Fix first-time run of docker_dev.sh

[AJGranowski]

What are these changes?

This change makes octane commands wait a bit for manifest.json to exist if it's missing.

Why are these changes being made?

The first run of bin/docker_dev.sh causes both nginx and php-dusk fail.

php:

Exception
The manifest does not exist.
at app/Singletons/AssetsManifest.php:22
   18▕     {
   19▕         $manifestPath = public_path('assets/manifest.json');
   20▕
   21▕         if (!file_exists($manifestPath)) {
➜ 22▕             throw new Exception('The manifest does not exist.');
   23▕         }
   24▕
   25▕         $this->manifest = json_decode(file_get_contents($manifestPath), true);
   26▕     }
1   app/Providers/AppServiceProvider.php:96
    App\Singletons\AssetsManifest::__construct()

php-dusk:

php-dusk exited with code 157

nginx:

[emerg] host not found in upstream "php-dusk" in /etc/nginx/conf.d/default.conf:53
nginx exited with code 1

Successive runs pass and serve the website, suggesting a race condition in docker compose up. Running docker compose run --rm assets beforehand seems to fix this.

[nanaya]

what should be done is adding health check on assets service, checking for existence of the manifest file.

[AJGranowski]

I'll take a look into that.

[AJGranowski]

Hmm, looks like things break if I just use health checks... I think assets depends on composer install?

PHP Warning:  require(/app/vendor/autoload.php): Failed to open stream: No such file or directory in /app/artisan on line 18
PHP Fatal error:  Uncaught Error: Failed opening required '/app/vendor/autoload.php' (include_path='.:/usr/share/php') in /app/artisan:18

Extracting localizations...
[webpack-cli] Error: ENOENT: no such file or directory, open '/app/resources/builds/messages.json'
    at Object.openSync (node:fs:573:18)
    at Object.readFileSync (node:fs:452:35)
    at getAllMesssages (/app/resources/js/cli/generate-localizations.js:36:22)
    at extractLanguages (/app/resources/js/cli/generate-localizations.js:20:20)
    at Object.generateLocalizations [as callback] (/app/resources/js/cli/generate-localizations.js:66:21)
    at /app/webpack.config.js:292:40
    at Array.forEach (<anonymous>)
    at configFunction (/app/webpack.config.js:292:11)
    at loadConfigByPath (/app/node_modules/webpack-cli/lib/webpack-cli.js:1445:37)
    at async WebpackCLI.loadConfig (/app/node_modules/webpack-cli/lib/webpack-cli.js:1515:38) {
  errno: -2,
  code: 'ENOENT',
  syscall: 'open',
  path: '/app/resources/builds/messages.json'
}
error Command failed with exit code 2.

[nanaya]

maybe add a compose install before the yarn install in the watch command

[AJGranowski]

Yep, that seems to work. Let me know if you were thinking of a different approach.

[nanaya]

looking again at how the service is actually used, maybe just monitor for the file (with a while loop) inside the octane specific entrypoint function instead

(also, moving the composer install before setting up github token for it is bad)

[AJGranowski]

Were you thinking of something like this latest change?

[AJGranowski]

Migrated the manifest check to run.sh. On my machine, the script waits between 15 and 70 seconds for the manifest file to generate.

Here's what the stderr output looks like:

[nanaya]

also please squash the commits

[notbakaneko]

What's wrong with putting an empty manifest.json and letting the workers reload once the new one finishes building, instead of jumping through all the hoops checking for the file?

[nanaya]

that's a good point 🤔

[AJGranowski]

That seems to work! I just get served some error pages until it's finished.

[AJGranowski]

Here's an alternative PR:

• Add empty manifest.json #11792

Either work, it just depends on the UX you're looking for when starting (either getting served nothing, or an error page during startup).

[nanaya]

let's just go with this one for now (there's different problem with the other one in its current state and I would probably take a different approach)

[AJGranowski]

Sounds good! Thank you for being patient with me.