Eric Huber

Just to keep everything in one place, I have a separate example here that reproduces the issue more consistently by injecting a delay. (This assumes the fix is defeated/omitted first,...

Exact steps to reproduce (thanks @tdy for suggestions): ```bash #!/bin/bash mkdir -p "$HOME/pl_ag_jobs" mkdir -p "$HOME/pl_repro" cd "$HOME/pl_repro" if [ ! -e workspace_reboot ] ; then git clone https://github.com/echuber2/PrairieLearn.git workspace_reboot...

I'll try to reproduce on some other machines, but based on the fact that I do experience the issue and the proposed fixes have been working for me, we can...

Here I have a repro commit that works in WSL on a different machine. (Initially I couldn't reproduce the issue, but inserting an artifical delay after the database update commands,...

^ Detailed repro steps here: https://github.com/PrairieLearn/PrairieLearn/issues/10716#issuecomment-2389440033 I was going to replace the hard-coded 10 retries for 1 sec intervals with something that would retry 7 times for 1/10 intervals of...

Looks like the earlier version that just installed "3.10" also grabbed the cpython version by default, judging from one of the earlier jobs that had timed out: ``` 136.2 +...

Since Amazon Linux 2023 doesn't have 3.10 as a package (apparently), it might also be worth trying uv on top of the conda/mamba build of the runtime.

I found that moving `export UV_PYTHON_PREFERENCE=only-managed` after the venv is created helps to get around the issue, or it could simply be removed. (I was about to push another experiment...

Welp, this is still timing out. I wonder if the Python version is really the issue. Maybe it will help to update based on master, or try layering `uv pip...

It may aid debugging to split out all of the uv stuff to a separate `RUN` script after the `dnf` and `npm` parts, at least temporarily. This makes rebuilds much...