Is auto stop/restart of site still needed to prevent memory build up

Some simple restart scripts we used at UW Oshkosh would have the load balancer (pound) disable a pound client, wait longer than the squid timeout period, restart the ZEO client, wait a while, prime the ZEO client by loading some sites that it served, then reenable it as an active pound client.

good news from CloudFlare: no more single point of failures with the new Traffic Manager feature.

I think quite a lot of the memory snowballing mythos comes from the days of Python 2.4 and Plone 3.x prior to this:

https://mail.python.org/pipermail/patches/2005-May/017592.html

Quite a chunk of the memory usage variation can also be, as already expressed in this thread, attributed to hugely varying sizes of ZODB objects and the object count based caching strategy.

I would be quite interested in how to get the byte size defined ZODB cache into a better state, but first, did this change any of the assumptions or beliefs people hold on that not working as one might expect: