Understanding persistent storage

[Yang Zhang]

On Wed, Apr 20, 2011 at 10:21 AM, Per Buer <perbu at varnish-software.com> wrote:
> On Tue, Apr 19, 2011 at 6:44 PM, Yang Zhang <yanghatespam at gmail.com> wrote:
>>
>> What I found last time was:
>>
>> (1) Fetch a page. Miss first time, hit subsequent times.
>>
>> (2) Restart Varnish.
>>
>> (3) Fetch same page. Miss.
>
> Right. This is actually valid behavior. The goal of the persistent
> storage isn't to salvage _all_ of the data across a restart, only to
> salvage _most_ of the data. Salvaging all of the data would require
> database type storage semantics which is more or less impossible to
> achieve without scarifying a lot of persistence.
>
> As -spersistent is implemented, the storage silo which is open at the
> time of the crash/restart is discarded completely, so most of the
> objects inserted into the cache the last few seconds before the crash
> will  certainly be lost. Remember, we're a cache, not a data store.

Those are the semantics I'm interested in.  That's perfect, thanks for
your explanation - sounds like -spersistence may work for us.

-- 
Yang Zhang
http://yz.mit.edu/