The following PR was posted to FreeBSD ports. I know someone did a lot
of ab work recently.
Any thoughts on this ? I'm in a position to test things on FreeBSD.
--- Begin Message ---
>Number: 125464
>Category: ports
>Synopsis: [patch] Apache Bench in www/apache22 doesn't work
>Confidential: no
>Severity: serious
>Priority: medium
>Responsible: freebsd-ports-bugs
>State: open
>Quarter:
>Keywords:
>Date-Required:
>Class: sw-bug
>Submitter-Id: current-users
>Arrival-Date: Thu Jul 10 07:40:01 UTC 2008
>Closed-Date:
>Last-Modified:
>Originator: MQ
>Release: 7.0-RELEASE-p2
>Organization:
>Environment:
FreeBSD q6600.macro 7.0-RELEASE-p2 FreeBSD 7.0-RELEASE-p2 #0: Wed Jun 18
06:48:16 UTC 2008 [EMAIL PROTECTED]:/usr/obj/usr/src/sys/GENERIC amd64
>Description:
Apache Bench in www/apache22 doesn't work, it always encounter connection reset.
# ./ab -c3 -n100 http://localhost/
This is ApacheBench, Version 2.3 <$Revision: 655654 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/
Benchmarking localhost (be patient)...Send request failed!
apr_socket_recv: Connection reset by peer (54)
Total of 2 requests completed
#
This problem is caused by Apache Bench's improper use of the kqueue support in
APR.
>How-To-Repeat:
cd /usr/ports/www/apache22; make BATCH=yes; cd
/usr/ports/www/apache22/work/httpd-2.2.9/support; ./ab -n100 -c10
http://localhost/
>Fix:
Use the attached patch could solve this problem. But I haven't test how https
connections would be influenced.
There are three pieces of modified codes. The second and third modifications
can increase the performance of Apache Bench.
Patch attached with submission follows:
--- support/ab.c.orig 2008-05-28 20:58:01.000000000 +0800
+++ support/ab.c 2008-07-10 15:28:11.000000000 +0800
@@ -670,6 +670,13 @@
c->state = STATE_READ;
c->endwrite = lasttime = apr_time_now();
+ {
+ apr_pollfd_t remove_pollfd;
+ remove_pollfd.desc_type = APR_POLL_SOCKET;
+ remove_pollfd.desc.s = c->aprsock;
+ remove_pollfd.reqevents = APR_POLLOUT;
+ apr_pollset_remove(readbits, &remove_pollfd);
+ }
{
apr_pollfd_t new_pollfd;
new_pollfd.desc_type = APR_POLL_SOCKET;
@@ -1676,6 +1683,7 @@
const apr_pollfd_t *pollresults;
n = concurrency;
+ apr_time_now();
status = apr_pollset_poll(readbits, aprtimeout, &n, &pollresults);
if (status != APR_SUCCESS)
apr_err("apr_poll", status);
@@ -1759,22 +1767,6 @@
write_request(c);
}
}
-
- /*
- * When using a select based poll every time we check the bits
- * are reset. In 1.3's ab we copied the FD_SET's each time
- * through, but here we're going to check the state and if the
- * connection is in STATE_READ or STATE_CONNECTING we'll add the
- * socket back in as APR_POLLIN.
- */
- if (c->state == STATE_READ) {
- apr_pollfd_t new_pollfd;
- new_pollfd.desc_type = APR_POLL_SOCKET;
- new_pollfd.reqevents = APR_POLLIN;
- new_pollfd.desc.s = c->aprsock;
- new_pollfd.client_data = c;
- apr_pollset_add(readbits, &new_pollfd);
- }
}
} while (lasttime < stoptime && done < requests);
>Release-Note:
>Audit-Trail:
>Unformatted:
_______________________________________________
[EMAIL PROTECTED] mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports-bugs
To unsubscribe, send any mail to "[EMAIL PROTECTED]"
--- End Message ---
