64290 - Turning on Keep Alive makes page load indefinitely

Status: VERIFIED INVALID

(Core :: Networking: HTTP, defect)

Description

[slipszi]

Turn on Keep Alive in Preferences > Debug > Networking, and try to load the URL:

http://www.yunord.net/cgi/click_counter.cgi?dump=ftp

The page will be loaded but the throbber and the status bar will keep on running
indefinitely. If you turn off the Keep Alive, the page loads without a problem.

Note that the page has no Content-Length specified.

This was tested on Win98 running Mozilla build 2000122805.

Comment 1

[gabriel]

I think this might be a dupe of bug 39310.

Comment 2

[gabriel]

My mistake. It's a dupe of bug 68811.

Comment 3

[Stephan Niemz]

I think both are duplicates of the same bug.  Please reopen if you disagree.

*** This bug has been marked as a duplicate of 39310 ***

Comment 4

[John Morrison]

Reopening. bug 39310 is lumping too many causes together under the banner
'make it stop spinning'. This bug has a nice, clear definition of a problem
and it is 100% reproducible in current builds: loading the above URL will never
stop for linux/win32/mac; however, if you disable Keep-Alive, the win32/linux
will fire onload (but mac will spin whether Keep-Alive is on or off).

I poked at the server with libwww-perl, and the first socket read gets only
155 bytes, which is just the headers. The second read is only 48 bytes. 
Subsequent reads are all 'short' (ask for 4096 but get ~1400 per read). These 
are the raw headers:

Server: Netscape-Enterprise/3.0C
Date: Sun, 18 Mar 2001 05:24:54 GMT
Content-type: text/plain; charset=iso-8859-2
Connection: close

Comment 5

[neeti]

*** Bug 70531 has been marked as a duplicate of this bug. ***

Comment 6

[Mike Pinkerton (not reading bugmail)]

since keepalive is on by default, i'd say this should be fixed before beta1

Comment 7

[Bob McElrath]

Recent builds only grab one file per HTTP/1.1 connection (which probably stops
the throbber problem, but is not HTTP/1.1 compliant).  That is, mozilla forcibly
closes the connection after each file despite being sent a 'Connection:
keep-alive' header.

I'm seeing this in the context of a HTTP/1.1 proxy.  Let me know if filing a
separate bug would be appropriate, I'm not sure this is related.

Comment 8

[Darin Fisher]

with keep-alives enabled, mozilla receives the following headers:

HTTP/1.1 200 OK
Server: Netscape-Enterprise/3.0C
Date: Tue, 27 Mar 2001 19:13:16 GMT
Content-type: text/plain; charset=iso-8859-2
Connection: keep-alive

The server provides no indication of when the connection should close.  This
seems like a misconfigured server to me.  It needs to either send a "Connection:
close" header or a "Content-length" header or use chunked transfer encoding.
Otherwise, the browser will have no way of knowing when to close the connection.

Marking INVALID.  Please fix your CGI script.

Comment 9

[John Morrison]

Well, duh. I didn't send a Keep-Alive token when I probed earlier, so yeah
I get a 'Connection: close'. 

Yeah, the server is not doing the right thing. However, aren't we supposed
to shut down the socket if the server end disconnects (assuming the server
is disconnecting in this case).

Comment 10

[Darin Fisher]

john: the server sent "Connection: keep-alive" which means that it is not
going to close its end right away.

Comment 11

[John Morrison]

Yes. But what do we do when the server shuts down its connection even though
it has sent Keep-Alive. That is a bogus, brain-dead thing for the server to
do, but the web has no shortage of misconfigured servers and flaky connections. 
If the socket is closed by the server, shouldn't we bail out?

Comment 12

[Darin Fisher]

we bail whenever the server closes its end of the connection.  in this case,
the server does not close its connection right away, b/c it is configured (via
keep-alive) to wait for other http requests.  we don't send any more requests,
b/c we don't know that we're done processing the current request.  hence, the
page load does not complete until the server connection times out.

Comment 13

[John Morrison]

Ah, thanks. (Light bulb comes on; I was assuming that the server was shutting 
down the connection, despite the Keep-Alive, but that's not the case).

Comment 14

[Tom Everingham]

verified