
# poudriere buld -j main -f SOMEFILE-LISTING-OTHER-ORIGINS

based on using ALLOW_MAKE_JOBS=yes .

Part of this I based on your on-going llvm10-10.0.1_5
build shows load averages (example):

4.53 4.49 4.40

so all 4 cores are busy with a little backlogged work
already. It is also part of the explanation for:

bad_C++_code	24:30:53 for both building at the same time
vs.
bad_C++_code	06:59:12 for only llvm10 building.

(I'm not making claims for overall elapsed time.)

You wrote in http://www.zefox.org/~bob/readme :

MAX_EXECUTION_TIME_PACKAGE=432000 (since increased to 1724000, builds still stop at 24 hours) 

I think you may have guessed wrong about what
MAX_EXECUTION_TIME_PACKAGE covers: it is for
after staging the build, just creating the package
from the staged material. It is not for the
overall time turning the port into a package.

The time to build (through staging?) is controlled
by something you have left commented out and have
not adjusted:

# This defines the max time (in seconds) that a command may run for a build
# before it is killed for taking too long. Default: 86400
#MAX_EXECUTION_TIME=86400

My prior notes had listed:

# Cortex-A53 and such are slow for the purpose, allow 4 times the defaults:
MAX_EXECUTION_TIME=432000

But the figures that I'd used never dealt with something like
rust on something like an RPi3B+. So the figure may well be
too small even if rust is never built in parallel with anything
else. (A similar point goes for all my example MAX_EXECUTION_TIME*
figures.) I did do various llvm* builds, but rust is bigger than
any one of those by a long shot.

Parallel builds of things like llvm10 and rust in significantly
overlapping time frames put the load average well over 4 and
likely cause periods if significant paging/swapping. This can
greatly expand the elapsed-time for the individual jobs (builders).


===
Mark Millard
marklmi at yahoo.com
( dsl-only.net went
away in early 2018-Mar)
Received on Thu Jun 24 2021 - 00:58:56 UTC

Original text of this message