Re: DEBUG_FULL and other DEBUG_*

On Tue, Feb 09, 2010 at 10:29:02AM -0800, Paul Hirose wrote:
> On Tue, Feb 9, 2010 at 9:48 AM, Willy Tarreau <w#1wt.eu> wrote:
> > On Mon, Feb 08, 2010 at 12:35:32PM -0800, Paul Hirose wrote:
> >> The error earlier reported during -dev7 about not compiling with
> >> DEBUG_FULL is still present in rc1 and ss-20100208.  As that msg
> >> indicated, changing it f rom req to rep allowed the compile.
> >
> > grrr, it seems I lost the patch. Care to resend it please ?
>
> It wasn't a diff-type patch so much as a sentence describing it :) In
> src/session.c around line 736 in function process_store_rules() you
> just have "req" in the DPRINTF statement, and it should be a "rep".
> I found mention of it at
> http://marc.info/?l=haproxy&m=126454210721922&w=2

Ah OK thanks, I'll do that.

(...)
> > If you get the chance to find a user who can reliably reproduce the
> > issue, do not hesitate, start a tcpdump capture and you'll be fixed !
>
> That'll be the next step. Alas, it's not reproducible on demand, but
> happens "reliably" after a while. We know it will happen, just not
> quite when or why or how :) Just depends on how big a tcpdump we want
> :)

I wrote a script a long time ago to rotate tcpdump captures. That made it possible to capture for several days if required. Now tcpdump supports a file rotate, so that script has less use. Once you know the date and IP, it's easy to isolate that user's traffic. Using very simple scripts, it's easy too to spot retransmits and connection failures.

Regards,
Willy Received on 2010/02/09 19:42