| 摘要 |
A system for managing failover in a server cluster. In response to detecting a failed server, subscription message processing of a failover server is stopped. A subscription queue of the failed server is opened. A marker message is published to all subscribers of a particular messaging topic. The marker message includes an identification of the failover server managing the subscription queue of the failed server. Messages within the subscription queue of the failed server are processed. In response to determining that a message in the subscription queue of the failed server is the marker message, the subscription queue of the failed server is closed. Then, the failover server resumes processing of its original subscription queue looking for the marker message, while processing yet unseen messages from the queue. Once the marker message is found in the original subscription queue, normal operation is resumed.
|
