Tag support for Symfony HttpCache

[dantleech]

Part of #26

Any advice on how feasable it would be to implement tags for the HTTP cache?

e.g. I would imagine it would be possible by creating filesystem indexes

app/
    cache/
        http_cache/
            tags/
                articles
                article-42
                article-21
                media
                media-42
                page-32

$ cat app/cache/htp_cache/tags/articles
url/to/purge
url/to/purge
url/to/purge

(p.s. just started to look at this so sorry if I am misunderstanding some things here)

[dantleech]

/cc @chirimoya

[dantleech]

Would this be handled better as a StackPHP middleware?

[dbu]

the structure you are proposing would be the way to look up what cache items to remove when a tag is invalidated? i would see some problems:

1. how do you trim this list when cache items are removed for other reasons?
2. how do you prevent the system to overload on invalidation with a large cache?
3. how to prevent race conditions with editing those files without waiting for locks to be released?

About 2: As explained in this comment varnish just keeps a list of bans that it checks on access to cached content and runs a "ban lurker" that gets active when the system load is not too high to sift old cache entries and get rid of ban rules that are older than all cached content.
I would expect a symfony implementation to need the same solution - unless we assume this to only ever work for a low number of cached pages where an instant-ban is no big effort. that might be a legitimate assumption - a larger system hopefully gets run with a real varnish / nginx anyways.

[dbu]

@lolautruche does ez already have solution / experience with the cache tagging topic?

[dbu]

btw, #168 could be a first step to deepen our knowledge of the symfony built-in reverse proxy, before digging this more complex topic.

[dantleech]

yeah, I am thinking now of something much more like Varnish and basically similar to what @lolautruche has implemented in EZPublish I think.

[lolautruche]

Hi

Actually I'm currently considering the use of a cache library such as Stash. It would be used as a dictionnary, pretty much like @dantleech is suggesting, but would cover pitfalls mentioned by @dbu .

Would this approach be considerable for you guys?

[dantleech]

Yeah that would be nice and flexible, there is also the Doctrine cache library: https://github.com/doctrine/cache/tree/master/lib/Doctrine/Common/Cache which is more likely to be present already in Symfony projects.

[lolautruche]

I'm not familiar with Doctrine cache, but does it support anti-dog pile effects and hierarchical cache? These are 2 critical features for this job IMHO

[dantleech]

Maybe @Ocramius can answer that?

[dantleech]

Also, the method I was thinking of was more like a list of rules, like in varnish, i.e.

0xb75096d0 1318329475.377475    10      obj.http.x-url ~ test
0xb7509610 1318329470.785875    20G     obj.http.x-url ~ test

Where the request is tested against each expression (evaluated by the Symfony ExpressionLanguage) and the list tidied up when new bans are added.

I guess thats a different method than the one you are planning?

[lolautruche]

Yes, looks like a different approach.