What's the correct way to config Alertmanager to use a forward proxy?

[eugene-chow]

In this thread with @brancz, I was seeking a way to config Alertmanager to use a forward proxy. One way is to use the https_proxy env var detailed here.

Is this the correct way to do it in the long haul, or should it be in Alertmanager's config. Thanks!

[brancz]

As mentioned in the other thread, all Prometheus projects are exclusively configured via config files, and that this is configurable via an environment variable right now is purely a coincidence because of the underlying http client implementation.

I'd propose that an http_proxy config maybe to be consistent with Prometheus we should also call it proxy_url (see https://prometheus.io/docs/operating/configuration/#scrape_config). That configuration should then be available/implemented for all existing notification providers.

[brian-brazil]

For blackbox I'm getting the common http bits broken out to a library in common, which we could also use here.

[brancz]

@brian-brazil is this already in process?

[brian-brazil]

prometheus/common#81 I must look at it again

[brian-brazil]

That's submitted over in common, so should be good to go here.

[brancz]

With the changes merged from prometheus/common#81, this can now be implemented here.

@brian-brazil what are your thoughts on where/how to integrate the HTTP config into the alertmanager config? An optional global and then an optional on every receiver config interacting with HTTP?

[brian-brazil]

That's fine in principle, the problem is how we'd deal with the case where there's both a global and a per-receiver config. How should they be merged and/or overridden? Probably overridden.

[brancz]

Yes I think if specified in a receiver then it overrites the global, no merging would be my suggestion as well.

[brian-brazil]

That just leaves how to implement that then. Maybe do this as a named "http_config" rather than the usual inlining?

[brancz]

Yes that was my idea as well.

[brian-brazil]

This was done.