test: Embed the test data inside the test container image #545
Conversation
There was a problem hiding this comment.
Interesting idea; thank you!
What do you think, would it be worth to remove & prune the image after the test is done? I'm not sure how docker's layer will work in this case: won't we be eating out more and more disk space with every run of the tests?
Also, I have one tiny request about the code, see below.
| let docker_image: RunnableImage<GenericImage> = | ||
| GenericImage::new(DOCKER_PLEX_IMAGE_NAME, &image_tag) | ||
| GenericImage::new("plex-api-tests", "latest") | ||
| .with_wait_for(WaitFor::Healthcheck) | ||
| .into(); | ||
|
|
||
| #[cfg_attr(windows, allow(unused_mut))] |
There was a problem hiding this comment.
Can you remove this along the way? Since the variable is not mut anymore.
|
Unfortunately, I don't think it'll help us to resolve the transcoding crashes in any way :( Volumes on macOS and Linux work in very different ways. |
|
Just to clarify my previous comment(s), as I think I didn't communicate my thoughts clearly. I'll do my best to avoid providing quick-ish responses during business hours in the future. The change is good and valuable: it solves the issue arising during the development using macOS. But it doesn't look like it might fix the problem with crashes during transcoding requests :( |
You're probably right. So I guess the question is do you think it is actually worth doing this? It will slow down running tests, though should be by a pretty tiny amount and I guess adds an additional point of failure. |
|
Let's keep this open for now and use the old flow locally. If the issue with the missed media will show itself in the next, let's say, month, it'll be a good sign that we need this change. |
andrey-yantsen
added
the
S-Postponed
This is something of a test to see if it might resolve #541. I discovered that running tests locally got a lot better when switching Docker Desktop's file sharing implementation from VirtioFS to gRPC FUSE which got me wondering whether bind mounting the data into the container is an issue. That would seem strange since that's how I run my own server but still I thought it might be worth a try. This change creates a new container image where instead of mounting the data we copy it into the image before launching so there are no longer any bind mounts involved.