Centralise frontend network error reporting

[sarayourfriend]

Fixes

Fixes #3468 by @krysal

Description

I've kept this PR as a draft because:

• it is just a proposal for an idea for how to solve this problem, but is not necessarily the simplest (but I think it's worth it)
• it still needs additional testing and probably some small refactoring, but I don't want to invest time into that until someone else has looked at this and given the general approach/idea a 👍

Testing Instructions

Force an error to happen in a network request on the frontend. For example, append a bogus string to the end of media IDs so that they result in 404s. Confirm that those don't get sent to Plausible or Sentry anymore (try it on main as well to see that it goes to Sentry).

I don't know how to force a network error. You may be able to do it using your browser's network debugging tab and explicitly blocking a request.

Checklist

• My pull request has a descriptive title (not a vague title likeUpdate index.md).
• My pull request targets the default branch of the repository (main) or a parent feature branch.
• My commit messages follow best practices.
• My code follows the established code style of the repository.
• I added or updated tests for the changes I made (if applicable).
• I added or updated documentation (if applicable).
• I tried running the project locally and verified that there are no visible errors.
• [N/A] I ran the DAG documentation generator (if applicable).

Developer Certificate of Origin

Developer Certificate of Origin

Developer Certificate of Origin
Version 1.1

Copyright (C) 2004, 2006 The Linux Foundation and its contributors.
1 Letterman Drive
Suite D4700
San Francisco, CA, 94129

Everyone is permitted to copy and distribute verbatim copies of this
license document, but changing it is not allowed.


Developer's Certificate of Origin 1.1

By making a contribution to this project, I certify that:

(a) The contribution was created in whole or in part by me and I
    have the right to submit it under the open source license
    indicated in the file; or

(b) The contribution is based upon previous work that, to the best
    of my knowledge, is covered under an appropriate open source
    license and I have the right under that license to submit that
    work with modifications, whether created in whole or in part
    by me, under the same open source license (unless I am
    permitted to submit under a different license), as indicated
    in the file; or

(c) The contribution was provided directly to me by some other
    person who certified (a), (b) or (c) and I have not modified
    it.

(d) I understand and agree that this project and the contribution
    are public and that a record of the contribution (including all
    personal information I submit with it, including my sign-off) is
    maintained indefinitely and may be redistributed consistent with
    this project or the open source license(s) involved.

[sarayourfriend]

@krysal interesting in your view on this, in particular, if you have any opinions about this approach.

The simpler alternative I can think of was to either copy/paste the conditionals of error handling to the appropriate areas, or to create a similar unified fetch error processing function that just takes context as an argument, rather than creating a plugin. The plugin feels like a convenient approach to me, but it could cause additional pains with Nuxt 3, I'm not sure.

Curious for your thoughts regardless!

[github-actions]

Full-stack documentation: https://docs.openverse.org/_preview/3721

Please note that GitHub pages takes a little time to deploy newly pushed code, if the links above don't work or you see old versions, wait 5 minutes and try again.

You can check the GitHub pages deployment action list to see the current status of the deployments.

[AetherUnbound]

I think this approach makes sense! I like that we can differentiate the tool the error is sent to for certain kinds of failures.

[obulat]

I love centralized recording of errors, and separation of the actionable ones from the non-actionable, @sarayourfriend!
I was thinking whether we should create an Error class for our custom fetchingError instead of an object, in case we need to throw this error somewhere (based on some work in the Nuxt 3 branch). What do you think?

[obulat]

Have you checked that this always works?

The reason that sendCustomEvent was passed as a parameter here was that you cannot use a composable that requires Nuxt instance outside of setup. There is an edge case useSearch is called, if I remember correctly, in asyncData, and Nuxt instance is not available, so a call to useAnalytics inside useSearch produces a warning.

[sarayourfriend]

I see. But then should useSearch return a product based on calling useAnalytics? I'll revert this change, but that to me sounds like a bit of code smell, if the composable isn't able to call other composables at the top level. Maybe I am misunderstanding the edge case. Either way, I'll remove this from the PR, it definitely doesn't need to be here.

[krysal]

I love the centralization of error management, too! 💯 I think the plugin approach is right; not so sure about the high coupling with Axios since it may go away with Nuxt 3, but it is what it is right now. If we also move the API calls to a plugin, I think the error management will probably merge with it.

[sarayourfriend]

should create an Error class for our custom fetchingError instead of an object, in case we need to throw this error somewhere

In JS you can throw anything. We can wrap it in a class, no problem. But is the idea to be able to check it with instanceof? We can do this in a separate issue as well, now it will be easy to switch to a different format for the error if needed.

[sarayourfriend]

Thanks for the preliminary reviews, Krystle and Olga. I'll fix up this PR and get it ready for a final review.

[obulat]

I tried this locally, and sent the following error to Sentry:

Do you know where Sentry gets the "recordError" title from, and whether we can change it?
I also left 2 non-blocking comments inline.

[obulat]

Maybe we should treat the 429 errors the same as NETWORK_ERRORs? There's nothing we can do about them in Sentry, too.

[sarayourfriend]

Good question! Those we can see in Cloudflare, so I don't know how much value we get from sending them to Plausible, but I agree that we can skip them in Sentry 👍

[obulat]

Then it's even better, we can completely ignore 429s here :)

[obulat]

What if we test the$processFetchingError instead of the plugin itself? This way, the setup and mocking would be simpler.

[sarayourfriend]

I did start with that, but the function the plugin puts into $processFetchingError still requires $sentry and useAnalytics. It also would require further exporting things from the plugin module that aren't really needed anywhere other to make them testable, and would be testing such small units that it would kind of miss the bigger picture (is how I felt).

Happy to change this if you see a way to make the tests much simpler, and not require mocking $sentry and useAnalytics still, but if they still require that, then the only change is one that also additionally tests that the plugin is configured correctly (calls inject with the right parameters).

[obulat]

Not directly related to this comment, but after spending some time this morning investigating the problems with "search time" event, I realized that this change would also require the changes that would make useAnalytics usable on the server (see #3775)

[sarayourfriend]

Do you know where Sentry gets the "recordError" title from, and whether we can change it?

I suppose from the function name, but I don't know how to change it. Maybe in breadcrumbs? I'll have to look at the sentry JS docs, but it's a good one to change, absolutely.

[obulat]

I'm requesting changes because I realized that our plausible tracker breaks on the server (it's basically only loaded on the client side). We would need to make sure that the server-side error reporting does not cause new errors. And related to that, we should probably add e2e tests for errors because the unit tests do not have server/client distinction.

[sarayourfriend]

Yes! I actually have staged changes for that I forgot to push, thanks for the reminder!

[sarayourfriend]

I remember what I got stuck on, with Playwright. I don't know how to trigger a network error as Axios identifies them. Maybe using page.route and route.abort will work, I will try it. Just wanted to mention in case you have ideas before I'm able to get back to this, @obulat

[obulat]

WebStorm is complaining that JSDoc does not describe the parameters correctly (error instead of context, originalError and fetchingError.

[obulat]

I remember what I got stuck on, with Playwright. I don't know how to trigger a network error as Axios identifies them. Maybe using page.route and route.abort will work, I will try it. Just wanted to mention in case you have ideas before I'm able to get back to this, @obulat

You can use Playwright's route.abort like this to trigger a network error:

await page.route(new RegExp(`v1/(images|audio)/`), async (route) => {
      route.abort("internetdisconnected")
    })

I tried adding e2e error tests to this PR, and found that useAnalytics does not work when we need to call it from the store and not from the setup function. The context is not passed to this composable, even if the store's fetchMedia function is called from within the setup function. The context is not available, and the onMounted also throws an error. We could move the onMounted to sync analytics to the layout files to prevent it from running for every mounted component that uses it. However, the solution for the context not being available is more complex. Maybe we should move sendEvent to a plugin?

[sarayourfriend]

Maybe we should move sendEvent to a plugin?

I was thinking about this as well. There are a few composables like this that might be better suited for plugins instead of composables.

[sarayourfriend]

This PR doesn't work due to the analytics composable, so I'm going to close it for now, but leave the branch up. I'll open another PR to move analytics composable into a plugin.