Spurious network errors for `index.crates.io` about simpleinfra HOT 4 CLOSED

See also https://github.com/ferrous-systems/rust-exercises/actions/runs/5967326186/job/16188788089

from simpleinfra.

One theory I brought up in https://rust-lang.zulipchat.com/#narrow/stream/318791-t-crates-io/topic/repeated.20errors.20in.20fetching.20.60syn.60 is that Cargo doesn't seem to close the TCP connection when retrying, which means that we're almost certainly hitting the same CloudFront server on all the retries. I think it would be worth seeing if we can change the implementation to re-open the connection (ideally also re-poking DNS) to try to get a different answer.

from simpleinfra.

The discussion in https://rust-lang.zulipchat.com/#narrow/stream/246057-t-cargo/topic/timeout.20investigation suggests that the problem might be an issue with curl < 8.1 and Rust < 1.72.

from simpleinfra.

We haven't received any new reports about this since September and there is been no further discussion on Zulip. I'm going to close the issue under the assumption that this was an issue with a specific version of curl. A workaround is available in the linked Zulip thread above.

from simpleinfra.