[Zope3-Users] recovering from a lost db connection

[Roy Mathew]

Brian Sutherland writes:
 > On Wed, Jan 10, 2007 at 08:13:27PM -0500, Roy Mathew wrote:
 > > What is the recommended way to recover from a db connection that is
 > > held by a Psycopgda adapter instance? If for some reason postgres
 > > goes down, I get the following error:
 > > 
 > >   File "/var/lib/zope3/instances/instance.barry2007/lib/python/sqlos/adapter.py", line 83, in _runWithConnection
 > >     val = meth(conn, *args)
 > >   File "/usr/lib/python2.4/site-packages/SQLObject-0.8dev_r1814-py2.4.egg/sqlobject/dbconnection.p ", line 351, in _queryOne
 > >     self._executeRetry(conn, c, s)
 > >   File "/var/lib/zope3/instances/instance.barry2007/lib/python/sqlos/adapter.py", line 77, in _executeRetry
 > >     raise DatabaseException(str(exc.args))
 > >   DatabaseException: ('no connection to the server\n',).
 > > 
 > > I then have no recourse but to restart the zope server. In the Java
 > > world one would put a try/catch around the the SQL operation and
 > > attempt a reconnect, but I am wondering if there is a better way
 > > (read: automatic recovery) in a z3 world.
 > 
 > The way ZPsycopgDA does it is exactly that. But it can be a little
 > complex when you look at the detail.
 > 
 > In the Z3 case, I would:
 >     * If the error is a "concurrent update" error, raise a Retry
 >       exception. The same request gets tried again.
 >     * If the error is an OperationalError, close the connection and
 >       re-raise the error (perhaps as a Retry error, though I can think
 >       of cases where this is a bad idea). On the next request, the
 >       connection is re-opened.

Brian, Thanks for your response.

I am confused - that means that if this is being done *currently* in
Psycopgda, I should simply be able to repeat my client request and
succeed in the operation. Or is ZPsycopgda something other than plain
old psycopgda?

I am essentially stuck once I get this error. The stacktrace above
shows that a retry was attempted but did not succeed. Sorry that I am
missing your point, but I don't get it!

 >     * On other errors, let the error propagate.
 > 
 > All this stuff should probably be done at the zope.rdb level.

-- 
Thanks,
Roy Mathew.