Gordon Mohr wrote a while back (2007-09-20):
> When all available cache space is used, and requests for new resources
> arrive, does Varnish discard older objects to make space for new? Is its
> policy for doing so configurable?
>
> Thanks for any details/pointers,
Still wondering if anyone can comment on this question.
Once upon a time I would have assumed the answer was something like "of
course, that's what a cache does", but responses like this one from DES
back in May gave me pause:
http://projects.linpro.no/pipermail/varnish-dev/2007-May/000602.html If I understand the implications of that thread correctly, at the time
of its writing, Varnish would crash rather than evict lesser-accessed
cached data.
In particular, I am considering using Varnish in front of a 'data set'
of > 1PB. Many (or most) of the URLs in the set will never be accessed,
but over the course of hours or days more will be accessed than can be
allocated the cache in RAM or disk. Still, we'd like to cache the 'hot'
items to ease the load on the backend and improve end-user responsiveness.
- Gordon @ IA