| Summary: | sci-libs/caffe2-2.2.1-r1: MAKEOPTS uses -j instead of -l threads | ||
|---|---|---|---|
| Product: | Gentoo Linux | Reporter: | APN-Pucky <alexander> |
| Component: | Current packages | Assignee: | Tupone Alfredo <tupone> |
| Status: | RESOLVED INVALID | ||
| Severity: | minor | CC: | alexander, ionen, tupone |
| Priority: | Normal | ||
| Version: | unspecified | ||
| Hardware: | All | ||
| OS: | Linux | ||
| Whiteboard: | |||
| Package list: | Runtime testing required: | --- | |
| Bug Depends on: | |||
| Bug Blocks: | 654178 | ||
|
Description
APN-Pucky
2024-03-19 11:07:33 UTC
ninja (or make) is allowed to start 113 compilation jobs and will stop adding jobs if the load is more than 16 %. -j is jobs and -l is load at the beginning the load is near 0 so it start all the jobs he can. But then the load goes very high. When one jobs terminate it will not start another jobs if the load is still over the threshold. Try with -j10 and -l1 and you will see 50 jobs started, and then only one will remains There are plenty of packages that either do not understand/respect load-average limitations or implement it poorly and may still spawn a lot of jobs at once as noted above, and then may not be able to use distcc either (don't believe it's handled for nvcc) to spawn these jobs on another machine. In general I don't think these MAKEOPTS are going to be safe and it may not be worth trying to handle this ebuild-side on every problematic packages. If want to keep using this, I'd suggest to set per-package exceptions using package.env (albeit that won't stop bad surprises the first time).
>
> Try with -j10 and -l1 and you will see 50 jobs started, and then only one
> will remains
you will see 10 jobs started (sorry)
|
