Apache shouldn't be prematurely disconnecting sockets in the middle
of a response unless there is a serious problem (Eg. the Apache
child process is crashing). Could you describe how to reproduce this?
As for the patch, could you make this configurable with a command-line
option? That way the current functionality can stay default (meaning,
all recv() errors are fatal) and for those circumstances where the
user knows that there is some network-level or Apache-level problem
causing intermittent recv() errors, they can still get performance
results out of AB.
-aaron
On Mon, Feb 26, 2007 at 01:06:14PM -0700, Filip Hanik - Dev Lists wrote:
> I've created a small patch that lets ab continue even if it encounters
> an error on apr_socket_recv
>
> quite commonly, when servers are overloaded they disconnect the socket,
> ab receives a 104 (connection reset by peer) and the ab test exits.
>
> This patch logs the error, both counters correctly, cleans up the
> connection and continues.
>
> thoughts?
>
> Filip
> Index: ab.c
> ===================================================================
> --- ab.c (revision 511976)
> +++ ab.c (working copy)
> @@ -1332,7 +1332,10 @@
> err_except++; /* XXX: is this the right error counter? */
> /* XXX: Should errors here be fatal, or should we allow a
> * certain number of them before completely failing? -aaron */
> - apr_err("apr_socket_recv", status);
> + //apr_err("apr_socket_recv", status);
> + bad++;
> + close_connection(c);
> + return;
> }
> }
>
