Ensure Timely Event Processing
The fault tolerance callback function is the only channel of information from Rendezvous fault tolerance software to your program. Rendezvous fault tolerance software can call the callback function at virtually any time—whether active or inactive, a member program must be ready for a callback event.
To ensure timely processing of fault tolerance events, follow these guidelines:
| • | Associate fault tolerance member events with a high priority queue, so events on other queues do not delay fault tolerance callback processing. |
| • | Ensure that callback functions return promptly. This rule applies to all callback functions that dispatch in the same thread as the fault tolerance callback function. |
If a callback function monopolizes the dispatch thread (for example, by blocking, or with a lengthy computation), then the fault tolerance callback function might remain in its queue waiting for dispatch. Such backlog could cause problems such as these:
| • | Delay a program from activating, resulting in interrupted service. |
| • | Delay a program from deactivating, resulting in redundant service. |
| • | Delay a timer, or the arrival of a fault tolerance control message, interfering with fault tolerance software. |
| • | Delay the arrival of regular messages, interfering with the program performance. |