You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@httpd.apache.org by Ryan Bloom <rb...@ntrnet.net> on 2000/03/20 19:35:50 UTC
FreeBSD core'ing.
I have been able to figure out why FreeBSD is core'ing with Apache 2.0.
Or if not core'ing, is at least not running. We are choosing the wrong
locking mechanism for FreeBSD. We should be using flock or fcntl, but APR
is choosing SYSVSEM. I'll be looking into this verys oon, unless somebody
else wants to fix it. :->
Ryan
_______________________________________________________________________________
Ryan Bloom rbb@apache.org
406 29th St.
San Francisco, CA 94131
-------------------------------------------------------------------------------
Re: FreeBSD core'ing.
Posted by Jeff Trawick <tr...@bellsouth.net>.
SysV sems work fine on FreeBSD. It is the apache 2.0
code which isn't cleaning them up correctly.
--
Jeff Trawick | trawick@ibm.net | PGP public key at web site:
http://www.geocities.com/SiliconValley/Park/9289/
Born in Roswell... married an alien...
Re: FreeBSD core'ing.
Posted by rb...@apache.org.
But, that's just it. SysV sems shouldn't be chosen if they don't work.
They aren't the first or default choice. They are the second choice, and
they aren't used unless they work on a given platform, at least that's the
idea. It is a bug, but it is in choosing the locking mechanism. I guess
I am looking at the fact that the code hasn't changed between 1.3 and 2.0,
so we weren't reporting the error before, because the error doesn't exist
if good choices are made.
Ryan
On Mon, 20 Mar 2000, Greg Stein wrote:
> But a valid point was made (forget who): if SysV sems *are* chosen, then
> it shouldn't crash. That should be fixed at some point.
>
> If you select a different locking mechanism, then the sem bug is still
> there.
>
> Cheers,
> -g
>
> On Mon, 20 Mar 2000 rbb@apache.org wrote:
> > I have looked at these really quickly, and I don't think they are
> > necessary. I will be looking at these in mroe detail later today. It
> > seems to me that if we change the lock choosing logic to be correct, we
> > don't need some of these changes. I'll post more detail opn my opinions
> > later.
> >
> > Ryan
> >
> > > > If you have my patches from Feb 29 it seems to work fine
> > > > w.r.t. locks. Changing it not to use SysV sems may be
> > > > the right thing to do for FreeBSD performance-wise, but
> > > > that simply moves the bad lock symptoms to some obscure
> > > > platform.
> > >
> > > Same here. I believe they where commited.
> >
> >
> > _______________________________________________________________________________
> > Ryan Bloom rbb@apache.org
> > 406 29th St.
> > San Francisco, CA 94131
> > -------------------------------------------------------------------------------
> >
> >
>
> --
> Greg Stein, http://www.lyra.org/
>
>
_______________________________________________________________________________
Ryan Bloom rbb@apache.org
406 29th St.
San Francisco, CA 94131
-------------------------------------------------------------------------------
Re: FreeBSD core'ing.
Posted by Greg Stein <gs...@lyra.org>.
But a valid point was made (forget who): if SysV sems *are* chosen, then
it shouldn't crash. That should be fixed at some point.
If you select a different locking mechanism, then the sem bug is still
there.
Cheers,
-g
On Mon, 20 Mar 2000 rbb@apache.org wrote:
> I have looked at these really quickly, and I don't think they are
> necessary. I will be looking at these in mroe detail later today. It
> seems to me that if we change the lock choosing logic to be correct, we
> don't need some of these changes. I'll post more detail opn my opinions
> later.
>
> Ryan
>
> > > If you have my patches from Feb 29 it seems to work fine
> > > w.r.t. locks. Changing it not to use SysV sems may be
> > > the right thing to do for FreeBSD performance-wise, but
> > > that simply moves the bad lock symptoms to some obscure
> > > platform.
> >
> > Same here. I believe they where commited.
>
>
> _______________________________________________________________________________
> Ryan Bloom rbb@apache.org
> 406 29th St.
> San Francisco, CA 94131
> -------------------------------------------------------------------------------
>
>
--
Greg Stein, http://www.lyra.org/
Re: FreeBSD core'ing.
Posted by rb...@apache.org.
I have looked at these really quickly, and I don't think they are
necessary. I will be looking at these in mroe detail later today. It
seems to me that if we change the lock choosing logic to be correct, we
don't need some of these changes. I'll post more detail opn my opinions
later.
Ryan
> > If you have my patches from Feb 29 it seems to work fine
> > w.r.t. locks. Changing it not to use SysV sems may be
> > the right thing to do for FreeBSD performance-wise, but
> > that simply moves the bad lock symptoms to some obscure
> > platform.
>
> Same here. I believe they where commited.
_______________________________________________________________________________
Ryan Bloom rbb@apache.org
406 29th St.
San Francisco, CA 94131
-------------------------------------------------------------------------------
Re: FreeBSD core'ing.
Posted by Dirk-Willem van Gulik <di...@webweaving.org>.
Jeff Trawick wrote:
>
> If you have my patches from Feb 29 it seems to work fine
> w.r.t. locks. Changing it not to use SysV sems may be
> the right thing to do for FreeBSD performance-wise, but
> that simply moves the bad lock symptoms to some obscure
> platform.
Same here. I believe they where commited.
DW.
Re: FreeBSD core'ing.
Posted by Jeff Trawick <tr...@bellsouth.net>.
If you have my patches from Feb 29 it seems to work fine
w.r.t. locks. Changing it not to use SysV sems may be
the right thing to do for FreeBSD performance-wise, but
that simply moves the bad lock symptoms to some obscure
platform.
--
Jeff Trawick | trawick@ibm.net | PGP public key at web site:
http://www.geocities.com/SiliconValley/Park/9289/
Born in Roswell... married an alien...
Re: FreeBSD core'ing.
Posted by Dirk-Willem van Gulik <di...@webweaving.org>.
Ryan Bloom wrote:
>
> I have been able to figure out why FreeBSD is core'ing with Apache 2.0.
> Or if not core'ing, is at least not running. We are choosing the wrong
> locking mechanism for FreeBSD. We should be using flock or fcntl, but APR
> is choosing SYSVSEM. I'll be looking into this verys oon, unless somebody
> else wants to fix it. :->
Hmm on 3.4, 4.0 and 5.0 it no longer core's for me since those init fixes
where applied. That was for SYSVSEM.
There are not that many thouhg in the default kernel install.
Dw
Re: FreeBSD core'ing.
Posted by rb...@apache.org.
> > The problem isn't access to a FreeBSD machine, it is fast access to on.
> > Currently, my only access is a 56K modem. I am working on getting DSL,
> > and as soon as I have an office, this will be even less of an issue. I
> > just can't tie up my phone line for three hours to fix this right now.
>
> Wander down to Borders bookshop, or take the BART up to walnut creek and
> fetch the FreeBSD cdrom. It's the last station :-) For what you need
> the most minimal install would be fine. Jordan might give you a CD if
> you show up in person at the walnut creek offices and ask nicely :-)
> I'd send you one if it was cost effective :-) But italy is too far.
I'll do that later today, I guess I can make due with two Linux boxes at
home. :-)
_______________________________________________________________________________
Ryan Bloom rbb@apache.org
406 29th St.
San Francisco, CA 94131
-------------------------------------------------------------------------------
Re: FreeBSD core'ing.
Posted by Dirk-Willem van Gulik <di...@webweaving.org>.
rbb@apache.org wrote:
>
> > > I would like to have this fixed before the 2.0a2. I am waiting until I
> > > have faster access to a FreeBSD box (hopefully tomorrow) before I devote
> > > real time to fixing this.
> >
> > Send my your public ssh key and I can arrange this.
>
> The problem isn't access to a FreeBSD machine, it is fast access to on.
> Currently, my only access is a 56K modem. I am working on getting DSL,
> and as soon as I have an office, this will be even less of an issue. I
> just can't tie up my phone line for three hours to fix this right now.
Wander down to Borders bookshop, or take the BART up to walnut creek and
fetch the FreeBSD cdrom. It's the last station :-) For what you need
the most minimal install would be fine. Jordan might give you a CD if
you show up in person at the walnut creek offices and ask nicely :-)
I'd send you one if it was cost effective :-) But italy is too far.
Dw
Re: FreeBSD core'ing.
Posted by rb...@apache.org.
> > I would like to have this fixed before the 2.0a2. I am waiting until I
> > have faster access to a FreeBSD box (hopefully tomorrow) before I devote
> > real time to fixing this.
>
> Send my your public ssh key and I can arrange this.
The problem isn't access to a FreeBSD machine, it is fast access to on.
Currently, my only access is a 56K modem. I am working on getting DSL,
and as soon as I have an office, this will be even less of an issue. I
just can't tie up my phone line for three hours to fix this right now.
Ryan
_______________________________________________________________________________
Ryan Bloom rbb@apache.org
406 29th St.
San Francisco, CA 94131
-------------------------------------------------------------------------------
Re: FreeBSD core'ing.
Posted by Dirk-Willem van Gulik <di...@webweaving.org>.
rbb@apache.org wrote:
>
> I think FreeBSD runs out of semaphores very quickly with the default
> compilation. On locus, we aren't core'ing anymore, but we get an error
> message saying there aren't enough semaphores to create a new one. We are
> generating the correct error message, but we shouldn't be receiving that
> message, because we should be using a different locking mechanism.
>
> I would like to have this fixed before the 2.0a2. I am waiting until I
> have faster access to a FreeBSD box (hopefully tomorrow) before I devote
> real time to fixing this.
Send my your public ssh key and I can arrange this.
Dw
Re: FreeBSD core'ing.
Posted by Chuck Murcko <ch...@topsail.org>.
Yes, I think the compiled default is 10 semaphores.
rbb@apache.org wrote:
>
> I think FreeBSD runs out of semaphores very quickly with the default
> compilation. On locus, we aren't core'ing anymore, but we get an error
> message saying there aren't enough semaphores to create a new one. We are
> generating the correct error message, but we shouldn't be receiving that
> message, because we should be using a different locking mechanism.
>
> I would like to have this fixed before the 2.0a2. I am waiting until I
> have faster access to a FreeBSD box (hopefully tomorrow) before I devote
> real time to fixing this.
>
--
Chuck
Chuck Murcko
Here: chuck.murcko@trintech.com
There: chuck@topsail.org
Re: FreeBSD core'ing.
Posted by rb...@apache.org.
I think FreeBSD runs out of semaphores very quickly with the default
compilation. On locus, we aren't core'ing anymore, but we get an error
message saying there aren't enough semaphores to create a new one. We are
generating the correct error message, but we shouldn't be receiving that
message, because we should be using a different locking mechanism.
I would like to have this fixed before the 2.0a2. I am waiting until I
have faster access to a FreeBSD box (hopefully tomorrow) before I devote
real time to fixing this.
Ryan
> That's only half an answer, though, surely SYSVSEM should work on
> FreeBSD even if it isn't appropriate?
>
> I assume this means there's aither a fault in the error reporting or in
> SYSVSEM which should be fixed _before_ switching FreeBSD to using
> another method?
>
> BTW ... damn! I'd come to this conclusion before and forgotten about it.
>
> Cheers,
>
> Ben.
>
> --
> SECURE HOSTING AT THE BUNKER: http://www.thebunker.net/hosting.htm
>
> http://www.apache-ssl.org/ben.html
>
> Coming to ApacheCon Europe? http://ApacheCon.Com/
>
_______________________________________________________________________________
Ryan Bloom rbb@apache.org
406 29th St.
San Francisco, CA 94131
-------------------------------------------------------------------------------
Re: FreeBSD core'ing.
Posted by Ben Laurie <be...@algroup.co.uk>.
Ryan Bloom wrote:
>
> I have been able to figure out why FreeBSD is core'ing with Apache 2.0.
> Or if not core'ing, is at least not running. We are choosing the wrong
> locking mechanism for FreeBSD. We should be using flock or fcntl, but APR
> is choosing SYSVSEM. I'll be looking into this verys oon, unless somebody
> else wants to fix it. :->
That's only half an answer, though, surely SYSVSEM should work on
FreeBSD even if it isn't appropriate?
I assume this means there's aither a fault in the error reporting or in
SYSVSEM which should be fixed _before_ switching FreeBSD to using
another method?
BTW ... damn! I'd come to this conclusion before and forgotten about it.
Cheers,
Ben.
--
SECURE HOSTING AT THE BUNKER: http://www.thebunker.net/hosting.htm
http://www.apache-ssl.org/ben.html
Coming to ApacheCon Europe? http://ApacheCon.Com/
Re: FreeBSD core'ing.
Posted by Chuck Murcko <ch...@topsail.org>.
Marc Slemko wrote:
>
> On Mon, 20 Mar 2000, Ryan Bloom wrote:
>
> >
> >
> > I have been able to figure out why FreeBSD is core'ing with Apache 2.0.
> > Or if not core'ing, is at least not running. We are choosing the wrong
> > locking mechanism for FreeBSD. We should be using flock or fcntl, but APR
> > is choosing SYSVSEM. I'll be looking into this verys oon, unless somebody
> > else wants to fix it. :->
>
> But shouldn't SYSVSEM work anyway on FreeBSD, assuming the kernel is
> compiled right? Or is it that the kernel isn't compiled right but we
> still choose it?
I think the distributions may not have SYSVSEM enabled by default, so
you'd have to rebuild the kernel to use. I thought that had changed
since 3.0, though. 2.x never had SYSVSEM built.
--
Chuck
Chuck Murcko
Here: chuck.murcko@trintech.com
There: chuck@topsail.org
Re: FreeBSD core'ing.
Posted by rb...@apache.org.
> > I have been able to figure out why FreeBSD is core'ing with Apache 2.0.
> > Or if not core'ing, is at least not running. We are choosing the wrong
> > locking mechanism for FreeBSD. We should be using flock or fcntl, but APR
> > is choosing SYSVSEM. I'll be looking into this verys oon, unless somebody
> > else wants to fix it. :->
>
> But shouldn't SYSVSEM work anyway on FreeBSD, assuming the kernel is
> compiled right? Or is it that the kernel isn't compiled right but we
> still choose it?
I am under the impression that we are choosing SYSVSEM regardless of how
the kernel is compiled.
Ryan
_______________________________________________________________________________
Ryan Bloom rbb@apache.org
406 29th St.
San Francisco, CA 94131
-------------------------------------------------------------------------------
Re: FreeBSD core'ing.
Posted by Marc Slemko <ma...@znep.com>.
On Mon, 20 Mar 2000, Ryan Bloom wrote:
>
>
> I have been able to figure out why FreeBSD is core'ing with Apache 2.0.
> Or if not core'ing, is at least not running. We are choosing the wrong
> locking mechanism for FreeBSD. We should be using flock or fcntl, but APR
> is choosing SYSVSEM. I'll be looking into this verys oon, unless somebody
> else wants to fix it. :->
But shouldn't SYSVSEM work anyway on FreeBSD, assuming the kernel is
compiled right? Or is it that the kernel isn't compiled right but we
still choose it?