[Bug]: Yarn workspace foreach is considerably slower compared to other alternatives

[pastelsky]

Self-service

• I'd be willing to implement a fix

Describe the bug

I have a yarn workspace with around 20 packages. When I run a simple echo test —

"scripts": {
    "echo-test": "yarn workspaces foreach -v --parallel exec echo 'hi'",
}

It takes 14-15 seconds for echo to execute. I have 6 physical cores.

This is a lot compared to me running this directly through bash, or using other utilities like —
https://www.npmjs.com/package/workspaces-run

which take under 1 second here.

To reproduce

Unfortunately, I wasn't able to create a minimal repro for this. If I try to emulate the same workspace structure in a fresh project, I am not able to achieve the same high command execution times for yarn workspace foreach exec.

I did notice however, that increasing parallelism makes the performance worse. For example, on increasing this to parallel of 40, it takes 18+ seconds to execute.

I also noticed that when I run this, a node process with 150% CPU usage throughout the lifetime of the run appears.
This is strange, because the executed command itself isn't CPU intensive at all.

Environment

System:
    OS: macOS 11.6
    CPU: (12) x64 Intel(R) Core(TM) i7-9750H CPU @ 2.60GHz
  Binaries:
    Node: 14.17.4 - /private/var/folders/_j/128s9ngn26v2z9sx5c28bdd00000gn/T/xfs-38cb4655/node
    Yarn: 3.1.0-rc.8 - /private/var/folders/_j/128s9ngn26v2z9sx5c28bdd00000gn/T/xfs-38cb4655/yarn
    npm: 7.21.1 - ~/.nvm/versions/node/v14.17.4/bin/npm
  npmPackages:
    jest: ^26.6.3 => 26.6.3

Additional context

No response

[merceyz]

Duplicate of #2575

[merceyz]

While this issue mentions exec it's the same problem as #2575

[pastelsky]

@merceyz would that explain why it's only one node process that shoots up in CPU usage to 150%?
If this were #2575 , isn't it expected that the performance cost is paid by each newly spawned process instead?