When Zope connects to ZEO it is configured with a database cache, where objects are cached by Zope so that it doesn’t have to constantly return to ZEO. These caches can be transient (they don’t persist past a Zope restart) or persistent (where they do).
If all you do is uncomment the example ZEO client configuration in your zope.conf you’ll have a 20MB transient cache. If you follow one of the many buildout recipes on the web you’ll likely have a transient cache of anywhere from 30MB (plone.recipe.zope2instance) up to 300MB (from this howto on plone.org).
Zope uses the Python tempfile module to decide where to put these transient caches, which on Debian defaults to /tmp. Our default production server installation has /tmp as a subdirectory of /, with the root partition being only 1GB. This means we have a maximum of about 400MB for temporary files on our servers.
With the default 20MB or 30MB this is fine, even in multiple Zope/ZEO installations on the same box. However, when we recently tried to up the size of the caches on a system that needed it we quickly found ourselves running out of diskspace in the root partition, which in turn caused some very strange behaviour, ultimately resulting in the site being unavailable.
The trick is to move these transient files to a more sensible location (like /var/tmp). You can do this (with thanks to mauro on zodb-dev) by setting an environment variable in your zope.conf, either manually:
<environment> TMPDIR /var/tmp </environment>
Or in your buildout:
[instance] recipe = plone.recipe.zope2install ... environment-vars = TMPDIR /var/tmp