refactor: separate the CDN upload process and add a retry script

[daine]

Summary | Résumé

Refactor: separate the CDN upload process and add a retry to wait for npm publishing to complete
Fix for:

• Compile and publish race condition on components design-gc-conception#907

This adds a retry functionality to the part where the files get uploaded to the CDN. The script usually fails if the package is not yet available on NPM, especially if it's a new package.

New Github workflow file upload-cdn.yml

This separates the CDN upload part in a script called upload_to_cdn.sh, which runs on new the GitHub workflow file upload-cdn.yml.

• The workflow action runs after the compile-and-publish.yml workflow is completed.
• We can run it manually if we need to -- you can run it multiple times because it just updates the files based on the NPM version of the specified package.

Tested with my scratch account at the moment

I ran our terraform files on my scratch account and linked our CDN to my custom domain.

Here are the links for quick testing, to make sure they do upload correctly:

• https://cdn.ds.daine.app/@cdssnc/gcds-components@0.22.1/dist/gcds/gcds.css
• https://cdn.ds.daine.app/@cdssnc/gcds-components@latest/dist/gcds/gcds.css
• https://cdn.ds.daine.app/@cdssnc/gcds-components@0.22.1/dist/gcds/gcds.esm.js
• https://cdn.ds.daine.app/@cdssnc/gcds-components@latest/dist/gcds/gcds.esm.js
• https://cdn.ds.daine.app/@cdssnc/gcds-components@0.22.1/dist/gcds/gcds.js
• https://cdn.ds.daine.app/@cdssnc/gcds-components@latest/dist/gcds/gcds.js

Testing @latest and @0.22.1 are exactly the same:

$ wget https://cdn.ds.daine.app/@cdssnc/gcds-components@0.22.1/dist/gcds/gcds.css
$ wget https://cdn.ds.daine.app/@cdssnc/gcds-components@latest/dist/gcds/gcds.css
$ ls -alh
    38K 18 Jun 09:26 gcds.css
    38K 18 Jun 09:26 gcds.css.1
$ diff gcds.css gcds.css.1
// nothing; exactly the same

---

S3 Bucket

gc-design-system-daine-scratch-cdn
@cds-snc/gcds-components@latest/

Confirmed: Contents are the same, file size is the same.

[patheard]

I like this pattern - it's a nice way to work around the problem of the package not being instantly available.

[daine]

I've made a few changes and ran them on my forked repo.

@ethanWallace I added sleep timer (60 seconds) to allow time for npm to publish the web package before proceeding to publish the rest of the packages. Here's the log to see the publish package workflow run w/sleep timer

^ Above action also forces a failure so I can test the unable-to-deploy job.
@patheard Here's the log for running a failed publishing workflow run

[ethanWallace]

LGTM!