[ZODB-Dev] Does the ZODB eat memory?

[Tim Peters]

The test program doesn't leak memory under Zope3/ZODB4, so there's reason to
hope Jeremy's backports will fix all the leaks not specific to the ZODB3
cache details.

Under a magnifying glass (watching sys.gettotalrefcount() under a
debug-build Python), the reference behavior in ZODB4 is complex, *appearing*
to leak references at the start (probably due to filling the ZODB4 cache),
but settling down then:  I've been running the program for a few hours now,
and the total-refcount highwater mark was reached over 24,000 open/close
iterations ago.  The peak memory usage hasn't budged since a few minutes
after starting it.