Squid in front of Mogilefs tracker

[Brad Fitzpatrick]

We cache the results of the get_paths request to the tracker in memcached,
so we don't hit it as much.


On Mon, 26 Jun 2006, jb wrote:

> Thanks Mark.
>
> In fact, my goal is to reduce the step : client > tracker > mysql > tracker
> ...
> I'd like to be able to have one MogileFS installation, cache docs at the top
> level, and if required in a few months, to replicate on another location my
> squid farm (idea is to build a little custom CDN).
>
> If I add appropriated headers (Expires, Last-Modified..) to the MogileFS
> client API when it returns content, Squid can cache the result, no ?
>
>
> 2006/6/26, Mark Smith <mark at plogs.net>:
> >
> > > Is it possible to put Squid in front of MogileFS tracker ?
> >
> > In front of the storage nodes, yes.  Not the tracker - that speaks a
> > custom non-HTTP protocol.
> >
> > But if you're talking about the part that serves the files, sure.  You'd
> > have to run both a webdav server (such as mogstored, or lighttpd) as
> > well as your squid cache - just put them on separate ports.
> >
> > Then, you can configure your hosts to set the alternate GET port to send
> > all of the retrievals to your squid cache.  For example, if you take a
> > standard storage node, called store1, with mogstored/lighttpd running on
> > port 7500 (default), you could start squid on port 7600, and then do:
> >
> > $ mogadm host modify store1 --getport=7600
> >
> > That would modify store1 and change the GET port to 7600.  Once that's
> > done, all URLs returned by the tracker will point at :7600 on the
> > destination machine - but only for GET requests.
> >
> > That should do what you want...
> >
> >
> > --
> > Mark Smith
> > mark at plogs.net
> >
>