plugin reloading is broken on non cluster pages (like settings)

[illume]

Describe the bug

For me reloading with "npm start" is broken except on the home page.

Note, only seems to happen only on non-cluster pages that don't do backend requests.

To Reproduce

Steps to reproduce the bug:

• Open headlamp app (in dev mode or with compiled app)
• Go to a non home page (eg settings)
• cd plugins/examples/podcounter/ && npm i && npm start
• Note how the app doesn't reload
• Make a change to plugin, note how it doesn't reload

Environment (please provide info about your environment):

WSL ubuntu, running the app

[illume]

This also affects backend lists of clusters I think. Similar problem.

[rajdama]

Headlamp caches plugins in the ~/.config/Headlamp/plugins directory, which means that even if you modify the source code, it continues to use the previously built version instead of automatically updating. Additionally, Vite’s hot reloading doesn’t work in this setup because Headlamp isn't monitoring changes in the dist/ folder where the plugin is built. As a result, any updates to the plugin require manually deleting the cached version and restarting Headlamp to see the changes.

I can work on this issue..

[Faakhir30]

Not exactly same issue but facing similer issue with helm deployment.

@illume, when should this cache be invalidated for in-cluster deployments? I was assuming backend is maintaining a watch on -plugins-dir arg path. So if plugins at my -plugins-dir=/headlamp/plugins are updated, the cache should have been updated.
Needed this in the main headlamp container for #2787.

f@f:~$ kubectl exec -it headlamp-55bbd688c8-n9tht -c headlamp -- /bin/sh
/ $ ls ~/.config/Headlamp/
kubeconfigs  plugins
/ $ ls ~/.config/Headlamp/plugins/
/ $ ls /headlamp/plugins/
appcatalog_headlamp_plugin  prometheus_headlamp_plugin
/ $ 

Also, visiting cluster info pages and other settings pages from UI is also not updating the cache

[kallal79]

Hi @illume, I'd love to work on this issue! Please assign it to me, and I’ll start investigating potential fixes. Thanks!

[illume]

@Faakhir30 I think I covered your question on slack. Sorry I didn't see this mention.

For passers by: It's to do with the lack of watching when in-cluster (now there is an option in main).