Timeouts for post_to_esp()

[nuschk]

We recently had networking troubles which lead to certain (Sendgrid) SSL handshakes stalling forever. This lead all our mail sending workers to eventually stall and lock up all background tasks.

Usually, for other HTTPS requests, we add a (generous) timeout (see http://docs.python-requests.org/en/master/user/quickstart/#timeouts). The timeout is chosen such that it only triggers in those rare, extreme circumstances. We mostly use 30 seconds, but Stripe, as another example, uses even 80 seconds (https://github.com/stripe/stripe-python/blob/f948b8b95b6df5b57c7444a05d6c83c8c5e6a0ac/stripe/http_client.py#L97).

I think anymail should have a default timeout, too. Doesn't even have to be configurable, and the exception could be passed straight to user code.

[medmunds]

Oh, good call. It should be straightforward to set a timeout for all the requests-based backends (currently all but SparkPost), and easy enough to make this an Anymail setting.

I'm leaning toward a 30 second default timeout. Frankly, this feels really long (though I don't have a good sense for every ESP's historical performance). Thoughts?

[nuschk]

I completely agree on a long timeout, which should never be reached by accident. 30s seems completely fine to me and would certainly work for us.

I could imagine people sending recipient lists with several tens of thousands of entries to the ESP. If your line is bad or congested, this may take some time. OTOH, this seems like a very special use case, and if users run into that, it should be very straightforward to diagnose and fix.

So yeah, I'm fine with it. :-)

And thanks for that awesome package and the quick migration away from mandrill! That really made my day back then.

[medmunds]

Released in v1.3