bug(?) + docs patch & question

[ed...@apache.org]

Hi --

I installed apache from the head of the CVS tree this morning (not the
2.0.15 tarball), and have seen some flakiness.  This is on SuSE 6.2 :

ed@ambient:~/apache > uname -a
Linux ambient 2.2.13 #1 Tue Nov 9 00:10:15 GMT 1999 i686 unknown

I'm using the default MPM (threaded) and I've seen the following occur
with the default set of modules.  The part which seems broken is
graceful restarts -- if I try that, I get a bunch of errors like the
following in error_log (one for each of the 8 children I've got):

[Sun Mar 25 11:00:08 2001] [warn] (9)Bad file descriptor: write pipe_of_death

Afterwards, bin/apachectl stop doesn't stop all httpd children (two are
left around); subsequent attempts to bind to the port fail 'cause there
are other processes still bound to it.  Is this a known issue/does this
affect others?  Possible a related issue -- when I use apachectl stop,
it includes one line which looks like the following:

[Sun Mar 25 11:43:31 2001] [warn] child process 31237 still did not
exit, sending a SIGTERM


Also -- the information in INSTALL about which modules would be enabled
by default didn't match what I found : mod_cgi and mod_status were not
included by default.  The former seems to be MPM specific, and it's
described in that way -- but I don't know about the latter.  Should
mod_status be disabled by default, or does it depend on the MPM
(perchild doesn't get it at all, as I understand)?  

I've attached a patch to fix a file in apache-site, which contains links
to the unofficial contrib stuff.

FYI -- I'm looking into writing a module to do limiting connections by
IP... I did something like this long ago, for Apache 1.2/1.3.  I'll be
happy to outline the approach I'd like to try, if anyone is interested.

thanks --

Ed

Re: bug(?) + docs patch & question

[rb...@covalent.net]

> I installed apache from the head of the CVS tree this morning (not the
> 2.0.15 tarball), and have seen some flakiness.  This is on SuSE 6.2 :

Not enough has changed with 2.0.16-dev to modify this behavior at all.
These are all issues in 2.0.15.  :-(

> ed@ambient:~/apache > uname -a
> Linux ambient 2.2.13 #1 Tue Nov 9 00:10:15 GMT 1999 i686 unknown
>
> I'm using the default MPM (threaded) and I've seen the following occur
> with the default set of modules.  The part which seems broken is
> graceful restarts -- if I try that, I get a bunch of errors like the
> following in error_log (one for each of the 8 children I've got):
>
> [Sun Mar 25 11:00:08 2001] [warn] (9)Bad file descriptor: write pipe_of_death

Hmmmmm...  I am not sure that piped logs and a threaded MPM have ever been
tried together.  I will try to look into this issue tonight or tomorrow
morning.

> Afterwards, bin/apachectl stop doesn't stop all httpd children (two are
> left around); subsequent attempts to bind to the port fail 'cause there
> are other processes still bound to it.  Is this a known issue/does this
> affect others?  Possible a related issue -- when I use apachectl stop,
> it includes one line which looks like the following:

This is not a known issue.  Please put it in the STATUS file.

> [Sun Mar 25 11:43:31 2001] [warn] child process 31237 still did not
> exit, sending a SIGTERM
>
>
> Also -- the information in INSTALL about which modules would be enabled
> by default didn't match what I found : mod_cgi and mod_status were not
> included by default.  The former seems to be MPM specific, and it's
> described in that way -- but I don't know about the latter.  Should
> mod_status be disabled by default, or does it depend on the MPM
> (perchild doesn't get it at all, as I understand)?

Mod_cgi is MPM specific.  If you are using a threaded MPM, you should get
mod_cgid.  This needs to be documented better.  All MPMs should work with
mod_status.  As for whether it is enabled by default, it looks like it is
only enabled when using --enable-module=most.  This is easy to change in
either the code or the docs.  I will look at what 1.3 did and follow suit.

> FYI -- I'm looking into writing a module to do limiting connections by
> IP... I did something like this long ago, for Apache 1.2/1.3.  I'll be
> happy to outline the approach I'd like to try, if anyone is interested.

Please outline it.  I would be very interested in the approach you are
going to take, and how quickly you are going to reject a request.

Ryan

_______________________________________________________________________________
Ryan Bloom                        	rbb@apache.org
406 29th St.
San Francisco, CA 94131
-------------------------------------------------------------------------------