Giters

containerbase / base

The containerbase project's base image source

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Support readonly root filesystem

viceice opened this issue · comments

commented

general

  • /usr/local/bin is symlinked to /opt/containerbase/bin
  • /usr/local/lib is symlinked to /opt/containerbase/lib
  • tool prepare state is stored at /var/lib/containerbase
  • all cache files are stored at /tmp/containerbase
  • files and folders in user home are symlinked to /tmp/containerbase/cache (eg .cache, .npmrc, .npm)

binarySource=global

  • needs /tmp writable

binarySource=install

  • needs /tmp writable
  • needs /opt/containerbase writable

related issues

commented

@rarkins There is one more issue! npm always tries to write to ~/.npm/_logs.

commented

It cannot be stopped??

commented

not i'm aware of. i'm now redirecting home to /opt/containerbase/home, see linked PR.

We need to think about the existing data at /opt/containerbase when mounting a volume which will hide all existing. 🤔

commented

I think we need to split installs again.

So when running as root and it's a docker build, then install tools to some other persistent path.
Otherwise all files and prepare information are gone when mounting a volume to /opt/containerbase 🤔

I would add a new containerbase-cli restore command, which recreates all missing folders and files.
It then symlinks the root installed files there, so the layout looks like now.

containerbase-cli restore will be called automatically by the entrypoint script and should be called by custom images with their own entrypoint.

@rarkins WDYT? Can you follow me?

commented

I'm not sure I fully follow, but I guess you mean that when the user mounts a volume to /opt/containerbase then anything we've put there prior is lost. So you're planning to put it in another location and symlink it over whenever the container runs?

commented

Yes, that's the core idea.

When later running install-tool as user it will simply install to /opt/containerbase as now.
I can propably also check and run the containerbase-cli restore command implicit from install-tool when something is missing inside /opt/containerbase.

commented

Would it be a viable alternative to allow the user to configure the containerbase rw location via env variable? If configured then the restore happens from /opt/containerbase otherwise /opt/containerbase is used? Or is it too hard to set up PATH etc?

commented

we can't ensure path then. a user would need to explicit set it 😞