Re: The store interface

Kevin Littlejohn wrote:

> Note, on the squidfs code I've got (which the company, in their infinite
> wisdom, have scheduled me to work on early August *sigh*), file 'name' is
> not known until _after_ file is opened - as any identifier for any file is
> allocated by the fs-specific code on opening. For me, it makes more sense
> to follow Henrik's suggestions.

And in the very slimmed squidfs code I am planning, "filename" is not
known until at least two "block" of data is written (or all, depending
on which path I select). Hence my worries and suggestions for a change
in naming (sfileno) and hints (tell the filesystem the estimated size of
the object if known). Here open() for write does not really exists at
the filesystem level, only at the interface level (creates a
storeIOHandle). No disk data gets allocated until there actually is
something to store, and for most objects ("small") nothing actually gets
stored on disk prior to closing the object.

My filesystem will also require another store manager (recycling
policy/method), hence the discussion on a three tier store (access
methods used to read/write objects, filesystem for persistent storage,
manager to keep track of all objects and recycle space when needed) with
well defined interfaces betewen the different parts.

/Henrik
Received on Tue Jul 29 2003 - 13:15:59 MDT