Potential issues when deploying with capistrano

[rachitpant]

Description 📖

Thank you for your help yesterday and for this gem in general. I am trying to deploy a rails application that I migrated from webpacker to vite to our staging environment. I was able to deploy a working copy but there are some issues with the contents of public directory on the server that I have tried to describe below. You will notice my publicOutputDir is "". I am not sure if the following are actually issues so I have started this discussion and not opened a bug.

Vite Ruby Info

bin/vite present?: true
vite_ruby: 3.0.9
vite_rails: 3.0.6
rails: 6.1.4.4
node: v14.16.1
npm: 8.3.0
yarn: 1.22.17
pnpm: 
ruby: ruby 2.7.2p137 (2020-10-01 revision 5445e04352) [x86_64-darwin21]

Details 📜

Problem 1

When watched files have not changed and you try to deploy with capistrano, vite build would not build new assets and deploy existing contents within the public directory. There is a false assumption that the contents of last build would reside locally in public directory. Thats true only if you build locally with a command like RAILS_ENV=staging DEBUG=vite-plugin-ruby:* bin/vite build.
However when deploying with a capistrano deploy script , the assets are directly generated on the server and
In this scenario, the public directory in the release folder on the deployment server will not include the manifest files.

Problem 2

If there are existing files in the public directory ( eg. robots.txt ) , those files would not get deployed on the server in case of a successful vite build. Please see the screenshot below. You can see the manifest files and the assets generated by vite , but the existing public files are not present.

On the other hand when vite doesn't actually build because watch files haven't changed, you can find the existing public files on the server but no manifest files. ( essentially this is same as problem 1 , described above)

Ideal scenario

Ideally, both the existing static files on the src public folder (eg. robots.txt ) as well as manifest files should get generated every time on the server public directory irrespective of whether the watched files have changed or not. It shouldn't be an either/or scenario. Right now the result is one of the above two screenshots.

Problem 3.

When vite does a build ( either the first time or watch files have changed ) , it recreates the assets folder instead of following an existing symlink to the shared assets directory if there is one ( by default capistrano does this ) .
If you compare the two screenshots above, you will notice in the first screenshot vite-ruby replaced has replaced the assets symlink with a new assets folder and generated the assets there.

[ElMassimo]

About builds being skipped, you should ensure that either:

• The content of the publicOutputDir dir is persisted across builds (which is the case in most setups)
• The buildCacheDir is cleared across builds if you don't intend to reuse previous assets

Regarding robots.txt, you can avoid that by moving robots.txt to app/frontend/public, or avoid using "publicOutputDir": "" (which is not recommended, it's better to keep the "publicOutputDir": "vite" default).

About the third problem, if you avoid setting "publicOutputDir": "", it wouldn't affect any symlinks already in place in public.

---

I'm not available for further advice regarding the particulars of a Capistrano-based deployment.

[rachitpant]

• The content of the publicOutputDir dir is persisted across builds (which is the case in most setups)

is there something that should be done so that build is outputted to publicOutputDir and its persisted. The two screenshots above are consecutive builds and the manifest file generated in the first build was not present in the second build.