This is exactly what I needed James. Thanks!

> They should be asynchronous. Daniel Florey wrote the 
> notification mechanism, so he can confirm, but I'm pretty 
> sure on this. 

Daniel, can you confirm that notifications are asynchronous? Or is there a
write-up on the notifications that I can look at?

And I have another question on the cluster cache :-) There is a "base-uri"
parameter that is specified in the configuration. If this is set to "/files"
will it pick up all changes to objects under /files? For example when a file
is checked out/in with a new revision will it also notify other slide
servers of /history path updates that stem from the change even though
/history isn't specifically configured as the base-uri?

Thanks,
Warwick


> -----Original Message-----
> From: James Mason [mailto:[EMAIL PROTECTED] 
> Sent: Tuesday, December 07, 2004 8:40 PM
> To: Warwick Burrows; Slide Users Mailing List
> Subject: Re: Cluster cache
> 
> 
> On Tue, 2004-12-07 at 18:10 -0800, Warwick Burrows wrote:
> > Hi James,
> >  
> > I was hoping you could answer some questions about the 
> cluster cache 
> > design. We would like to go into production with Slide but doing so 
> > requires the cluster cache implementation -- or no caching at all 
> > which is not even an option :-)  We need to know the cluster cache 
> > mechanics so we can cater for them.
> >  
> > - Are the updates (notifications) between servers performed 
> > synchronously or asynchronously?
> 
> They should be asynchronous. Daniel Florey wrote the 
> notification mechanism, so he can confirm, but I'm pretty 
> sure on this. When a resource is modified the server sends 
> notifications to all subscribers that there are events 
> waiting for them. The subscribers (other servers) then POLL 
> the server and get a list of changed resources. Each 
> subscriber then removes the resource in the list from their cache.
> 
> > - If an update to a server fails will it be retried?
> 
> It depends on where the failure is. If a subscriber is unable 
> to POLL a server, it will retry. In fact, each subscriber 
> periodically POLLs the server just to be sure it hasn't 
> missed any notifications. The POLL frequency should be configurable.
> 
> >  
> > Any other design points I should be aware of?
> 
> Not that I can think of.
> 
> >  
> > Incidentally I had mentioned that locks could be an issue with the 
> > cache design in that two lock requests coming in at the 
> same time may 
> > both return with the lock. It isn't an issue for us though 
> as locking 
> > is controlled by our DAV applications and the lock information they 
> > share is stored in a DB. Only one app is guaranteed to get the lock 
> > and so create a Slide lock. One way to mitigate this 
> problem for other 
> > clients may be to allow for the "lockstore" store to be 
> specifically 
> > configured so that only lock related metadata isn't cached.
> 
> ExtendedStore has five (I think) separate caches. You can set 
> the size of each cache by setting parameters on the store. 
> One of the caches is for locks. Take a look at the source for 
> ExtendedStore, since I'm not sure where, if anywhere, the 
> parameters are documented.
> 
> -James
> 
> >  
> > Thanks,
> > Warwick
> >  
> >  
> 

---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]

Reply via email to