CrashingZopeWithOracle

John Jarvis reported to the zope-dev list:

Hi all,

I've been having severe problems with Zope stability lately.

I wanted to provide a more detailed report, but I can't seem to find out what Zope is doing when it crashes--it seems completely random. I'd file a bug report, but I'm not sure exactly what to call a bug here or if there really is a bug and not just a bad setup on my part.

Well, here goes:

• Zope drops connections to Oracle DB

  Based on what I saw with Zope 1, I thought that DB connections were persistent. Since moving to Zope 2, I've noticed that quite often the DB connections are lost. After watching the v$session table for a while, I noticed that when running a number of queries, Zope opens a number of connections equal to the number of Zope threads. When the traffic stops for a time (not long), these connections all disappear. Is this normal?

  Often it seems that Zope can't reopen the connection and crashes or gives an error: "xxxx not connected to a database."

  I thought maybe it was a setting on the DB side, but idle time, etc. is set for infinity for the Zope users. Also, this happens quite frequently under heavy use.

• Zope crashes when using the Open/Close Database Connection button.

  Related to the above? Or is the above related to this?

  Most often crashes after editing the connection properties or getting the "xxxx not connected to a database." error, noticing that the connection is marked as "Not Connected", and then clicking Open Database Connection.

• Zope crashes after heavy continued useage

  Feeling adventurous, I wrote a script that uses ZClient? to request a query report page using random id's as search parameters. The script looks something like this:

      #--------begin script
      from ZPublisher import Client
      from time import time, ctime
      from random import random
      i = 0     # number or passes
      while 1:
          i = i + 1
          id = int(random()*1757) + 20000000
          print "[" + ctime(time()) + "]", "Pass:", i, "No:", id
          Client.call(http://appserver:8080/action/list_mon_detail?
                      'call_no='+id, anonymous, '')
      #---------end script
  
  

  With Zope set for four threads and five instances of the script running on a different machine, Zope will silently crash within a half hour. ZDaemon? restarts the process if it can, but no errors are reported (even in debug mode). I've also noticed that memory usage increases constantly until the crash. I still have gobs left, though.

  I've noticed no difference in running through Apache/PCGI or Medusa. A typical traceback looks something like this:

      #----------begin traceback
      Traceback (innermost last):
        File "C:\PROGRA~2\hammers\HAMMER~1.PY", line 21, in ?
         Client.call('http://myserver:8080/action/list_mon_detail?call_no=%i\
                &lng_id=0&dsp_menubtn=on'%call_no, anonymous,'')
        File "C:\PROGRA~1\GEORGE\lib\python\ZPublisher\Client.py", line 344,\
                in call
         return apply(Function(url,username=username, password=password),\
                (), kw)
        File "C:\PROGRA~1\GEORGE\lib\python\ZPublisher\Client.py", line 225,\
                in __call__
         self.handleError(query, ec, em, headers, response)
        File "C:\PROGRA~1\GEORGE\lib\python\ZPublisher\Client.py", line 240,\
                in handleError
         raise t, RemoteException(t,v,f,l,self.url,query,ec,em,response)
        bci.ServerError: -1 (File: Unknown Line: Unknown)
        -1  for http://myserver:8080/action/list_mon_detail?call_no=20000695\
                &lng_id=0&dsp_menubtn=on
  
      #----------end traceback
  
  

  My system:

      Zope test server
        OS: Linux 2.2.14, glibc 2.0
        CPU: Celeron 466MHz
        Memory: 128MB
        Disk Space: Gobs and Gobs
        Zope Version: 2.1.6 (built from source dist.)
        ZOracleDA Version: 2.2 (problem exists w/2.1, too)
        Connection Method: PCGI, Medusa
        Oracle Client: Net8 (from 8.0.5 dist.)
        Lan: 100Mbps
      Zope client:
        OS: Windows NT
        Client: ZPublisher (from binary dist)
        Browser: Netscape 4.61 (others use IE4.0)
        Lan: 10Mbps
      Oracle server:
        OS: HPUX
        Hardware: Bigger than my refrigerator * 2
        Oracle Version: 7.3.1
        Lan: 100Mbps (on same segment as Zope server)
  
  

  I really wish I had more details. I'll send an update if I find anything.