Display original error on failure

[djc]

No description provided.

[cpu]

@djc This looks like an uncontroversial fix that might be nice to clear out of the PR queue.

There was a Macos CI failure but the job logs have already been purged and I don't have permissions to re-run. Could you kick the build and see if it was something transient?

[djc]

It's interesting that the apple test fails on beta/ubuntu-20.04 1.59.0/ubuntu-20.04 but passes on stable/ubuntu-20.04, nightly/ubuntu-20.04, stable/macos and stable/windows. Seems like an intermittent issue, maybe they're doing some kind of rate limiting?

I tried running it a bunch of times in rapid succession locally, but it passed every time.

[cpu]

Seems like an intermittent issue, maybe they're doing some kind of rate limiting?

I was curious (and already looking at some CI failures today) so I tried hammering apple.com a bit from a test GHA repo. I was able to reproduce the issue with openssl s_client in a tight loop, but not reliably. In approx. 1 in 10 runs, I'd get a failure after ~25 connections. I saw similar tweaking the rustls-native-certs GHA to run just the apple.com smoke test in a loop of 50 runs. Most completely successfully, one failed on the 26th connection. I concur that it seems like rate limiting or a WAF 🤷 Frustrating!

[djc]

@ctz do you remember if there's a specific reason to include apple.com here? Do we suspect that they run a SecureTransport-based server which we'd like to try, or something?

[djc]

(Meanwhile, we probably shouldn't block this PR on the intermittent CI issues.)

[ctz]

@ctz do you remember if there's a specific reason to include apple.com here? Do we suspect that they run a SecureTransport-based server which we'd like to try, or something?

It's really just a random selection of big company websites that came to mind at the time.

[djc]

Okay, maybe we should just get rid of the apple test? Seems like it's more trouble than it's worth at this point.