Node.js servers being down fails the CI workflows

[sheerlox]

When using Nodelix in CI (e.g. through semantic_release), sometimes the Node.js servers are down and the whole CI fails.

We should consider implementing an exponential backoff retry to mitigate that issue and eliminate the need to manually relaunch the workflow (sometimes multiple times).

[Lucassifoni]

WIP here : alzo-archi@9ab9452

The idea would be to make the backoff optional (the default max_attempts \ 1 param makes it optional) to allow people to control this behavior from the outside (personnally I would launch the install in an Oban job as part of a workflow DAG).

Would that be direction that would suit you ?

I'll have to separate the actual downloading from this, to unit test the backoff separately in my next commits.

[sheerlox]

Looks great, like the direction this is taking!

Regarding the default for max_attempts, I'd personally put a sane number as default (3 or 5), and leave specific use-cases (like running in a DAG) to override it to 1.
I think setting the default to 1 would make sense if the async installation was the main use case, which I don't think it is for now.

What do you think?

[Lucassifoni]

I think you're right, in that case I'd write a tighter loop that sleeps for max 50ms at a time and checks if a time boundary has been exceeded to decide whether to sleep again or trigger a retry, because I think I remember that sleeping for more than a few tens of milliseconds is suboptimal for the BEAM's preemptive scheduling.
I'll update both code and this ticket after looking that up.

[Lucassifoni]

Looks I'm wrong on that one, after searching, sleeping is generally discouraged because it's often the wrong tool vs message passing but it's fine here.
I'll switch max attempts to 3 and add tests.

[Lucassifoni]

Now working on that in #23 .

[sheerlox]

🎉 This issue has been resolved in version 1.0.0-alpha.15 🎉

The release is available on:

• GitHub release
• v1.0.0-alpha.15

Your semantic-release bot 📦🚀