I'm debugging the 'MySQL server has gone away' errors. There is a proposed solution, which more or less works:
from twisted.enterprise import adbapi
from twisted.python import log
import MySQLdb
class ReconnectingConnectionPool(adbapi.ConnectionPool):
"""Reconnecting adbapi connection pool for MySQL.
This class improves on the solution posted at
http://www.gelens.org/2008/09/12/reinitializing-twisted-connectionpool/
by checking exceptions by error code and only disconnecting the current
connection instead of all of them.
Also see:
http://twistedmatrix.com/pipermail/twisted-python/2009-July/020007.html
"""
def _runInteraction(self, interaction, *args, **kw):
try:
return adbapi.ConnectionPool._runInteraction(self, interaction, *args, **kw)
except MySQLdb.OperationalError, e:
if e[0] not in (2006, 2013):
raise
log.msg("RCP: got error %s, retrying operation" %(e))
conn = self.connections.get(self.threadID())
self.disconnect(conn)
# try the interaction again
return adbapi.ConnectionPool._runInteraction(self, interaction, *args, **kw)
Taken from here: http://www.gelens.org/2009/09/13/twisted-connectionpool-revisited/
However, the exception gets caught no only in our ReconnectingConnectionPool, but in _runInteraction itself too. And it triggers log.err, which (apparently) does two things:
I could hack adbapi, but that's probably not the best idea. Are there any better ways to handle this?
Are you asking how to make a unit test pass even if an error is logged? Try flushLoggedErrors. If you're asking something else, please clarify.