build driver-os-monitor (0.5.2-dev) fails with seg fault

[cgreenhalgh]

When I try to build driver-os-monitor, branch 0.5.2-dev, npm/node exit with segmentation fault (e.g. at npm install step):

$ docker build -t databoxsystems/driver-os-monitor-amd64:0.5.2  .
Sending build context to Docker daemon  262.1kB
Step 1/10 : FROM amd64/alpine:3.8
 ---> 491e0ff7a8d5
Step 2/10 : WORKDIR /app
 ---> Using cache
 ---> 4d2f5e899dae
Step 3/10 : RUN addgroup -S databox && adduser -S -g databox databox && apk --no-cache add build-base pkgconfig nodejs npm libzmq zeromq-dev libsodium-dev python  && npm install zeromq@4.6.0 --zmq-external --verbose && apk del build-base pkgconfig libsodium-dev python zeromq-dev
 ---> Using cache
 ---> 97161b42183f
Step 4/10 : ADD ./package.json package.json
 ---> Using cache
 ---> 28b0e780cad0
Step 5/10 : RUN npm install --production && npm run clean
 ---> Running in f363a5802d4f
Segmentation fault
The command '/bin/sh -c npm install --production && npm run clean' returned a non-zero code: 139

strangely, if i edit the Dockerfile to end at that point and run it then node / npm segfault, but if I apk del then re-add npm it seems to work again. So the seg fault only appears when npm has been installed at an earlier step in the docker build (i.e. in a previous build container).

This is on an alpine VM running Docker version 18.09.1-ce.

[Toshbrown]

Hi Chris,

I've just run docker build -t databoxsystems/driver-os-monitor-amd64:test . --no-cache on my build machine and can't replicate the issue. Can you rerun with --no-cache and see if it builds?

[cgreenhalgh]

I have replicated this issue but only on an Alpine VM (e.g. it doesn't occur in Docker for Mac or in a Debian Stretch VM). It is repeatable with --no-cache.

The Alpine VM Is based on Vagrant's "alpine/alpine64" base image, plus

sudo apk update
sudo apk upgrade

and docker installed as per alpine docs

I've confirmed that alpine (which didn't work) and debian (which did) both report docker is using the overlay2 storage engine.

So my best guess is that there is something different about the overlayFS support in the alpine kernel vs debian, or something like that :-/

So I'm using debian for now in Intel.

[Toshbrown]

would that be Alpine 3.9 or 3.8?

[cgreenhalgh]

/etc/alpine-release says 3.9.0.

/etc/apk/repositories are

http://nl.alpinelinux.org/alpine/v3.7/community
http://nl.alpinelinux.org/alpine/edge/main
http://nl.alpinelinux.org/alpine/edge/community