Re: [Libevent-users] Possible bug in http.c

[Ronen Mizrahi <ronen@xxxxxxxxxxxx>]

It looks like the patch does create issues with subsequent HTTP requests made on the same connection (for persistent connections) and therefore we won't submit it.

The original issue reported however is still very much in effect. It is particularly problematic for long-polling type requests where the server uses evhttp_send_reply_start() and then every once in a while does evhttp_send_reply_chunk(). If the client closes the connection, the server won't detect it during the period between different chunks but rather only when attempting to send the next chunk. This period can be several minutes so this can be a real issue for servers that handle many connections.

On Mon, Mar 3, 2014 at 3:30 PM, Nick Mathewson <nickm@xxxxxxxxxxxxx> wrote:

On Mon, Mar 3, 2014 at 3:06 PM, Ronen Mizrahi <ronen@xxxxxxxxxxxx> wrote:
> Thanks Nick. We have been testing it for the last few days and it seems to
> work well. What do you mean in terms of testing? Is there any automated set
> of tests that need to be made for proper verification?

It should pass the unit/regression tests.  (They run with "make
check".)  It would be good if they were include a check for the
specific case that this patch fixes, to make sure that the patch
really fixes it.

The current evhttp tests are in test/regress_http.c , but any kind of
little test program that fits into the testing infrastructure would be
fine with me.  (Don't take too many style cues from the contents of
test/regress_http.c -- many of the tests there follow a legacy testing
style.)

best wishes,
--
Nick
***********************************************************************
To unsubscribe, send an e-mail to majordomo@xxxxxxxxxxxxx with
unsubscribe libevent-users    in the body.