Each container job now looks up the image's index, manifest,
and config digests from the registry,
keys an `actions/cache` entry on the config digest,
and only uses a cached image if its ID matches one of those digests;
otherwise it is discarded and the image is pulled (with retries) instead.
A poisoned or stale cache entry can therefore never be used,
and a cache hit needs only a couple of small manifest requests,
so this scales regardless of pull rate limits.
DockerHub's anonymous pull rate limit applies per IP for hours,
so retrying cannot ride it out on shared runners;
`mirror.gcr.io` serves the same official images without that limit,
as the io.js source-install job already does.
Mirror outages have outlasted the existing retries,
which covered only `apt-get update` (and not the WSL `additional-packages` install).
Installs now go through `.github/scripts/apt-install.sh`,
which retries each request within apt and retries the whole update+install with backoff for about 10 minutes.
Upstream's `.gitmodules` uses a relative submodule url (`../nvmrc.git`),
which resolves against the superproject's origin:
on a fork without its own `nvmrc` fork that is `<owner>/nvmrc`,
which 404s and fails `actions/checkout` with `submodules: true`.
Keep `submodules: true` (so a fork's own `nvmrc` is used when present),
but mark checkout `continue-on-error` and, only when it failed,
re-point the submodule at `nvm-sh/nvmrc` and update.
harden-runner runs with `egress-policy: block`, and the allow-list only included
`production.cloudflare.docker.com`. DockerHub serves image blobs from either its
Cloudflare or its CloudFront CDN; when a pull was routed to CloudFront
(`production.cloudfront.docker.com`) the connection was dropped, causing
`error pulling image configuration: ... connect: connection refused` and exit
125 in the xenial, installation_node, and fast (httpbin) suites. Allow both CDNs.
Container-based suites and the `nvm_download` httpbin check hard-fail whenever DockerHub is briefly unreachable
(observed: `dial tcp ...:443: connect: connection refused` while pulling images),
even though the change under test is fine.
This is unrelated to any test logic.
- tests-xenial / tests-installation-node: retry the `docker pull` up to 5 times before `docker run`, mirroring the existing apt-get retry
- `nvm_download` test: retry the httpbin pull and skip the auth-header checks (rather than fail) when the image cannot be pulled or run, and make cleanup tolerant of a missing container.