[squid-users] caching and changing content

[Amos Jeffries]

On 12/09/19 8:25 pm, fansari wrote:
> In my scenrio (squid 3.5.23) I have several clients which download content.
> 
> Now I want to achieve two things: if one client has already downloaded the
> content the second client requesting the same content should take it from
> the squid cache.
> 
> But: when this content resource on the internet is changing the squid should
> not give a HIT but a MISS and refresh the content the next time a client is
> requesting it.

No. HTTP specifies that the cache should update its copy of the content,
and deliver the new result. This is *not* a MISS. It is a REFRESH in the
logs, or a "Near-HIT" in the traffic statistics report.

> 
> It is not clear to me how to achieve this behaviour.

You do not have to do anything. For content which this behaviour is
possible, Squid does it by default.

> 
> The application (server on internet) sends headers like this:
> 
> cache-control: public, max-age=0

'public' - this object is allowed to be stored in caches.

'max-age=0' - the cache is required to contact the server for updates
before delivering this object to any clients.


> last-modified: Thu, 12 Sep 2019 07:30:56 GMT
> etag: W/"6699e3-16d2461dba1"
> 
> So far I could only achieve HITs using refresh_pattern with min value > 0.
> 
> refresh_pattern .               30       20%     1440 override-expire
> 

refresh_pattern provides default values for Squid to use in its caching
algorithm if the server does not provide them.

The headers you show above contain clear and precise cache-control
instructions. So refresh_pattern is not normally relevant to that object.

By adding the override-expires option you are *forcing* the object to be
cached for no less than 30 minutes and no more than 1 week...


> But also this configuration is not exactly what  I want. 
> 
> Now even if the content of a resource on the internet is changing I get HITs
> for 30 min.
> 

Not just that resource. By using the '.' pattern you are forcing this
30min HIT on *every* object any client fetches through this cache.

It is best to be a specific as you can in the pattern field. Especially
when using the ignore-* and override-* options. To reduce nasty side
effects.


PS. The server is delivering cache controls, what makes you think the
server developer and admin are wrong about their own content and systems?


Amos