Failed to get sources of alt-ergo.1.01 (and cmdliner-2.0.0)

[jmid]

On #28841 (and elsewhere) we are seeing the same pattern of lots of tests (failed: Failed to get sources of alt-ergo.1.01: curl failed).

When I initially investigated it, I discovered that alt-ergo.1.01 and others were still using http which was then permanently forwarded to the corresponding https url. #28896 fixed that, but didn't solve the Failed to get sources issue.

On 4.14 for example, it happens during lower bound search - hence the pre-historic alt-ergo.1.01.
https://opam.ci.ocaml.org/github/ocaml/opam-repository/commit/4792a6c73213655c2a0d420c9c1bf26d0fe51a10/variant/compilers,4.14,frama-c-metacsl.0.10~beta,tests

Looking at the log I see:

#=== ERROR while fetching sources for cmdliner.2.0.0 ==========================#
OpamSolution.Fetch_fail("https://erratique.ch/software/cmdliner/releases/cmdliner-2.0.0.tbz (curl failed: \"/usr/bin/curl --write-out %{http_code}\\\\n --retry 3 --retry-delay 2 --user-agent opam/2.5.0~beta1 -L -o /tmp/opam-7-574620/cmdliner-2.0.0.tbz.part -- https://erratique.ch/software/cmdliner/releases/cmdliner-2.0.0.tbz\" exited with code 6)")

#=== ERROR while fetching sources for alt-ergo.1.01 ===========================#
OpamSolution.Fetch_fail("https://alt-ergo.ocamlpro.com/http/alt-ergo-1.01/alt-ergo-1.01.tar.gz (curl failed: \"/usr/bin/curl --write-out %{http_code}\\\\n --retry 3 --retry-delay 2 --user-agent opam/2.5.0~beta1 -L -o /tmp/opam-7-650c0e/alt-ergo-1.01.tar.gz.part -- https://alt-ergo.ocamlpro.com/http/alt-ergo-1.01/alt-ergo-1.01.tar.gz\" exited with code 6)")

According to man curl on my machine:

EXIT CODES

       6      Could not resolve host. The given remote host could not be resolved.

Interestingly,

• it is these same two packages that fail to have sources fetched for all failing jobs,
• the two are notable for self-hosting, rather than pointing to GitHub like most other sources, and
• it happens on opam-ci workers, not on GHA

Wildly speculating I'm wondering whether (lack of?) CDN or too long- or short-lived caching may be involved... 🤔

Polite ping @mtelvers