[ZODB-Dev] Deadlock in ZEO shutdown protocol

[Dieter Maurer]

On our fastest servers (and only there)
(Dual 2 GHz AMD, 1 GB RAM, Linux, Python 2.1.3, ZODB 3.1 CVS from "Zope_2.6branch")
the tests in "ZEO/tests/testConnection.py" often deadlock.

At the time of the deadlock, the client has sent the "close"
message to the server and waits in "waitpid" for its termination.
The server does not process the close message. Therefore, it
does not feel any need to terminate. Due to a bug in GDB 5.3,
I have been unable to analyse the server state.
Unfortunatly, the problem disappears as soon as logging is enabled.

I worked around this problem by not using the shutdown protocol.
Instead, I send the server a terminate signal which leads to
a reliable shutdown (though without cleanup in "run_server").


Dieter