make fn scoped public in retry.rs

[enricoschaaf]

Hi 👋🏽,

nice to see reqwest adding it's own retry helpers. Would it be possible to make this function

reqwest/src/retry.rs

Line 89 in 994b8a0

fn scoped<F>(func: F) -> Builder

public so it's possible to setup retrying for all hosts?

[seanmonstar]

Yes, the goal was to eventually make the API more public to allow more things. But in order to do so, there's some conceptual design challenges that need to handled, at least to help guide the user to making wise choices when they don't realize there is danger.

The reason for the scopes in the first place is that budgets likely shouldn't be applied to all hosts. But budgets are extremely important. It's explained more in this blog post I wrote while designing the feature: https://seanmonstar.com/blog/reqwest-retries/

So, to open that up, I'd want a design that might allow more hosts, while still encouraging budgets. Want to think about that design space?

[enricoschaaf]

My use case is quite simply. I would like to internally handle the retry of 429 errors. Currently we just externally retry requests without looking much at the status code so it retries 429 errors. Instead of improving the external retry I thought I look at reqwests new build in retries. Budgets totally make sense but we have a client builder that wraps the reqwest one and I would ideally like to be able to define a default retry policy that uses per host budgets. With the currently exposed for_host fn this is a bit award because I would have to ask people for all the hosts a client calls. That usually is not many but often env specific. (Actually we have one case where we send webhook events to customer where the client has a high number of hosts). So long story short not ideal. A design that I would see is keeping the for_host fn to allow overriding for a specific host but allow something like are retry_policy() function that applies to all hosts but basically keeps a Map of all hosts and their budget. I am happy if it's not allowed in the design to use a global budget but being able to use a default retry policy and override this for certain hosts with for_host would be needed in my eyes. Maybe an unstable_retries feature flag would be nice to explore this? Not sure how it's handled here