Webware for Python

Sorry--I don't know where the subject line on the last message went.
 
One more data point--there's an </table> after the last uid line in the
output I get, but that's it.  No additional information, no </body></html>,
etc.
 
Cheers! 
-- 
David Hancock | dhancock@... | 410-266-4384 

-----Original Message-----
From: Hancock, David (DHANCOCK) [mailto:DHANCOCK@...] 
Sent: Friday, April 26, 2002 2:23 PM
To: 'Geoffrey Talvola'; 'webware-discuss@...'
Subject: [Webware-discuss] (no subject)


I first noticed this problem using IE-on-Windows.  I also see the problem in
several other client situations (these are all on Linux):
 
1. Using a Python script with urllib.py
2. Using wget
3. Using Netscape on Linux
 
Number 3 gives the most status information:  The error page displays, right
down to the uid/gid/etc. information, and the status bar states "12K read
(stalled)" and the "wait cursor" (the little wristwatch) displays until I
press Stop, at which point the status bar states "Document done."  At the
end of the page, then, is a horizontal rule, followed by "Transfer
interrupted!" in bold.
 
Now that I see what Netscape on Linux shows, there may be a clue.  Is there
supposed to be more error information AFTER the list of uid/gid/pid, etc.
The last think I see is 'getuid  500' (500 is the nobody user on my box).
If there's more information being collected, perhaps that's where the hang
is.

Cheers! 
-- 
David Hancock | dhancock@... | 410-266-4384 

-----Original Message-----
From: Geoffrey Talvola [mailto:gtalvola@...] 
Sent: Friday, April 26, 2002 1:02 PM
To: 'Hancock, David (DHANCOCK)'; 'webware-discuss@...'
Subject: RE: [Webware-discuss] WebKit Error traceback pages don't close th
eir connection?


OK.  It doesn't happen here at work on Windows boxes, but I can try it this
weekend on my linux box at home.
 
Have you by any chance tried other clients beside IE-on-Windows?  I'll only
be able to test it with Linux browsers at home.
 
- Geoff 

-----Original Message-----
From: Hancock, David (DHANCOCK) [mailto:DHANCOCK@...]
Sent: Friday, April 26, 2002 10:28 AM
To: Geoffrey Talvola; 'webware-discuss@...'
Subject: RE: [Webware-discuss] WebKit Error traceback pages don't close th
eir connection?


Oops--sorry for incomplete information earlier.
 
OS: Linux (RedHat 6. b 2 distribution)
mod_webkit adapter
Python 2.2
Webware 0.7 (released, nothing newer from CVS)
NOT configured to email error reports
 
Cheers! 
-- 
David Hancock | dhancock@... | 410-266-4384 

-----Original Message-----
From: Geoffrey Talvola [mailto:gtalvola@...] 
Sent: Friday, April 26, 2002 10:19 AM
To: 'Hancock, David (DHANCOCK)'; 'webware-discuss@...'
Subject: RE: [Webware-discuss] WebKit Error traceback pages don't close th
eir connection?


Which OS, adapter, Python version, and Webware version?  It it configured to
send email or not?
 
It doesn't happen to me, and like you, I get a lot of error pages :-)

- Geoff 

-----Original Message-----
From: Hancock, David (DHANCOCK) [mailto:DHANCOCK@...]
Sent: Thursday, April 25, 2002 10:56 PM
To: 'webware-discuss@...'
Subject: [Webware-discuss] WebKit Error traceback pages don't close their
connection?



When I get an error traceback page (it happens a lot, I make a lot of
mistakes), it seems like the HTTP connection doesn't get closed.  The
symptoms I see are that the traceback page loads, down to the end (I think),
where the IDs print, but the IE "globe" keeps spinning until I press the
Stop button on the browser.

When I request the /Examples/Error servlet via urllib in a Python script, it
hangs at the point of retrieving the page. If I press Ctrl-C, there's
nothing in the output file.

Any words of advice about what's going on and how I could fix it?  I'm
trying to write a cheesy automatic test for a servlet page that will simply
look for the presence of the traceback.  What I'm doing as a workaround is
assuming that no page will take more than 30 seconds to load (browser and
servlet on the same network), and handling an alarm if the timeout is
exceeded.

I appreciate any advice or pointers to information you can provide. 

Thanks.  By the way, the fancy traceback is a great boon for us--much easier
than finding and tailing a logfile. 

Cheers! 
-- 
David Hancock | dhancock@... | 410-266-4384

FancyTracebacks is set to 0, FancyTraceBackContext is set to 5.  I set
FancyTracebacks to 1, restarted the application server, and got similar
behavior, except that the page hangs after the following in the output:
 
ImportError: No module named UnknownModule
      args = ('No module named UnknownModule',)
 
I am seeing the behavior with the included Examples/Error servlet.  One
installation that demonstrates the problem can be seen at:
 
    http://dhancock.2y.net/ADC/Examples/Error
<http://dhancock.2y.net/ADC/Examples/Error> 
 
And one more data point--while the above is a machine at home (I'm taking
the day off!), but I see the same behavior on several Linux machines at work
(similar configurations, just WAY faster than my home box).
 
I appreciate all the attention to this issue--I hope we can find something
and fix it.

Cheers! 
-- 
David Hancock | dhancock@... | 410-266-4384 


-----Original Message-----
From: Geoffrey Talvola [mailto:gtalvola@...] 
Sent: Friday, April 26, 2002 2:42 PM
To: 'Hancock, David (DHANCOCK)'; 'webware-discuss@...'
Subject: RE: 


The Ids section is the last part of the exception report unless you have
IncludeFancyTracebacks set to true, in which case a section with the title
"Fancy Traceback" comes after it.  Do you have IncludeFancyTracebacks set to
true, and if so, are you in fact seing the "Fancy Traceback" section?  Maybe
it's hanging in the code that generates the fancy traceback.
 
Also, does this fail with the standard Examples/Error servlet or only with a
servlet you wrote yourself?

- Geoff 

-----Original Message-----
From: Hancock, David (DHANCOCK) [mailto:DHANCOCK@...] 
Sent: Friday, April 26, 2002 2:33 PM
To: 'Geoffrey Talvola'; 'webware-discuss@...'
Subject: RE: [Webware-discuss] WebKit Error traceback pages don't close th
eir connection?


Sorry--I don't know where the subject line on the last message went.
 
One more data point--there's an </table> after the last uid line in the
output I get, but that's it.  No additional information, no </body></html>,
etc.
 
Cheers! 
-- 
David Hancock | dhancock@... | 410-266-4384 

-----Original Message-----
From: Hancock, David (DHANCOCK) [mailto:DHANCOCK@...] 
Sent: Friday, April 26, 2002 2:23 PM
To: 'Geoffrey Talvola'; 'webware-discuss@...'
Subject: [Webware-discuss] (no subject)


I first noticed this problem using IE-on-Windows.  I also see the problem in
several other client situations (these are all on Linux):
 
1. Using a Python script with urllib.py
2. Using wget
3. Using Netscape on Linux
 
Number 3 gives the most status information:  The error page displays, right
down to the uid/gid/etc. information, and the status bar states "12K read
(stalled)" and the "wait cursor" (the little wristwatch) displays until I
press Stop, at which point the status bar states "Document done."  At the
end of the page, then, is a horizontal rule, followed by "Transfer
interrupted!" in bold.
 
Now that I see what Netscape on Linux shows, there may be a clue.  Is there
supposed to be more error information AFTER the list of uid/gid/pid, etc.
The last think I see is 'getuid  500' (500 is the nobody user on my box).
If there's more information being collected, perhaps that's where the hang
is.

Cheers! 
-- 
David Hancock | dhancock@... | 410-266-4384 

-----Original Message-----
From: Geoffrey Talvola [mailto:gtalvola@...] 
Sent: Friday, April 26, 2002 1:02 PM
To: 'Hancock, David (DHANCOCK)'; 'webware-discuss@...'
Subject: RE: [Webware-discuss] WebKit Error traceback pages don't close th
eir connection?


OK.  It doesn't happen here at work on Windows boxes, but I can try it this
weekend on my linux box at home.
 
Have you by any chance tried other clients beside IE-on-Windows?  I'll only
be able to test it with Linux browsers at home.
 
- Geoff 

-----Original Message-----
From: Hancock, David (DHANCOCK) [mailto:DHANCOCK@...]
Sent: Friday, April 26, 2002 10:28 AM
To: Geoffrey Talvola; 'webware-discuss@...'
Subject: RE: [Webware-discuss] WebKit Error traceback pages don't close th
eir connection?


Oops--sorry for incomplete information earlier.
 
OS: Linux (RedHat 6. b 2 distribution)
mod_webkit adapter
Python 2.2
Webware 0.7 (released, nothing newer from CVS)
NOT configured to email error reports
 
Cheers! 
-- 
David Hancock | dhancock@... | 410-266-4384 

-----Original Message-----
From: Geoffrey Talvola [mailto:gtalvola@...] 
Sent: Friday, April 26, 2002 10:19 AM
To: 'Hancock, David (DHANCOCK)'; 'webware-discuss@...'
Subject: RE: [Webware-discuss] WebKit Error traceback pages don't close th
eir connection?


Which OS, adapter, Python version, and Webware version?  It it configured to
send email or not?
 
It doesn't happen to me, and like you, I get a lot of error pages :-)

- Geoff 

-----Original Message-----
From: Hancock, David (DHANCOCK) [mailto:DHANCOCK@...]
Sent: Thursday, April 25, 2002 10:56 PM
To: 'webware-discuss@...'
Subject: [Webware-discuss] WebKit Error traceback pages don't close their
connection?



When I get an error traceback page (it happens a lot, I make a lot of
mistakes), it seems like the HTTP connection doesn't get closed.  The
symptoms I see are that the traceback page loads, down to the end (I think),
where the IDs print, but the IE "globe" keeps spinning until I press the
Stop button on the browser.

When I request the /Examples/Error servlet via urllib in a Python script, it
hangs at the point of retrieving the page. If I press Ctrl-C, there's
nothing in the output file.

Any words of advice about what's going on and how I could fix it?  I'm
trying to write a cheesy automatic test for a servlet page that will simply
look for the presence of the traceback.  What I'm doing as a workaround is
assuming that no page will take more than 30 seconds to load (browser and
servlet on the same network), and handling an alarm if the timeout is
exceeded.

I appreciate any advice or pointers to information you can provide. 

Thanks.  By the way, the fancy traceback is a great boon for us--much easier
than finding and tailing a logfile. 

Cheers! 
-- 
David Hancock | dhancock@... | 410-266-4384

On Friday April 26, 2002 02:52 pm, Hancock, David (DHANCOCK) wrote:
> I am seeing the behavior with the included Examples/Error servlet.  One
> installation that demonstrates the problem can be seen at:
>
>     http://dhancock.2y.net/ADC/Examples/Error
> <http://dhancock.2y.net/ADC/Examples/Error>

I just visited this link - and it does wedge.

> And one more data point--while the above is a machine at home (I'm taking
> the day off!), but I see the same behavior on several Linux machines at
> work (similar configurations, just WAY faster than my home box).
>
> I appreciate all the attention to this issue--I hope we can find something
> and fix it.
>
> Cheers!

I can't reproduce the problem on my Linux box at home.  I'm running Mandrake 
8.2 with a mod_webkit built from CVS on March 24, and I'm using Webware from 
the tip of CVS.  And Python 2.2.  The error page comes out just fine and the 
connection closes -- no wedge.

At this point I would recommend:

1) Try changing to the CGI adapter and see if that helps.

2) Throw in a bunch of print statements into the WebKit code to see if you 
can find a point where it's getting wedged.

3) Try updating Apache if you're using a version significantly older than 
1.3.23 which is what I'm using.

4) Recompile mod_webkit from the head of CVS.

5) Find out from other people on the list whether they have the same problem.

- Geoff

Geoff and list: I had to do some other work on Monday and Tuesday.  Here's
the state of my investigations (and I appreciate knowing if others have seen
this problem):

1. Used the wkcgi adapter, and the page still hangs.

2. I haven't put prints into the WebKit code yet; I'm not familiar enough
with it to know where to start.

3. I'm running Apache 1.3.23, same as you are.  I unloaded the mod_python
DSO, but still get the same results.

4. I recompiled and reinstalled mod_webkit from the head of CVS; same
results as before.

I see this behavior on two different machines on two completely different
networks.

If others see this behavior, please weigh in.

And if I could get one or two pointers to where to start putting print
statements to track down the point of hang, I'd be grateful.  I'm a
procedural sort of person trying to get by in an object-oriented world :-)

Cheers!
--
David Hancock | dhancock@... | 410-266-4384


-----Original Message-----
From: Geoffrey Talvola [mailto:gtalvola@...] 
Sent: Sunday, April 28, 2002 9:34 PM
To: Hancock, David (DHANCOCK); 'Geoffrey Talvola';
'webware-discuss@...'
Subject: Re: [Webware-discuss] WebKit Error traceback pages don't close th
eir connection?


On Friday April 26, 2002 02:52 pm, Hancock, David (DHANCOCK) wrote:
> I am seeing the behavior with the included Examples/Error servlet.  
> One installation that demonstrates the problem can be seen at:
>
>     http://dhancock.2y.net/ADC/Examples/Error
> <http://dhancock.2y.net/ADC/Examples/Error>

I just visited this link - and it does wedge.

> And one more data point--while the above is a machine at home (I'm 
> taking the day off!), but I see the same behavior on several Linux 
> machines at work (similar configurations, just WAY faster than my home 
> box).
>
> I appreciate all the attention to this issue--I hope we can find 
> something and fix it.
>
> Cheers!

I can't reproduce the problem on my Linux box at home.  I'm running Mandrake

8.2 with a mod_webkit built from CVS on March 24, and I'm using Webware from

the tip of CVS.  And Python 2.2.  The error page comes out just fine and the

connection closes -- no wedge.

At this point I would recommend:

1) Try changing to the CGI adapter and see if that helps.

2) Throw in a bunch of print statements into the WebKit code to see if you 
can find a point where it's getting wedged.

3) Try updating Apache if you're using a version significantly older than 
1.3.23 which is what I'm using.

4) Recompile mod_webkit from the head of CVS.

5) Find out from other people on the list whether they have the same
problem.

- Geoff

Hancock, David (DHANCOCK) wrote:
> And if I could get one or two pointers to where to start putting print
> statements to track down the point of hang, I'd be grateful.  I'm a
> procedural sort of person trying to get by in an object-oriented world :-)


Try:

http://www.cyberclip.com/webdebug/

It's easy to setup, you start it running
and tell your browser to proxy through
port 2001. Then you should get all the
printed data on the webdebug admin
page to give you an idea what the
last character coming out of WebWare
is.

One caveat is that webdebug proxies
to port 80, and there might be an issue
with the redirect to the WebKit port.

-- 
Bill Eldridge
Radio Free Asia
bill@...

Bill Eldridge suggested using WebClip to see where in the output stream the
hang occurs, and that actually helped me find two things:

1. Without the WebClip webdebug proxy, the hang is right BEFORE
</body></html> at the end of the page. This is reliably true with or without
FancyTraceback turned on.  It turns out that patience (like several minutes
of patience) results in the trailing close tags coming across the wire, and
the page is complete.  (That's a bit too long to wait.)

2. Using the webdebug proxy, the page hangs and loads nothing (only the
request shows up in webdebug's menus) until about 300 seconds later, then
the whole page is loaded.

What I think this says is that the page is nearly completed and sent, and
only the last few bytes are "hanging" out there.  When (supposition only)
some sort of timeout occurs, the rest of the bytes are sent out.

I'd tried webdebug at Bill's suggestion the other day, but stupidly tried to
proxy through 2021, not 2001, the default port.

Here's a netstat on the server machine "steady state" followed by while the
error page is wedged:

[dhancock@... dhancock]$ netstat -an | grep 9034
tcp        1   8640 127.0.0.1:9034          127.0.0.1:3483          CLOSE
tcp        0      0 127.0.0.1:9034          0.0.0.0:*               LISTEN
[dhancock@... dhancock]$ netstat -an | grep 9034
tcp        1      0 127.0.0.1:9034          127.0.0.1:3491
CLOSE_WAIT
tcp        0      0 127.0.0.1:3491          127.0.0.1:9034
FIN_WAIT2
tcp        0      0 127.0.0.1:3490          127.0.0.1:9034
TIME_WAIT
tcp        0      0 127.0.0.1:3489          127.0.0.1:9034
TIME_WAIT
tcp        0      0 127.0.0.1:3488          127.0.0.1:9034
TIME_WAIT
tcp        0      0 127.0.0.1:9034          0.0.0.0:*               LISTEN
[dhancock@... dhancock]$

I can't profess to understanding these states, but maybe this will ring a
bell with somebody.

Once the five minutes elapses, netstat says:

[dhancock@... dhancock]$ netstat -an | grep 9034
tcp        1      0 127.0.0.1:9034          127.0.0.1:3491
CLOSE_WAIT
tcp        0      0 127.0.0.1:3491          127.0.0.1:9034
FIN_WAIT2
tcp        0      0 127.0.0.1:9034          0.0.0.0:*               LISTEN
[dhancock@... dhancock]$

Cheers!
--
David Hancock | dhancock@... | 410-266-4384


-----Original Message-----
From: Hancock, David (DHANCOCK) [mailto:DHANCOCK@...] 
Sent: Wednesday, May 01, 2002 1:40 PM
To: 'Geoffrey Talvola'; 'webware-discuss@...'
Subject: RE: [Webware-discuss] WebKit Error traceback pages don't close th
eir connection?


Geoff and list: I had to do some other work on Monday and Tuesday.  Here's
the state of my investigations (and I appreciate knowing if others have seen
this problem):

1. Used the wkcgi adapter, and the page still hangs.

2. I haven't put prints into the WebKit code yet; I'm not familiar enough
with it to know where to start.

3. I'm running Apache 1.3.23, same as you are.  I unloaded the mod_python
DSO, but still get the same results.

4. I recompiled and reinstalled mod_webkit from the head of CVS; same
results as before.

I see this behavior on two different machines on two completely different
networks.

If others see this behavior, please weigh in.

And if I could get one or two pointers to where to start putting print
statements to track down the point of hang, I'd be grateful.  I'm a
procedural sort of person trying to get by in an object-oriented world :-)

Cheers!
--
David Hancock | dhancock@... | 410-266-4384


-----Original Message-----
From: Geoffrey Talvola [mailto:gtalvola@...] 
Sent: Sunday, April 28, 2002 9:34 PM
To: Hancock, David (DHANCOCK); 'Geoffrey Talvola';
'webware-discuss@...'
Subject: Re: [Webware-discuss] WebKit Error traceback pages don't close th
eir connection?


On Friday April 26, 2002 02:52 pm, Hancock, David (DHANCOCK) wrote:
> I am seeing the behavior with the included Examples/Error servlet.
> One installation that demonstrates the problem can be seen at:
>
>     http://dhancock.2y.net/ADC/Examples/Error
> <http://dhancock.2y.net/ADC/Examples/Error>

I just visited this link - and it does wedge.

> And one more data point--while the above is a machine at home (I'm
> taking the day off!), but I see the same behavior on several Linux 
> machines at work (similar configurations, just WAY faster than my home 
> box).
>
> I appreciate all the attention to this issue--I hope we can find
> something and fix it.
>
> Cheers!

I can't reproduce the problem on my Linux box at home.  I'm running Mandrake

8.2 with a mod_webkit built from CVS on March 24, and I'm using Webware from

the tip of CVS.  And Python 2.2.  The error page comes out just fine and the

connection closes -- no wedge.

At this point I would recommend:

1) Try changing to the CGI adapter and see if that helps.

2) Throw in a bunch of print statements into the WebKit code to see if you 
can find a point where it's getting wedged.

3) Try updating Apache if you're using a version significantly older than 
1.3.23 which is what I'm using.

4) Recompile mod_webkit from the head of CVS.

5) Find out from other people on the list whether they have the same
problem.

- Geoff

Hancock, David (DHANCOCK) wrote:
> Bill Eldridge suggested using WebClip to see where in the output stream the
> hang occurs, and that actually helped me find two things:
> 
> 1. Without the WebClip webdebug proxy, the hang is right BEFORE
> </body></html> at the end of the page. This is reliably true with or without
> FancyTraceback turned on.  It turns out that patience (like several minutes
> of patience) results in the trailing close tags coming across the wire, and
> the page is complete.  (That's a bit too long to wait.)
> 
> 2. Using the webdebug proxy, the page hangs and loads nothing (only the
> request shows up in webdebug's menus) until about 300 seconds later, then
> the whole page is loaded.

Well that sounds like either 1) a module is
caluculating the hell out of something and
only finishes aftter some minutes, or
2) has sent a request for something and
only times out and forgets about it sometime
later.

ExceptionHandler does:
                html.append(self.htmlDebugInfo())

                 html.append('</body></html>')
                 return string.join(html, '')

which ends up calling:

         def writeHTML(self):
                 self.writeTraceback()
                 self.writeMiscInfo()
                 self.writeTransaction()
                 self.writeEnvironment()
                 self.writeIds()
                 self.writeFancyTraceback()

You can try commenting these out one by one
to see which one is hanging you up.

The different OneShot,ModPython,FCGIAdapters
include the </body></html> in with all the other
output, so they shouldn't be guilty.




-- 
Bill Eldridge
Radio Free Asia
bill@...