Placing a recent snapshot on daedalus?

[Justin Erenkrantz <je...@ebuilt.com>]

Would it be possible to place a fairly current snapshot of httpd-2.0
up on daedalus?

I'd really like to get the input filter changes hammered on.  If
it is broken, the sooner we get feedback, the better.  I know that
there are tons of other fixes (or "new features") that have entered
recently that should also be tested in the "real world."

I don't know if we can either place it on apache.org or put it on
a different port, but with a log replay.  I'll let those who do
this sort of thing handle that.  If it doesn't pass basic sanity
tests (like the looping reads I just fixed), I'd like to know that
as well.

I know SSL is still broken and proxy may or may not be working (Ian?),
but I don't believe that they are used by daedalus.  

I'm still frustrated with mod_ssl.  I may just commit what I have
so that others can possibly jump in here.  I dunno what the right
thing to do here is.  Should I wait until it is perfect, or should
I just commit what I have?  (Remember, it is broken right now.)
-- justin

Re: Placing a recent snapshot on daedalus?

[Daniel Stone <da...@sfarc.net>]

On Wed, Oct 10, 2001 at 12:36:51PM -0700, Ryan Bloom wrote:
> On Wednesday 10 October 2001 12:17 pm, Justin Erenkrantz wrote:
> > Would it be possible to place a fairly current snapshot of httpd-2.0
> > up on daedalus?
> >
> > I'd really like to get the input filter changes hammered on.  If
> > it is broken, the sooner we get feedback, the better.  I know that
> > there are tons of other fixes (or "new features") that have entered
> > recently that should also be tested in the "real world."
> 
> Daedalus is NOT a location to be sanity checking new code.  When
> www.apache.org goes down, people notice.  The current code should
> not be placed on www.apache.org until it has been tested extensively
> on a lot of other machines.

Depends. I made the call to put a CVS snapshot from the 1st of October
into the Debian package (which is waiting to be uploaded as I'm not a
Developer; still in the process), and another snapshot before that,
because it's been so long since a release and, by the looks of it,
several important fixes have gone in. I will continue to track CVS until
you can find a release stable enough to call beta, or do weekly releases
or whatever.

-d

-- 
Daniel Stone						    <da...@sfarc.net>
<Overfiend> OH, COME ON, LET ME FLAME THE FUCKER
<rcw> does anyone else think it's unhealthy for Overfiend to keep his feelings
to himself?

Re: Placing a recent snapshot on daedalus?

[Aaron Bannert <aa...@clove.org>]

On Wed, Oct 10, 2001 at 12:43:26PM -0700, Justin Erenkrantz wrote:
[...]
> I was also suggesting that a CVS snapshot be placed against a
> log replay - I think Greg and some others have done this in the
> past to "simulate" Daedalus's load.  But, this code really needs
> to interact with real browsers and real latency.

I've always been curious how you guys perform log replays, are the
scripts for doing this available anywhere? (assuming they are scripts :)

> This can't be
> easily simulated with any test tool. 

As the other 50% of the development inspiration of our beloved flood
load testing tool, I'll just pretend I didn't hear that... ;)

To be fair, taking CERN logs into flood is one of design goals that is
currently less than adequate, but can still be done with some script
magic.

-aaron

Re: Placing a recent snapshot on daedalus?

[Justin Erenkrantz <je...@ebuilt.com>]

On Wed, Oct 10, 2001 at 12:36:51PM -0700, Ryan Bloom wrote:
> Daedalus is NOT a location to be sanity checking new code.  When
> www.apache.org goes down, people notice.  The current code should
> not be placed on www.apache.org until it has been tested extensively
> on a lot of other machines.

I realize that, but I have tested it everywhere where I can do so
(it is running on apachelabs.org).  AFAIK, the core input filters
are working correctly.  There have been some minor quibbles which 
are getting fixed as fast as they are being pointed out.  At this
point, it just needs to be tested and used.

I was also suggesting that a CVS snapshot be placed against a
log replay - I think Greg and some others have done this in the
past to "simulate" Daedalus's load.  But, this code really needs
to interact with real browsers and real latency.  This can't be
easily simulated with any test tool. 

I was just thinking that this might be the quickest way to get it
some testing since people don't have the time to actually test
this stuff out on a regular basis.  This would force it to get 
used and all of the bugs to be worked out.  The sooner we find
them, the sooner the problems are fixed.  -- justin

Re: Placing a recent snapshot on daedalus?

[Jon Travis <jt...@covalent.net>]

On Wed, Oct 10, 2001 at 12:36:51PM -0700, Ryan Bloom wrote:
> On Wednesday 10 October 2001 12:17 pm, Justin Erenkrantz wrote:
> > Would it be possible to place a fairly current snapshot of httpd-2.0
> > up on daedalus?
> >
> > I'd really like to get the input filter changes hammered on.  If
> > it is broken, the sooner we get feedback, the better.  I know that
> > there are tons of other fixes (or "new features") that have entered
> > recently that should also be tested in the "real world."
> 
> Daedalus is NOT a location to be sanity checking new code.  When
> www.apache.org goes down, people notice.  The current code should
> not be placed on www.apache.org until it has been tested extensively
> on a lot of other machines.

Well, right now, anyone on the planet with the skill to write a couple
lines of Python and an internet connection can knock www.apache.org off the
net.  As long as this new build passes the testsuite, I think it would
be good for server development.  As I recall, bugs which took down Apache.org
when it first started running on 2.0 got fixed within minutes.

-- Jon

Re: Placing a recent snapshot on daedalus?

[Greg Ames <gr...@remulak.net>]

Greg Ames wrote:

> Amen to what Ryan said.  We're not ready to go live yet.  
> 
> * no error log message or dump after a seg fault on Linux.  

speaking of seg faults with no diagnostic info, I just tried a log
replay on my ThinkPad.  It ran for a little while then stopped dead.
Last week, it ran much longer. 

[gregames@gandalf gregames]$ hps
11161     1 040 do_sel S pts/1    00:00:00 httpd
11162 11161 040 unix_a S pts/1    00:00:00 httpd

looks like something happened to all the worker processes - these two
are the parent and the cgid daemon process.  hmmm, even so,
perform_idle_server_maintenance ought to wake up and create more, no
matter what.  gdb time.  this is prefork, btw.

Greg

Re: Placing a recent snapshot on daedalus?

[Greg Ames <gr...@remulak.net>]

Ryan Bloom wrote:
> 
> On Wednesday 10 October 2001 12:17 pm, Justin Erenkrantz wrote:
> > Would it be possible to place a fairly current snapshot of httpd-2.0
> > up on daedalus?
> >
> > I'd really like to get the input filter changes hammered on.  If
> > it is broken, the sooner we get feedback, the better.  I know that
> > there are tons of other fixes (or "new features") that have entered
> > recently that should also be tested in the "real world."
> 
> Daedalus is NOT a location to be sanity checking new code.  When
> www.apache.org goes down, people notice.  The current code should
> not be placed on www.apache.org until it has been tested extensively
> on a lot of other machines.

Amen to what Ryan said.  We're not ready to go live yet.  There have
been extensive changes since 2.0.24, and we're still dealing with the
fallout.  There's some mod_negotiation log msgs that need to be
investigated, and some 404's which may or may not be legit.

btw, I'd love it if someone wanted to pitch in and help with the log
replay testing for daedalus.  It's not perfect, it eats disk space and
your time, but it certainly seems to shake out bugs.  

Or just help out with these:

* no error log message or dump after a seg fault on Linux.  Behavior
unknown on other platforms.
* http://jakarta.apache.org/cvsweb/index.cgi gets a 404, even on the
live 2.0.24 server.  It seems to be a popular link, and is a symlink to
gstein's most excellent viewcvs.  whassup?

[gregames@daedalus gregames]$ ls -l
/www/jakarta.apache.org/cvsweb/index.cgi
lrwxrwxrwx  1 root  jakarta  38 Dec 30  2000
/www/jakarta.apache.org/cvsweb/index.cgi ->
/www/www.apache.org/websrc/viewcvs.cgi

Greg

Re: Placing a recent snapshot on daedalus?

[Ryan Bloom <rb...@covalent.net>]

On Wednesday 10 October 2001 12:17 pm, Justin Erenkrantz wrote:
> Would it be possible to place a fairly current snapshot of httpd-2.0
> up on daedalus?
>
> I'd really like to get the input filter changes hammered on.  If
> it is broken, the sooner we get feedback, the better.  I know that
> there are tons of other fixes (or "new features") that have entered
> recently that should also be tested in the "real world."

Daedalus is NOT a location to be sanity checking new code.  When
www.apache.org goes down, people notice.  The current code should
not be placed on www.apache.org until it has been tested extensively
on a lot of other machines.

Ryan


______________________________________________________________
Ryan Bloom				rbb@apache.org
Covalent Technologies			rbb@covalent.net
--------------------------------------------------------------

Re: Placing a recent snapshot on daedalus?

[ia...@apache.org]

In article <20...@ebuilt.com> you wrote:
<snip>
> 
> I know SSL is still broken and proxy may or may not be working (Ian?),
> but I don't believe that they are used by daedalus.  
> 
It seems to work, but i'm not confident about it's dechunking.

I'm off to Australia for 2 1/2 weeks, so I can't fix/break it anymore

> -- justin


..Ian