@santi, do you remember how long it took for the 3.11 dependencies to resolve? It's taking a rather long time (>1h) on my machine and am not sure whether I'm missing something.

@tillahoffmann As far as I can remember it took a while to generate the requirements.txt, but nowhere near an hour. I'll take a look tomorrow and check it out myself.

Great, thank you!

I've added a build to compile the requirements in #331. Here's an example of a slow build for 3.11 https://github.com/testcontainers/testcontainers-python/actions/runs/4681865266/jobs/8294990576. Any thoughts much appreciated!

@tillahoffmann Looking into it. A fresh install on my M1 Macbook Air takes about ~20 seconds, but I'll try spinning up a new Python container and see how it behaves there. Probably a lot of non-standard config on my computer.

From the logs in the run you provided it seems to struggle to find compatible dependencies, which leads to backtracking and trying out most of the distributions available. This does not happen on my computer, so there might be some already built or resolved cache locally. A fresh Ubuntu image will hopefully trigger the problem, I'll try that out as a first theory.

@tillahoffmann From a fresh install with ubuntu:20.04 running in a container, everything seems to work smoothly on the latest commit on master (a1aceac). Install takes about 30 seconds (running in Docker on a M1 Macbook Air), including downloads.

Only problem I had was installing pymyssql, which required a missing lib on standard Ubuntu distros.

A quick

apt update
apt install freetds-dev

fixed the problem and made everything install as expected, but this was probably not related to the backtracking error we saw in the Actions log.

Here is a full pastebin of the output from the run: https://pastebin.com/uADcPeSA

More recent runs also seem to complete in about 30 seconds, so unless the problem arises unexpectedly again, I guess this issue is resolved?

Thanks for the investigation, @santi! I think #333 resolved the issue. We had google-cloud-pubsub pinned to an ancient version which made the dependency resolution take a long time on 3.11.

@tillahoffmann Happy to help!

A bit unrelated, but I just wanted to highlight this to you, as I see you are planning a major version release pretty soon: #314

I've implemented a change which makes the behaviour of the library more similar to the other implementations of testcontainers, by using Ryuk for cleaning up containers instead of having to rely on a fallback with a garbage collector with the del method (and also opens up more possibilites for using a container without having to keep a reference to the container at all times).

There has been a discussion going on in that PR on whether to enable the Ryuk cleanup by default, which would make the lib work in the same way as other implementations. HOWEVER that change in behaviour should be considered a breaking change, and requires a major version bump if we are to follow semantic versioning. I am open to making the last few changes required for enabling Ryuk by default on short notice in the PR. Is that something you could consider including in the v4 major release?

Yes, let's hold off with the major version release until #314 is addressed. Sorry I haven't been able to get to it yet. I'll be out of the office for the coming week but will have a look when I'm back.