[Zope] Zope hiccuping

Chris Kratz chris.kratz@vistashare.com
Wed, 7 Nov 2001 14:50:10 -0500


Looks like I posted too quickly...

After posting this, I ran into some messages posted on the mailing list that
seem to say that py_malloc is the culprit for this particular problem.  I'm
in the process of trying this solution.  My apologies for the spam.  On the
other hand if there are any other thoughts, I would appreciate hearing them
as well.

-Chris
------------------------------
Chris Kratz
chris.kratz@vistashare.com


----- Original Message -----
From: "Chris Kratz" <chris.kratz@vistashare.com>
To: <zope@zope.org>
Sent: Wednesday, November 07, 2001 2:33 PM
Subject: [Zope] Zope hiccuping


> We have been noticing that periodically, we get an error from IE that says
> "Cannot find server or DNS error".  It is not easily reproducable (except
by
> just clicking links on the server) and a F5 refresh in the browser
[almost]
> always loads the page correctly.  I turned on logging today with the -M
> startup option and observed the following entries when it happened:
>
> B 145697132 2001-11-07T17:15:38 GET /OutcomeTracker/Dev_News
> I 145697132 2001-11-07T17:15:38 0
> A 145697132 2001-11-07T17:15:39 200 32155
> E 145697132 2001-11-07T17:15:39
> B 145934020 2001-11-07T17:15:40 GET
> /OutcomeTracker/PeopleOrganizations/index_html
> I 145934020 2001-11-07T17:15:40 0
> B 135053764 2001-11-07T17:15:56 POST /OutcomeTracker/Activities/index_html
> I 135053764 2001-11-07T17:15:56 2831
> B 146464388 2001-11-07T17:15:57 GET
> /OutcomeTracker/PeopleOrganizations/index_html
> I 146464388 2001-11-07T17:15:57 0
> A 146464388 2001-11-07T17:16:03 200 31200
> E 146464388 2001-11-07T17:16:03
>
> Notice how the Get /OutcomeTracker/PeopleOrganizations/index_html never
gets
> the A or E lines, but only has a B and I line.  The subsequent refresh
> finished the request.  Interestingly, the two incompleted requests are not
> logged to z2.log.  We can see the request before and the request after,
but
> that's it.  The other strangeness is that in the postgres log, we see a
> "pq_recvbuf: unexpected EOF on client connection".  This seemed to point
to
> zope threads dying.  Since I'm not getting anything in the logs(*see
below),
> I started running tests with one eye on the currently running processes.
> And sure enough, whenever I got that error at the browser (cannot find
> server...), *All* of the zope threads (except the main starter thread) die
> quietly and come back with new PIDs.  It really appears like it reruns the
> entire startup sequence again.  With Z_DEBUG_MODE on I can watch it go
> through the startup sequence again whenever this happens.  But, there are
no
> tracebacks.  It's just like somebody clicked restart in the middle of a
> process.
>
> The one glimmer of hope is in the stupid log file:
>
> 2001-11-07T19:30:23 ERROR(200) zdaemon zdaemon: Wed Nov  7 14:30:23 2001:
> Aiieee! 1925 exited with error code: 11
> ...restarting...
>
> Here's the questions,
>
> 1. It appears that something is causing those threads to crash (or end),
but
> nothing is getting put in the log file.  Is there any way to get the
> tracebacks I assume are happening or to find out what is going on?
> 2. Alternatively, is there a way to run zope in single threaded mode?
> Z_DEBUG_MODE appears to only apply to the main thread because it goes
ahead
> and spawns additional threads.  If I use -t 0 I get two processes, but no
> response from a web browser request.  If I use -t 1, I get three processes
> owned by nobody and the original one by root.
> 3. Any further ideas on how to debug this thing?  Where do I find what
error
> code 11 is?
>
> Thanks for you time and help,
>
> -Chris
>
> ------------------------------
> Chris Kratz
> chris.kratz@vistashare.com
>
>
>
> _______________________________________________
> Zope maillist  -  Zope@zope.org
> http://lists.zope.org/mailman/listinfo/zope
> **   No cross posts or HTML encoding!  **
> (Related lists -
>  http://lists.zope.org/mailman/listinfo/zope-announce
>  http://lists.zope.org/mailman/listinfo/zope-dev )
>