Document somewhere that npm can't install dependency without preparing the environment

[serut]

Context

I had some serious issues to run my test on gitlab ci using your image node:7.10.1 and node:8.2.1.

Nothing is documented on the docker node webpage, so I though it was as simple as that (.gitlab.yml):

image: node

stages:
  - test

test:
  stage: test
  script:
   - npm install
   - npm test

To get the same bug as me, you need to define a postinstall script, which calls others script :

(1)    "postinstall": "npm run build:dll && npm run build:production-dll",
(3)    "build:production-dll": "npm run build:production-core-dll && npm run build:production-coreoss-dll",
(2)    "build:dll": "webpack --progress --config webpack.dev.dll.config.js --display-error-details",
(4)    "build:production-core-dll": "webpack --progress --config webpack.prod.core.dll.config.js --display-error-details",
(5)    "build:production-coreoss-dll": "webpack --progress --config webpack.prod.coreoss.dll.config.js --display-error-details",

Logs

On my CI logs, I can see these are executed :

• npm install is well executed
• postinstall is started
• the first part, build:dll ends up correctly
• then a simple WARN is printed but it does not stop the build npm WARN lifecycle [email protected]~postinstall: cannot run in wd %s %s (wd=%s) [email protected] npm run build:dll && npm run build:production-dll
• build:production-dll is not called, neither build:production-core-dll and build:production-coreoss-dll And that's it, it starts to test with an unexpected environment.

WTF is happening ?

When you run npm run <smth> as root, you need to define unsafe-perm to true to avoid any issue.

Solution

There is two solutions so far :

• create a new user and run tests using it
• Add the following to your test file, which allows npm to be launched as root

   - echo "unsafe-perm=true" > ~/.npmrc 

Can you document that on the docker image ?

[SimenB]

We do include a node user which is recommend to use (USER node).

I'd love to see a PR improving our documentation in that area, though! Could you send one?

[serut]

Hmm, maybe later but I must be sure about what I am doing.

• [ ] Create a PR to improve the README about user rights and a warning about npm and root.

When we mount the folder like what gitlab ci does, we have user / group rights on files. It looks like there is no problem with root, but I'm not sure that user node has rights well configured !

• [x] Test that node user has access to my app folder while testing

Is this command cross platform (I'm not sure about node alpine) ? :

image: node

stages:
  - test

test:
  stage: test
  script:
   - su - node
   - cd comes/back/to/my/app
   - npm install
   - npm test

[serut]

Ok I tryed to use USER node instead of root and :

• the workspace that gitlab mounts is owned by root (at least in terms of uid gid) with chmod 666, so there is no problem with rights.
• ~~your USER node also has write access to /usr/local/lib/node_modules (no problem to install -g and npm link)~~ USER node can't use install -g and npm link
• ~~However, we loose gitlab ci environment variables by switching to another user.~~
• su is cross OS

What do you think about it ?

[serut]

We should also add an explanation about that log :

npm WARN lifecycle <module name>~<lifecycle step>: cannot run in wd %s %s (wd=%s)

Which basically means that your app have been misinstalled and will not work properly.

[gibfahn]

your USER node also has write access to /usr/local/lib/node_modules (no problem to install -g and npm link)

Really? When I tried this with node:8.4.0 and node:8.4.0-alpine I get permission denied on that folder.

# Alpine
/ $ ls -l /usr/local/lib | grep node_modules
drwxr-xr-x    3 root     root          4096 Aug 15 22:21 node_modules

# Debian
node@d5958ee4adde:/$ ls -l /usr/local/lib | grep node_modules
drwxrwxr-x 3  500   500 4096 Aug 15 16:24 node_modules
node@1ae43601ee81:/$ id node
uid=1000(node) gid=1000(node) groups=1000(node)

[gibfahn]

However, we loose gitlab ci environment variables by switching to another user.

It's odd that you're losing environment variables, they seem to be kept if you switch inside the image:

root@1ae43601ee81:/\# export foo=bar
root@1ae43601ee81:/\# su node
node@1ae43601ee81:/$ echo $foo
bar

[LaurentGoderre]

su -E preserves env var I believe

[serut]

It's odd that you're losing environment variables, they seem to be kept if you switch inside the image

Aaaalright ! That's my fault, I should have run su node instead of su - node. I tried the solution that propose @LaurentGoderre but su -E is not available on alpine :

/ # su -E node
su: unrecognized option: E
[..]
        -,-l    Clear environment, go to home dir, run shell as login shell

And I just end the research there. I saw somewhere that the user node was created and configured but no it's not configured to handle install -g and npm link ! Nice catch ! I will stroke everything that's not correct :S

[gibfahn]

I saw somewhere that the user node was created and configured but no it's not configured to handle install -g and npm link ! Nice catch ! I will stroke everything that's not correct :S

That's true, but you can fix it by doing something like:

export npm_config_prefix=$HOME/.npm

after you su node.