Comment 3 for bug 595378

The only way for this to happen that I can think of is that a
transaction is aborted after the commit process has started. I don't
know how this would happen in the wild.

In 3.10.3, I'm changing the ZEO tpc_abort call to the server to be
synchronous, which should address this case. Maybe there's another
case. Please report whether this goes way after upgrading to 3.10.3 or
higher.