https://bugs.exim.org/show_bug.cgi?id=1802
--- Comment #12 from Heiko Schlittermann <hs@???> ---
(In reply to vsevolod from comment #11)
> For the god sake, could you please explain me, why is it reasonable to use
> shutdown for Rspamd? I'm completely confused with your motivation.
>
> Unlike SA, Rspamd NEVER EVER required for a shutdown call: it required
> explicit 'Content-Length' thought. Hence, if you remove this call from
> Rspamd path, then Exim will be able to communicate with ANY version of
> Rspamd. So, again, what are you reasons to leave shutdown for Rspamd checks?
Maybe I wasn't able to express myself. I'll try again :)
I understand that Exim does a half close of the connection to RSpamd, after
all data has been sent. Currently I do not see what's wrong with this, except
that RSpamd doesn't like the half closed connection. (For my understanding
it's perfectly fine to shutdown() one direction of the communication, either
to signal that there is nothing more the remote end has to expect, or to
release ressources, flush buffers etc.)
Newer Rspamd seem to interpret this half-close as a "client drop" and stop
processing the input data. (I still do not see the benefit of doing so, but
you're the developer and the protocol design decisions are on your side.)
My suggestion was, to use the supplied patch, as long as we can guarantee that
newer Exims are able to talk to older Rspamds. If older Exims can't talk to
recent RSpamds, we can't do anything. But for this reason you implemented an
option for RSpamd, to be able to serve older Exims (and other older clients).
If older Rspamds do not rely on the half-close (shutdown()), we're fine keeping
the connection two-way open until we received the response from Rspamd.
If you say that even older Rspamd rely on Content-Length header already, we can
change Exim accordingly.
--
You are receiving this mail because:
You are on the CC list for the bug.
