Sounds very intelligent, and good to know the cache won't need to be
deleted.

We are using this to cache small images, and we plan on pushing this to
multi-millions of individual files, and about a TB of storage, so I'll
let you know what type of performance we see under those conditions at
startup time.

Thanks,

John




Igor Sysoev wrote:
> On Sat, May 02, 2009 at 11:53:18AM -0500, Resicow wrote:
>
>
>> Hello,
>>
>> I have a question on for nginx handles restarts and proxy_cache.
>>
>> Say I have 2 million objects cached, taking up 50GBs of space. Then I
>> get a kernel panic or some other error, and have to restart the physical
>> machine.
>>
>> When I start nginx up again the 2 million objects are still on the disk,
>> but nginx no longer has its internal caching table, correct? Are those
>> files just stuck on the disk forever, or do I need to ensure the cache
>> is empty if nginx needs to be restarted by manually deleting the objects?
>>
>
> While starting up workers look up disk cache on demand and update keys
> zones in shared memory. Also the cache manager process runs in background,
> reads step by step a whole disk cache, and updates the keys zones, too.
> After cache manager has read (it may take a long time for 50G) a whole
> disk cache, workers use only keys zones to look up existence of responses.
> Upon that the cache manager now delete inactive responses and looks after
> disk cache size.
>
>
>> Also on binary upgrades, will the internal proxy cache table survive?
>>
>
> Since keys zones are not shared between old and new processes, the new
> workers and cache manager does the same as on start up.
>
>
>