Problem

Turns out #9 + #10 did not solve the problem for a scenario where user omits the version and action needs to fetch addtional metadata to find out what is the latest one:

https://github.com/ipfs/distributions/runs/5136166478?check_suite_focus=true#step:3:92

This causes our nightlies to fail ipfs/distributions#631

Solution

I believe it is better to default to have some fallback or cached version than to fail CI when dist.ipfs.io is unstable.

Places which we need to harden:

• fetching /versions
• fetching dist.json for latest version

Thoughts @aschmahmann @galargh ?

This action uses curl multiple times. If any of the HTTP requests fails, the entire CI job will fail.
Unfortunately, out HTTP gateways are sometimes overloaded, and timeouts are more common that one would expect.

I believe we could improve devexp by retrying a few times before returning an error that would fail the CI.

Perhaps curl --retry 5 ? Afaik it is pretty good (it has exponential backoff enabled by default). We could also use --retry-all-errors if it is supported by the version of curl present on github.

Caveat: I don't think retrying plays well with redirected output – we may need to save output to a file instead, using explicit -o filename so it can be resumed correctly (see this warning)