Node 16.4 or higher on M1 Mac OOMs in Qemu/Docker

[alex-hunt-materialize]

Version

v16.9.0

Platform

Linux buildkitsandbox 5.10.57 #1-NixOS SMP Sun Aug 8 07:05:24 UTC 2021 x86_64 GNU/Linux

Subsystem

No response

What steps will reproduce the bug?

1. Get an M1 Mac
2. Install Docker Desktop
3. Attempt to build a Docker image that does npm install using nodejs 16.4 or higher targetting --platform linux/amd64.
   RUN npm install --production

Additional details of scenarios that reproduce this are in the related Docker issue docker/for-mac#5831

How often does it reproduce? Is there a required condition?

100% reproducible, as far as I can tell. It doesn't matter how much RAM you give Docker.
According to the Docker issue, this doesn't seem to affect older versions of nodejs, just 16.4 and newer. We've confirmed it works with node 14.

What is the expected behavior?

npm install to work, without OOM.

What do you see instead?

The build fails with OOM.

#18 [build 11/16] RUN npm install --production
 
#18 0.328 
#18 0.328 #
#18 0.328 # Fatal process OOM in Failed to reserve virtual memory for CodeRange
#18 0.328 #
#18 0.328 
#18 0.328 qemu: uncaught target signal 5 (Trace/breakpoint trap) - core dumped
#18 0.330 Trace/breakpoint trap
#18 ERROR: executor failed running [/bin/sh -c npm install --production]: exit code: 133

Additional information

Opening this here in case the issue is with something nodejs is doing. Related Docker issue is at docker/for-mac#5831

[evanlucas]

Have you tried limiting the max-old-space-size? You can do that by setting the NODE_OPTIONS env var to something like NODE_OPTIONS='--max-old-space-size=512' where the number should be less than the available RAM.

[aaronhmiller]

I've been following this thru the 5831 issue. Assuming I used your suggestion properly @evanlucas, I just tried it, but same error occurs:

 => [linux/amd64 4/5] RUN npm install --max-old-space-size=4096                                                                       12.2s
 => => # #
 => => # # Fatal process OOM in CodeRange setup: allocate virtual memory
 => => # #
 => => # qemu: uncaught target signal 5 (Trace/breakpoint trap) - core dumped

No error if using Node 16.3.

[aaronhmiller]

x-ref from issue 5831 note...Using Node 16.10+ should resolve this. h/t to @shamelesscookie

[whatwewant]

v16.3.0 is ok
latest (v16.10.0, 2021.10.10) is bad

[whatwewant]

x-ref from issue 5831 note...Using Node 16.10+ should resolve this. h/t to @shamelesscookie

v16.10.0 is still broken

[cmoore1776]

Hmm. Well it's fixed for me, but I just checked and past versions that used to be broken are fixed as well.. so I guess Docker (or Apple?) implemented a fix?

I'm running Docker Desktop 4.1.0 (69386)
macOS Monterey 12.0 Beta 9 (21A5543b)
MacBook Air (M1, 2020)

[Tei1988]

I was in same trouble.

Just my experience, when I updated Docker for Mac from 3.6 to 4.1, It seems to be resolved.

Mac OS Big Sur 11.4
MacBook Pro (M1, 2020)

[scottbessler]

I have a similar OOM happening both on my local M1 and on circleci, so not sure its an architecture thing specifically.

However, it does not OOM on 16.3.0 but does OOM on 16.11.0

Its a proprietary code base so I can't share, but its while running a test suite for me.

[thclark]

As @Tei1988 says (thanks!) it's resolved by the latest version of Docker Desktop for Mac....

But note: that's the latest MAJOR version (4.x) - you may be like me and be confident that your docker desktop installation is up to date (because it bugs you to install updates every few days), but they don't automatically update across major versions. I was on a completely up-to-date 3.x 🤣 .