Nope, I don't use it on this machine (only 8 Go RAM). I'll try with one job, and if it fails again, I'll buy an older workstation (with a i5 CPU / 16 GO RAM) and set a poudriere on it, at my home. It will be less expensive than using ec2 instances and ressources.
?
PACKAGE_FETCH_BRANCH=latest
PACKAGE_FETCH_WHITELIST="rust"
PACKAGE_FETCH_URL=pkg+http://pkg.FreeBSD.org/\${ABI} pkg info | grep rust returns:rust-1.71.0 Language with a focus on memory safety and concurrency poudriere bulk -j 132Ramd64 security/py-certbot, it still trying to build devel/rust...
DEFAULT_VERSIONS+=pycryptography=legacy
My old i5 (3470; 2 cores/4 threads; 16 GB) runs ZFS and poudriere. It's my "build" server. Yeah, it's not the fastest in the world, but it works just fine.
for j in ${SERVER_REPOS}; do
${POUDRIERE} bulk -j ${j} -p server -f ${BASEDIR}/rust.lst
${POUDRIERE} bulk -j ${j} -p server -f ${BASEDIR}/gcc.lst
${POUDRIERE} bulk -j ${j} -p server -f ${BASEDIR}/llvm.lst
${POUDRIERE} bulk -j ${j} -p server -f ${BASEDIR}/${j}-server-package.lst
doneYes, default setting of Poudriere.
USE_TMPFS=yestmpfs on /tmp (tmpfs, local)Sure it can help (and I still think it's the most convenient solution if you don't fiddle with anything that would affect the rust package). It just requires these extra steps making sure your ports tree is on the exact commit from which the package repo was built to work reliably....
Here's the script you posted in another thread to get the port hash - it does work but I haven't incorporated it into my build script to use Poudriere's Fetch feature:
# List of package globs that are not allowed to use tmpfs for their WRKDIR
# Note that you *must* set TMPFS_BLACKLIST_TMPDIR
# EXAMPLE: TMPFS_BLACKLIST="rust"
update-ports command is the one that does what zirias is describing:#!/bin/sh
commit=$(fetch -q -o - 'https://pkg-status.freebsd.org/api/1/builds?type=package&jailname=132amd64' | fx 'x["builds"].find(b => b.mastername == "132amd64-default" && b.status == "stopped:done:").buildname')
if [ -z $commit ]; then
echo 'Could not find completed build. Still running? https://pkg-status.freebsd.org/' 1>&2
exit 1
fi
cd freebsd-ports.git/main && git remote update && git reset --hard $commit-b latest (as used in the build-prod command) to fetch packages.-c to delete previously downloaded packages, because options have changed between the fetched version and git, and poudriere thinks it needs to rebuild.-n -v to see what triggers a build for a package that you expect to be downloaded.
As weird as it sounds, in such a case it works even better to just interrupt poudriere after it deleted packages not matching any more and immediately restart it ... then it will download just these packages ?
Another try with a better ec2 instance, with 16 Go RAM, 8 Go swap (no ZFS, /tmp as tmpfs, with one job) : stil a fail after 5 hours.
I understand your point. What I was trying to say is, for me, FreeBSD has always been a great OS for light and old machines. I use it since 2006, with... 256 Mo RAM (I could even build src with this). Of course, I had to increase the RAM and CPU years after years, yes. But increase from 4 Go to 64 Go is a huge gap... for a single port.
DEFAULT_VERSIONS+=pycryptography=legacy