List caching in the qrunners can cause the processes to grow unduly large.

[Mark Sapiro]

The qrunners implement a cache of list objects to save reloading the list from the file system when not necessary. This cache is implemented via a weakref dictionary which is intended to free cached objects when they are no longer referenced. However, there is a self reference in the default MemberAdaptor (OldStyleMemberships) that prevents cached list objects from ever being freed. Thus, the qrunners grow until they have a cached copy of every list. This can be quite large in an installation with many lists and/or multiple large lists.

[Mark Sapiro (msapiro)]

This is fixed by eliminating the qrunner list cache. It is not clear that this cache offers much benefit with modern file systems.