Feature request: os: os.availableProcessors()

[NicolasSchwarzer]

Is your feature request related to a problem? Please describe.
It's unable to get actual available amount of processors in a docker container running on image node:10.16.0, of which cpus is limited by docker container resource constraints. For detail information, please have a look at issue #28762.

Describe the solution you'd like
Provide os.availableProcessors() api to get the actual available amount of processors.

Describe alternatives you've considered
If possible, it's better to fix the output of os.cpus().length, e.g. by using Proxy to change the value of length property. Because most node modules use os.cpus().length to decide how many processes it should fork, which may exceed the limitation of cpus in the docker container.

[silverwind]

I would first start investigating if Docker/Moby or their upstream projects can fix it.

Both/proc/cpuinfoand getconf _NPROCESSORS_ONLN are heavily used methods to obtain number of CPU cores but for some reason both do not reflect Docker's CPU limits.

[juanarbol]

I think this issue happens also with memory, using os.totalmem(), does not take container resources, It'll take the Docker engine resources, I guess

[NicolasSchwarzer]

@silverwind Okay, thank you, waiting for your good news!

[NicolasSchwarzer]

@juanarbol I've did some investigation, it seems work with memory，please see below:

$ docker container run --rm --memory="3072m" -it node:10.16.0 bash
root@e39346d06455:/# node
> const os = require('os');
undefined
> console.log(os.freemem());
76783616
undefined
> console.log(os.totalmem());
2096164864
undefined
> .exit
root@e39346d06455:/# exit
exit

[juanarbol]

@NicolasSchwarzer I got the same is total memory as you did, but I just gave 100mb as maximum memory usage, I think we're getting the Docker engine total memory also

$ docker container run --rm --memory="100m" -it node:10.16.0 bash
root@fe57c5bcdc46:/# node
> const os = require('os');
undefined
> console.log(os.freemem())
119603200
undefined
> console.log(os.totalmem())
2096164864 
undefined
> .exit
root@fe57c5bcdc46:/# exit

Also free memory is around 119.6mb, that's greater than my given 100mb to that container, and total memory is more than 1gb of ram

[NicolasSchwarzer]

@juanarbol You're right, I've tried and it shows exactly as what you described.

@silverwind Please also have a look at the same kind bug of os. freemem() & os. totalmem().

[silverwind]

Guess you misunderstood me, I don't indent to chase this down to the correct project. I found this which might point someone in the right directions (kernel or runc). Out of curiosity, I did check in nginx's source and found it relies on _NPROCESSORS_ONLN which is also affected by this issue.

Anyways, if we want to do something about this in node, it should probably interface with /sys/fs/cgroup but the issue remains that this is something very Linux-specific and Node.js aims to be cross-platform, so it's a rather bad candidate for inclusion in core I think.

[NicolasSchwarzer]

Okay, I get it. So is there a confirmed conclusion about whether to support in node? Indeed this is a platform-specific issue, I could raise an issue to docker if you confirmed not acceptable.

[silverwind]

I'd say its pretty certain that we wouldn't accept something only relevant to one platform. There have been various proposals for Linux-specific features in the past and most of them have been rejected.

[mhdawson]

@NicolasSchwarzer most of the platform specific code is in https://github.com/libuv/libuv. You might look there to see if you can find the code Node.js uses in libuv and have a discussion in the libuv repo if that should reflect the limits withing a docker container. I personally think being able to find the actual number of processors available is useful.

[NicolasSchwarzer]

@silverwind Thanks for your answering, I will close this issue.

@mhdawson Okay, I will investigate libuv later, thanks for your advice.