CI on unrelated PR fails with an error on episode 15

[ericseppanen]

When queuing up a transcript for episode 008 (#57), I get a CI error:

015-twir-339.md: 7116560b (reported) != 5267717b (actual)

This seems a little weird-- why don't I see this error on PR #55?

@nellshamrell any ideas? It seems to behave as though commit 92154b7 is backwards or not applied, or we're serving the wrong mp3 file?

[jonhoo]

This is so weird. It looks like the file size of that file keeps fluctuating? As far as I can tell, the correct size is 5267717. But that was rejected in #55. I have no idea what's going on with that... Maybe the CI script is wrong?

[jonhoo]

I pushed a fix for that now. Let's see what CI does.

[jonhoo]

Err, that seems to have passed... So strange..

[ericseppanen]

This one seems wrong as well, but isn't throwing errors?

018-twir-343.md:length: "124"

This is roughly what you get if there's a 404 trying to fetch the mp3 file. Note that curl doesn't return an error when this happens. (Maybe curl -f would do better?)

[ericseppanen]

Ah, the episode 18 glitch is caused by there being a space after the URL. Curl is trying to fetch that and getting a 404.

I'll play with the script a little bit to see if I can get it to fail in a less obscure way on 404 errors.

[jonhoo]

Ah, yes, that would certainly do it! Thanks.

[ericseppanen]

Unfortunately, I don't think commit 78b9555 explains why the file size of episode 15 would have been fluctuating. As far as I can tell episode 18 had a completely separate glitch.

[jonhoo]

Yeah, that's true.. I guess I'll re-open in case it shows up again.

[ericseppanen]

I wonder if there's some kind of intrusive caching layer within github, so if your CI scripts run a lot and download big things, you're not wasting a ton of their bandwidth.

Now if that URL has changed over time (which I think did happen, as there was some discussion about audio quality on this episode), then it seems plausible that github actions might report nondeterministic file size fluctuations.

[jonhoo]

Yeah, I also think that URL may have changed once, and suspect that some kind of caching is involved. But hard to figure out without observing it reproducably..