From: Brian Candler <B.Candler@...>
Date: 2003-05-30T18:54:44+09:00
Subject: Re: Object Prevaylence vs. OODBMS or Madeleine vs. DyBase

On Fri, May 30, 2003 at 06:13:56PM +0900, Anders Bengtsson wrote:
> Since all "writes" to Madeleine have to go through the
> execute_command() method, which is synchronized, this won't be a
> problem.

As you say, reads don't benefit from this protection. It might not be as bad
as I thought: I was thinking that if you follow a reference from object A to
object B, and the reference was changed in the mean time, you could end up
pointing to a garbage object B. But actually you would be holding a
reference to the object yourself, so it wouldn't be garbage-collected.

But you could still end up with inconsistent reads, i.e. the objects are in
a state which is neither how they were before execute_command nor after it,
if execute_command is midway through its processing.

> Maybe I should add a way to access Madeleine's big write lock from the
> outside, if anyone wants to solve things the "global mutex" way.

Sure. Perhaps you could add a wrapper for the traditional model of a shared
read lock, and an exclusive write lock. (I'm not quite sure how to implement
this; I guess the read lock manager would have to grab the write lock
whenever 1 or more people asked for a read lock, and return it when this
number went to 0, but you want a way to prevent starvation when someone is
waiting for a write lock, e.g. no new readers are allowed until the write
has had a chance)

Cheers,

Brian.