/ Zope / Apsis / Pound Mailing List / Archive / 2010 / 2010-03 / Pound segfaults on first request

[ << ] [ >> ]

[ Pound mailing list emails considered spam with ... ] [ Problem with pound SSL / "Anne Moore" ... ]

Pound segfaults on first request
"(private) HKS" <hks.private(at)gmail.com>		 2010-03-10 23:20:52 [ SNIP ] 
I'm moving my Pound servers from OpenBSD to FreeBSD 8.0 AMD64.

I've compiled Pound 2.5 from source --with-maxbuf=16384 and included
PCRE. Each instance of Pound is throwing this error at its first
request:

pound: MONITOR: worker exited on signal 11, restarting

Subsequent requests are serviced normally. I'm using the same
configurations that I was using on the OpenBSD boxes (albeit with
relevant paths corrected). I've pasted my config below. Anybody seen
this before?


-HKS


User            "pound"
Group           "pound"
Control         "/var/run/pound-ctl.socket"
LogFacility     daemon
LogLevel        5
Alive           30
Client          30
TimeOut         900
Grace           30

# HTTP goes to HTTPS redirect
ListenHTTP
        Address 10.2.1.72
        Port 80
        RewriteLocation 2
        Service
                Backend
                        Address 10.2.1.11
                        Port 8080
                End
                Backend
                        Address 10.2.1.21
                        Port 8080
                End
                Backend
                        Address 10.2.1.22
                        Port 8080
                End
        End
End

ListenHTTPS
        Address 10.2.1.72
        Port 443
        Cert "/usr/local/etc/pound/private/cert.pem"
        Client 30
        RewriteLocation 2

        # Mercurial requests get big timeouts
        Service
                HeadRequire "[Aa][Cc][Cc][Ee][Pp][Tt]:
.*application\/mercurial.*"
                HeadRequire "[Uu][Ss][Ee][Rr]-[Aa][Gg][Ee][Nn][Tt]:
.*mercurial.*"
                Backend
                        Address 10.2.1.24
                        Port 80
                        # 12 hours
                        Timeout 43200
                End
                Backend
                        Address 10.2.1.30
                        Port 80
                        # 12 hours
                        Timeout 43200
                End
        End

        Service
                Backend
                        Address 10.2.1.24
                        Port 80
                End
                Backend
                        Address 10.2.1.30
                        Port 80
                End
        End
End

# Uh-Oh page in case everything else is hosed
Service
        Backend
                Address 127.0.0.1
                Port    8181
        End
End
Re: Pound segfaults on first request
"(private) HKS" <hks.private(at)gmail.com>		 2010-03-29 17:55:05 [ SNIP ] 
On Wed, Mar 10, 2010 at 6:20 PM, (private) HKS <hks.private(at)gmail.com>
wrote:
> I'm moving my Pound servers from OpenBSD to FreeBSD 8.0 AMD64.
>
> I've compiled Pound 2.5 from source --with-maxbuf=16384 and included
> PCRE. Each instance of Pound is throwing this error at its first
> request:
>
> pound: MONITOR: worker exited on signal 11, restarting
>
> Subsequent requests are serviced normally. I'm using the same
> configurations that I was using on the OpenBSD boxes (albeit with
> relevant paths corrected). I've pasted my config below. Anybody seen
> this before?
>
>
> -HKS
>
>
> User            "pound"
> Group           "pound"
> Control         "/var/run/pound-ctl.socket"
> LogFacility     daemon
> LogLevel        5
> Alive           30
> Client          30
> TimeOut         900
> Grace           30
>
> # HTTP goes to HTTPS redirect
> ListenHTTP
>        Address 10.2.1.72
>        Port 80
>        RewriteLocation 2
>        Service
>                Backend
>                        Address 10.2.1.11
>                        Port 8080
>                End
>                Backend
>                        Address 10.2.1.21
>                        Port 8080
>                End
>                Backend
>                        Address 10.2.1.22
>                        Port 8080
>                End
>        End
> End
>
> ListenHTTPS
>        Address 10.2.1.72
>        Port 443
>        Cert "/usr/local/etc/pound/private/cert.pem"
>        Client 30
>        RewriteLocation 2
>
>        # Mercurial requests get big timeouts
>        Service
>                HeadRequire "[Aa][Cc][Cc][Ee][Pp][Tt]:
> .*application\/mercurial.*"
>                HeadRequire "[Uu][Ss][Ee][Rr]-[Aa][Gg][Ee][Nn][Tt]:
> .*mercurial.*"
>                Backend
>                        Address 10.2.1.24
>                        Port 80
>                        # 12 hours
>                        Timeout 43200
>                End
>                Backend
>                        Address 10.2.1.30
>                        Port 80
>                        # 12 hours
>                        Timeout 43200
>                End
>        End
>
>        Service
>                Backend
>                        Address 10.2.1.24
>                        Port 80
>                End
>                Backend
>                        Address 10.2.1.30
>                        Port 80
>                End
>        End
> End
>
> # Uh-Oh page in case everything else is hosed
> Service
>        Backend
>                Address 127.0.0.1
>                Port    8181
>        End
> End
>


This seems to be related to configuring with --with-maxbuf=16384 on a
64-bit box (I don't have any 64-bit non-FreeBSD boxes, so I can't test
that). If I leave that out or use --with-maxbuf=8192, there is no
initial segfault.

Any idea what's going on here?

-HKS
Re: [Pound Mailing List] Re: Pound segfaults on first request
Robert Segall <roseg(at)apsis.ch>		 2010-03-30 16:54:35 [ SNIP ] 
On Mon, 2010-03-29 at 11:55 -0400, (private) HKS wrote:
> This seems to be related to configuring with --with-maxbuf384 on a
> 64-bit box (I don't have any 64-bit non-FreeBSD boxes, so I can't test
> that). If I leave that out or use --with-maxbufÂ92, there is no
> initial segfault.
> 
> Any idea what's going on here?

It might be a problem with the stack size, which is important on some
*BSD flavours. Try changing (version 2.5, pound.c line 359)

#ifdef  NEED_STACK
 /* set new stack size - necessary for OpenBSD/FreeBSD and Linux NPTL */
 if(pthread_attr_setstacksize(&attr, 1 << 18)) {
     logmsg(LOG_ERR, "can't set stack size - aborted");
     exit(1);
 }
#endif

to a higher value and see if it helps. Also make sure NEED_STACK is
defined.
-- 
Robert Segall
Apsis GmbH
Postfach, Uetikon am See, CH-8707
Tel: +41-32-512 30 19
Re: [Pound Mailing List] Re: Pound segfaults on first request
"(private) HKS" <hks.private(at)gmail.com>		 2010-03-30 17:38:12 [ SNIP ] 
On Tue, Mar 30, 2010 at 10:54 AM, Robert Segall <roseg(at)apsis.ch> wrote:
> On Mon, 2010-03-29 at 11:55 -0400, (private) HKS wrote:
>> This seems to be related to configuring with --with-maxbuf 384 on a
>> 64-bit box (I don't have any 64-bit non-FreeBSD boxes, so I can't test
>> that). If I leave that out or use --with-maxbuf 92, there is no
>> initial segfault.
>>
>> Any idea what's going on here?
>
> It might be a problem with the stack size, which is important on some
> *BSD flavours. Try changing (version 2.5, pound.c line 359)
>
> #ifdef  NEED_STACK
>  /* set new stack size - necessary for OpenBSD/FreeBSD and Linux NPTL */
>  if(pthread_attr_setstacksize(&attr, 1 << 18)) {
>     logmsg(LOG_ERR, "can't set stack size - aborted");
>     exit(1);
>  }
> #endif
>
> to a higher value and see if it helps. Also make sure NEED_STACK is
> defined.
> --
> Robert Segall
> Apsis GmbH
> Postfach, Uetikon am See, CH-8707
> Tel: +41-32-512 30 19
>
>
> --
> To unsubscribe send an email with subject unsubscribe to pound(at)apsis.ch.
> Please contact roseg(at)apsis.ch for questions.
>

Good shot: boosting the stack size to 512k (1 << 19) fixed it. I know
this probably slips into POSIX programming details, but can you help
me understand a bit more of why this was a problem?

-HKS
MailBoxer