Attribute error reported

[tjb@…]

I have just updated our trac to r10006 , sourced from the 0.12-stable branch, and found the following error.

Traceback (most recent call last):
  File "build/bdist.linux-i686/egg/trac/web/api.py", line 436, in send_error
    data, 'text/html')
  File "build/bdist.linux-i686/egg/trac/web/chrome.py", line 803, in render_template
    message = req.session.pop('chrome.%s.%d' % (type_, i))
  File "build/bdist.linux-i686/egg/trac/web/api.py", line 212, in __getattr__
    value = self.callbacks[name](self)
  File "build/bdist.linux-i686/egg/trac/web/main.py", line 298, in _get_session
    return Session(self.env, req)
  File "build/bdist.linux-i686/egg/trac/web/session.py", line 167, in __init__
    self.get_session(req.authname, authenticated=True)
  File "build/bdist.linux-i686/egg/trac/web/session.py", line 183, in get_session
    super(Session, self).get_session(sid, authenticated)
  File "build/bdist.linux-i686/egg/trac/web/session.py", line 57, in get_session
    cursor = db.cursor()
  File "build/bdist.linux-i686/egg/trac/db/util.py", line 102, in __getattr__
    return getattr(self.cnx, name)
AttributeError: 'tuple' object has no attribute 'cursor'

The same error had occurred at r10004, but not previously - I think the previous working version was r9999

I have also recently changed our trac runtime environment to use mod_wsgi instead of the prvious mod_python (apache 2.2). This setup was working prior to updating to trac0.12-stable r10004

Reverting to the prior working version fixes the problem

Our environment is:

• Ubuntu 8.04LTS
• Apache 2.2
• python 2.5
• subversion 1.6.11 (irrelevant I expect)
• mod_wsgi 3.3

[tjb@…]

Correction, the prior working revision was r10001

[rblank]

Unfortunately, this is an error during the reporting of an error, so it's difficult to find the cause from this secondary traceback. Could you please look into your trac.log and see if you can find the primary traceback, that is, the traceback right before this one?

[tjb@…]

Here is the contents of trac.log for the error - sorry for the delay.

2010-08-17 14:38:43,388 Trac[main] ERROR: Internal Server Error: 
Traceback (most recent call last):
  File "build/bdist.linux-i686/egg/trac/web/main.py", line 513, in _dispatch_request
    dispatcher.dispatch(req)
  File "build/bdist.linux-i686/egg/trac/web/main.py", line 235, in dispatch
    resp = chosen_handler.process_request(req)
  File "build/bdist.linux-i686/egg/trac/wiki/web_ui.py", line 117, in process_request
    page = WikiPage(self.env, pagename)
  File "build/bdist.linux-i686/egg/trac/wiki/model.py", line 44, in __init__
    self._fetch(name, version, db)
  File "build/bdist.linux-i686/egg/trac/wiki/model.py", line 55, in _fetch
    db = self.env.get_db_cnx()
  File "build/bdist.linux-i686/egg/trac/env.py", line 326, in get_db_cnx
    return get_read_db(self)
  File "build/bdist.linux-i686/egg/trac/db/api.py", line 90, in get_read_db
    return _transaction_local.db or DatabaseManager(env).get_connection()
  File "build/bdist.linux-i686/egg/trac/db/api.py", line 152, in get_connection
    return self._cnx_pool.get_cnx(self.timeout or None)
  File "build/bdist.linux-i686/egg/trac/db/pool.py", line 220, in get_cnx
    return _backend.get_cnx(self._connector, self._kwargs, timeout)
  File "build/bdist.linux-i686/egg/trac/db/pool.py", line 114, in get_cnx
    cnx.close()
AttributeError: 'NoneType' object has no attribute 'close'

[rblank]

Thanks, that should help.

[osimons]

I see this too on my development setup, and the error is reported consistently for certain requests. Exact same error(s). I just don't know why yet...

Christian, I'm on IRC as usual if you want to help me research this one... :-)

[cboos]

Well, the bug is clear now, but no idea why we didn't detect that mistake earlier when running the patch on t.e.o ;-)

trac/db/pool.py

@@ -158 +158 @@
         # Forth best option: Replace a pooled connection with a new one
         elif len(self._active) < self._maxsize:
             # Remove the LRU connection in the pool
-            self._pool.pop(0)
+            cnx = self._pool.pop(0)
             self._pool_key.pop(0)
             self._pool_time.pop(0)
-            return ('close', None)
+            return ('close', cnx)
 
     def _return_cnx(self, cnx, key, tid):
         # Decrement active refcount, clear slot if 1

Could you please try the above fix?

[cboos]

Replying to cboos:

why we didn't detect that mistake earlier when running the patch on t.e.o ...

... is because this code is only triggered when the same process serves different Trac environments (and different databases even, in case the environments correspond to different schemas), which is not the case on t.e.o which has dedicated processes for each environment.

[osimons]

I see that the type of request that fails for me are the "forge"-type requests where same process:thread connects with more than one project and hence needs to open more more than a single connection (separate databases).

The patch fixes this. Goodie.

[cboos]

Applied in r10039 for 0.12.1 and ported in trunk part of r10041.