On Thu, 2019-12-05 at 18:31 -0800, Ryan Schmitt wrote:
> I'm working on an HTTP client integration that will allow Apache 5's
> async
> client to be used as the HTTP backend for the AWS Java SDK [1]. One
> of the
> first smoke tests I ran failed because the server responded to my
> request
> with a TCP RST. I quickly realized that it was doing this because I
> was
> sending an unencrypted request over port 443. To make things weirder,
> this
> was taking place *after* a TLS session had been successfully
> negotiated.
>
> I was having this problem on beta6, and I noticed that the most
> recent
> commit to the client, 070f30fdc ("Fixed session i/o and wire logging
> in
> async clients"), had modified some async wire logging code that I was
> starting to grow suspicious of during my debugging. I upgraded to
> beta7-SNAPSHOT, and sure enough, the problem went away. I'd like to
> get a
> better idea of what the problem is, and in particular, why my other
> HttpClient 5 integration doesn't seem to have this problem. (Is there
> a
> workaround I can use on beta6?)
>
Hi Ryan
It was my bad. I tweaked i/o session initialization logic in core and
forgot to adapt logging code in client after upgrade to the latest core
BETA.
The defect only manifests itself when running with i/o session and wire
logging on, which is not the case by default. That is the reason the
regression has gone undetected for so long.
If you do not need to activate i/o session or wire logging the bug
should never trigger with beta6.
Oleg
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
