Warwick Burrows wrote:
Hi James, some questions about your cache update notifications:
I have a metadata store (backed by db2) based on the jdbcstore. Since it is the jdbstore then I assume that entries returned from the store are cached and update notifications will be used to notify other slides when a metadata object changes. If this is the case then what if there are two simultaneous requests to lock the same file at almost the same time such that the file is locked by the first caller and the lock metadata has been updated but the cache update notification hasn't yet been received by the second slide to which the next lock request has been directed. It would seem that the second slide will still get the outdated lock information and attempt to lock the object as well. Is this the case? If so can I disable caching for my jdbcstore-based so that metadata requests are made directly to the db?
What happens with locks is going to depend on the store implementation. Both instances will request locks, and either the first will get it and the second will fail, the second will overwrite the first, or they'll both get locks. I really don't know what the behavior will be. Feel like volunteering to test? :)
One thing that should work is setting the lock-cache-size parameter of your <store> node to -1 or 0. I think -1 will disable it, and if not 0 should make the cache worthless. This will disable caching just for locks. I don't know how this will effect performance, though.
Also, I notice that you can select which parts of the store are monitored for notifications. Can multiple URIs be specified? Eg. "/files/subtree1" and "/files/subtree2"? Can I monitor the "/history" subtree for changes too?
Subscribing to multiple paths would be an enhancement, but should be doable. The way notifications work it would mean a separate subscription for each path (actually four subscriptions for each path, currently), but that shouldn't be a problem.
Subscribing to / should be fine, though. Notifications are only sent when something changes, so dormant nodes such as /actions will be ignored.
-James
Thanks, Warwick
-----Original Message-----
From: James Mason [mailto:[EMAIL PROTECTED] Sent: Wednesday, August 11, 2004 2:11 PM
To: Slide Users Mailing List
Subject: Re: Cache refresh notification in two different instance of Slide
running on different boxes.
Just to clarify this, you no longer need to set the cache-mode parameter to cluster. You can actually remove the parameter entirely.
Clustering is now enabled using an event listener. See the wiki for details.
-James
Warwick Burrows wrote:
That was my understanding too, Carlos. James has improved the cluster cache solution to send updates notifications to the cluster participants. Cache update notifications are only sent when you are running in clustered mode as that is the only operating mode in which sending the notifications makes sense.
It is also the case that caching in clustered mode, to my understanding, is now enabled. Clustered cache mode in 2.1 M1 meant simply that the global cache was turned off altogether so that no caching was done. This was to get around the problem of an update in one servers cache not being reflected in the caches of the rest of the servers in the cluster. Now, because of James addition of cache update notifications for cluster participants the global cache now works and is turned on for clustered mode.
Daniel, is there a downside to running in clustered mode that you specifically don't want?
Warwick
-----Original Message----- From: Carlos Villegas [mailto:[EMAIL PROTECTED] Sent: Wednesday, August 11, 2004 11:21 AM To: Slide Users Mailing List Subject: Re: Cache refresh notification in two different instance of Slide running on different boxes.
Well, I don't know about the synchronization feature but I would think
that it's not possible to synchronize caches if the servers are not configured in a special mode which allows that. My understanding is that that special mode is precisely the clustering feature. Thus you can't have one without the other.
Well, maybe I'm wrong and there's another mode of operation...
Carlos
Daniel Varghese wrote:
Hello James,
Once again I will tell you, what I'm looking for
We have two instance of Slide running on two different Unix boxes
and
I'm not planing to use Cluster, In this senario how do we notify the cache information between different server instances.
So How do I configure this in Domain.xml for cache synchronization refresh between servers.
rgds Daniel
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
