Vsevolod Stakhov

> What part exactly do you plan not to implement? [Happy eyeballs](https://datatracker.ietf.org/doc/html/rfc6555) There are whole bunch of things that got broken in the async world. Let's imagine that you have...

> Thank you for your work! I think rspamd is a great project and there should not be conflicts and a fork because of such a small issue. > >...

> You could already receive multiple IPv4-addresses in a A lookup and even if the resolver does round robin, one should try all of these addresses, in case one of...

> I do see that there already is an LRU cache for the answers No, there is no such a thing so far - there is a cache merely for...

It's probably related to the problem with too few tokens or something like this. Please enable `bayes` and `lua_bayes` in the debug_modules.

Oh, that's `task` logging. Could you please also add `task` debug module and focus (see column 6 in log output) on log messages in functions: `rspamd_stat_backends_learn` and `rspamd_stat_preprocess` only. It...

Ok, so the culprit is obvious: `2021-11-05 21:34:00 #20650(controller) ; csession; rspamd_stat_classifier_is_skipped: learn condition for classifier bayes returned: already in class spam; probability 100.00%; skip classifier`. I'd say it is...

The problem was how to propagate error properly. I have desided to report that sort of error ' learn error: all learn conditions denied learning spam in default classifier' which...

I don't see an issue here. At least, it is not Rspamd issue. If we cannot learn, there is an error emitted. Why should it be changed I wonder?

So what? If an upstream service doesn't want to learn the error code, the path to resolve it is obvious: call rspamc like `rspamc ... >/dev/null 2&>1 || true`.