[JCS-10] Remote Cache Events Are Lost Between Primary Failure and Failover Connection

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: jcs-1.2.7.8, jcs-1.2.7.9
• Component/s: RMI Remote Cache
• Environment: All
• Assignee: Aaron Smuts
• Reporter: Aaron Smuts
• Created: 22/Aug/06 02:03
• Updated: 23/Jul/09 19:14
• Resolved: 22/Aug/06 14:53

Description

When the remote cache client encounters an error communicating with the server it substitutes a balking facade (a zombie) for the service and then initiates the failover / recovery process. Events on the remote cache event queue are sent to the zombie during the period between the failure and the failover connection. The Zombie simply balks. Hence the events are lost. In an environment with a failover-a hot standby-remote server, this is not desireable. It would be better if no events were lost to the Zombie.

The best solution is to make the Zombie a bounded queue. Events that flood into the Zombie should be put to the failover or the primary when the connection is established. The Zombie must be capable of limiting the size of this queue. When the remote cache restore process sets the remote server handle on client, the client should check to see if if the current handle is a zombie. If the current server is a zombie, it should put into service the good server and then walk the zombie queue.

Since failover switching is very quick, few items should be queued.

When handle error is called, the event should be passed to the zombie so it is not lost.