Re: store module abstractions

From: Robert Collins <robert.collins@dont-contact.us>
Date: Sun, 18 Feb 2001 22:27:39 +1100

----- Original Message -----
From: "Henrik Nordstrom" <hno@hem.passagen.se>
To: "Robert Collins" <robert.collins@itdomain.com.au>
Cc: "Kevin Littlejohn" <darius@bofh.net.au>; <squid-dev@squid-cache.org>
Sent: Sunday, February 18, 2001 7:49 PM
Subject: Re: store module abstractions

> Robert Collins wrote:
>
> > I don't see that as a problem. I'm not suggesting that all storefs
> > level layers use the third layer. As an analogy, if you choose to
> > you can write a storefs module now that ignore the replacement
> > policy module and implements its own. This might make sense with COSS
> > for instance if FIFO is all you can efficiently support.
>
> I see it as a problem, as you might well end up with only a few of the
> FS:es actually using that layer, while the others can't because of other
> needs, requirements of efficiency reasons.

That was my whole point actually: that coss is different to null is different to ufs|diskd|aufs is different to reiserraw. And that
the ufs|diskd|aufs being more alike than different should have the differences puilled out and put somewhere else, and the
similarities merged into a single entity.

> > > My preference at the moment is a very thin layer implementing store* calls -
> > > storeDirRebuild, storeRead, storeWrite, storeUnlink, etc. - and then
>
> > Right. That sounds like what i'm talking about. That layer is async itself.
> > It *should not* intergace to reiser/coss IMO. They don't have process's analagous to dirRebuild do they?
>
> Which makes it realistic to make these more like libraries than layers.

Yes. Uhmm if I referred to layers, sorry. I started out with abstractions rather than layers.

And yes a library concept is what I'm talking about.
My two main points being:
* the API to the library should be low-level and indentical for async/stdio/whateverdiskd uses.
* Ideally the choice of library is done at run time via squid.conf.

I agree with all your points in other two/thee mails Henrik. I probably expressed myself unclearly.

Rob
Received on Sun Feb 18 2001 - 04:25:42 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:13:32 MST