Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[stacked] Spike: Auto Layer Ordering #384

Closed
wants to merge 23 commits into from

Conversation

schneems
Copy link
Contributor

Prefix layers with sequential numbers starting at 0001_<name>. To support this it assumes any layers on disk with a number prefix is equivalent i.e. changing the order of layers will not invalidate the cache.

This has the benefit that:

  • Layers can be named something semantic without side effects.
  • Build and launch layer behavior is guaranteed to be the same (provided read_env is called and applied for every layer in main).

Note that this places `/layers/heroku_ruby/gems/bin` before `/layers/heroku_ruby/binruby` on the path. Related to, but doesn't entirely fix #380.
Using `-a` also shows that there's more than one value present.
It's common and expected that Rails applications will include a `bin` directory containing "binstubs" of executables their app depends on. For example https://github.com/heroku/ruby-getting-started/tree/5e7ce01610a21cf9e5381daea66f79178e2b3c06/bin. They're largely used to ensure that bundler is invoked/used so that you can run `bin/rails` rather than needing to use `bundle exec rails`. However it's not strictly limited to only that.

This change: Adds the `bin` folder in the root of the workspace to the PATH and changes the layer to `venv` so it is loaded after other layers (and takes precedence in the case of a PATH prepend).

This fixes the previously committed failing test.

Close #380
The Application Contract does not specify that commands such as `rake -P` will be called with `bundle exec` and the classic buildpack does not rely on `bundle exec` internally. This brings the CNB closer to parity with the classic buildpack.

In the container environment, the first gems on the PATH should be those installed by the buildpack, negating the strict need to call `bundle exec` as you would on a development machine.

Usually prepending a Ruby command with `bundle exec` will have no discernible difference for an application that's bug free. This is evidenced by all tests passing with this change. However someone can commit their own `bin/rake` or `bin/rails` and we should use this over the executable installed via `bundle install`.
@schneems schneems force-pushed the schneems/add-user-binstubs branch from 6f97cb5 to b17681c Compare January 10, 2025 22:51
@schneems schneems force-pushed the schneems/auto-layer-ordering branch from c51ddc6 to 1076e8b Compare January 10, 2025 22:51
At runtime, the alphabetical order of the layer name determines the order it is loaded. At build time, the order that the `env` variable is modified determines the order.

At both build and runtime we want the bin stubs to come first on the PATH when executing user defined code. This was already working for runtime, but wasn't for build time as the "gems" layer was being prepended to the path after the "venv" layer (because the `venv` layer was being defined first, last definition wins).

I originally tried to fix this by defining the PATH inside of the "gems" layer along with the gems path but ran into heroku/libcnb.rs#899. The libcnb.rs project loads the user defined PATH modification last, but I'm unclear if that's spec defined behavior or not https://github.com/buildpacks/spec/blob/main/buildpack.md#layer-paths.
@schneems schneems force-pushed the schneems/add-user-binstubs branch from b17681c to c91cee5 Compare January 10, 2025 22:57
Prefix layers with sequential numbers starting at `0001_<name>`. To support this it assumes any layers on disk with a number prefix is equivalent i.e. changing the order of layers will not invalidate the cache.

This has the benefit that:

- Layers can be named something semantic without side effects.
- Build and launch layer behavior is guaranteed to be the same (provided `read_env` is called and applied for every layer in main).
@schneems schneems force-pushed the schneems/auto-layer-ordering branch from 1076e8b to b6820b6 Compare January 10, 2025 22:57
@schneems
Copy link
Contributor Author

I think it makes more sense to have a single function/macro that does it instead of trying to re-order things behind the scenes automatically, though we still need the conversion code.

@schneems schneems force-pushed the schneems/auto-layer-ordering branch from 40b17aa to 9e17152 Compare January 10, 2025 23:56
Base automatically changed from schneems/add-user-binstubs to main January 13, 2025 22:03
@schneems schneems closed this Jan 13, 2025
@schneems
Copy link
Contributor Author

Ed pointed out that not all binaries can be relocated and might embed path information in the binary itself. I think ordering needs to be a first-class feature from CNB and opened buildpacks/rfcs#322 to explore what the interface might look like based on my experiences here.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

1 participant