145392 - some images do not load [apache CGI pipelining problem]

Status: RESOLVED WORKSFORME

(Core :: Networking, defect)

Description

[Darin Fisher]

some images do not load when the images are all served up via the same CGI
script. apache seems to have difficulty with pipelined requests to the same CGI
script, and it seems to fail in a way that is not recoverable.  we may have to
work around this bug by verifying that each pipelined request is for a unique
URL (minus the query and/or fragment).

Comment 1

[joakim_46]

WFM with build 2002051006. Reporter, please try with a newer build.

Comment 2

[Darin Fisher]

try enabling pipelining in 'Preferences->Advanced->HTTP Networking' ... you will
see the bug.

Comment 3

[Dave]

If you have pipelining enabled, then might this bug be a dupe of Bug 140561? 
(also see Bug 141179)

Comment 4

[Darin Fisher]

yup, this bug might be a dup of those other bugs, or it might not be.  i have to
investigate the cause of those bugs to know for certain.

Comment 5

[a_geek]

hi,

I'm unsure about this. With build 2002060721 (linux) I control both client and
server, and serve the images via an Apache ErrorDocument redirection. With
respect to _this_ bug (there are other problems as well, all not experienced
with Netscape 4.77) some images load, others don't, and the log contains:

(date) [info] [client x.x.x.x (32)Broken pipe: client stopped connection before
rflush completed

Some of the image doesn't load. Reloading the page several times yields
that sometimes eg. pictures 1, 3 and 7 load, while on a different attemt
pictures 1, 2, 3, and 6 load. Sometimes not even all requests are logged
in the error log where I usually find "File does not exist: /path/to/img.png"
(for example, with the path refering to the directory where the error
document is set up).

If this is a client or server problem, I don't know.

This is Apache 1.2.24+SSL as in OBSD 3.1 -stable (ie, patched for
chunked encoding and SSH/SSL problems) on the server side.

Btw, I have pipelining _OFF_.

Any hints on where to look (and how to correct this) are welcome!

Comment 6

[Darin Fisher]

[e-mail]: sounds like you're describing a different bug.  this bug is only
concerned with apache 1.3's problem with pipelined requests to CGI scripts. 
that said, you're welcome to file a new bug on the problem you're seeing, but
please try to provide a testcase so mozilla developers will be able to reproduce
the problem you are seeing.  thx!!

Comment 7

[Marek Z. Jeziorek]

batch: adding topembed per Gecko2 document
http://rocknroll.mcom.com/users/marek/publish/Gecko/Gecko2Tasks.html

Comment 8

[David G King]

Using the test URL, this WFM with build 20021001 on WinXP SP1.

Comment 9

[Michael Buckland]

Discussed in topembed bug triage.  Minusing.

Comment 10

[Darin Fisher]

-> default owner

Comment 11

[John Vandenberg]

The old server (tegu.mozilla.org) no longer exists.  netcraft says it was running Apache/1.3.12 and then Apache/1.3.26 with mod_throttle/3.1.2
http://toolbar.netcraft.com/site_report?url=tegu.mozilla.org

Could it have been running mod_throttle at the time this bug was reported?

The new server (build-graphs.mozilla.org) is running Apache/2.2.3 (Red Hat) so it isn't a valid test for this bug, but Apache 2 looks to be OK:
http://build-graphs.mozilla.org/multiquery.cgi?tboxes=comet,luna,sleestack,mecca,facedown,openwound,rheeeet&testname=startup&autoscale=1&days=0&units=&ltype=steps&points=1&avg=1

I can't find the source for multiquery.cgi in order to replicate what it does.

My little test perl script emits an image after a sleep of between 0 and 10secs.  I call it from a static html page to display the nine icons/pieX.gif bundled with Apache httpd.  It works as expected on apache 1.3.12, .24 & .42.

I've also set up http://www.ivor.org/throttle/, but that doesn't cause breakages.

Comment 12

[Matthias Versen [:Matti]]

Without testcase and comment#11 i vote on marking this wfm

Comment 13

[Ivan Baldo]

Maybe this bug should be closed now.