Try to automatically set maxWorkers correctly

[blimmer]

Background

Jest uses os.availableParallelism() and os.cpus().length to figure out how many Jest workers to spawn. However, on CircleCI Docker executors, these function calls return invalid results.

For example, on my medium+ executor with 3 vCPU cores:

circleci@00740471180a:~$ node
Welcome to Node.js v18.16.0.
Type ".help" for more information.
> const os = require('os')
undefined
> os.availableParallelism()
36
> os.cpus().length
36

This causes frequent OOM errors and crashes, which are hard for the average user to debug. There are several feature requests to CircleCI to make this better:

• Have nproc accurately reporter number of CPUs available to container
• Provide environment variables for resource limits

However, I'm not holding my breath on them fixing this on their end.

Possible Solution

It looks like using this snippet reports the correct number of vCPUs available:

circleci@2866814cd4c3:/tmp$ echo $(($(cat /sys/fs/cgroup/cpu/cpu.shares) / 1024))
3

So, it would be great to automatically set maxWorkers via this, instead of the inaccurate defaults used by jest.

Notes

• This would only be necessary on the docker executor, as the machine executor will correctly report CPUs: https://support.circleci.com/hc/en-us/articles/10653544013467-Docker-Executor-returning-Total-Host-CPU-Cores