pcntl_fork children are utilizing just a single core on FreeBSD 13.3+

[instaproofs]

pcntl_fork children are utilizing just a single core on FreeBSD 13.3+

[devnexen]

The next PHP major release will allow to have more control over processes being bound to specific cores/cpus (via pcntl). For now, maybe there is a change in term of cpu topology policy ? you may look at the following value

sysctl kern.sched.topology_spec

and see if changing it has an impact. But I may have a better look later.

[instaproofs]

We have narrowed this down to it being somehow related to Imagick. If we run the same code but do anything besides load Imagick in the forked processes, it works as expected and spreads the forked processes to all of the available CPUs. For example, if the images are resized using the GD library instead, it works properly. Or if we just read from /dev/random, it works properly as well.

The same script, using Imagick, works properly on everything below FreeBSD 13.3, so it's still a mystery as to why it is not functioning correctly after updating. If anyone has suggestions, we are still all-ears. Thank you!

[devnexen]

quick question: Do you know if your imagemagick port is (and/or was) compiled with openmp support ?

[kmpanilla]

quick question: Do you know if your imagemagick port is (and/or was) compiled with openmp support ?

Yes, our ImageMagick7 is compiled WITH OpenMP support. Funny enough, with OpenMP disabled, forked processes use all the CPU cores as expected. With OpenMP compiled in, the results are limited to one CPU.

[devnexen]

Good to know :-). Anyhow, not a php issue.