The Cluster Edition provides failover recovery, during which time the database remains online and can be used. However, if you request data that was modified by the failed instance, and must be recovered, the user process is blocked until the data is brought to a consistent state by the failover recovery process.
The Cluster Edition supports:
Recovery that occurs at cluster start-up, which is the same as the nonclustered Adaptive Server recovery that happens when you perform a cold restart of the server.
Failover recovery, which recovers data that a failed instance modified while other instances were using the same database in a shared-disk cluster. In this chapter, “failover” refers to an instance failover, not a client connection failover.
Recovery begins when the cluster membership service detects a failure. The recovery event handler on the coordinator instance recovers the system databases. Then surviving instances (including the coordinator instance) recover the user databases in parallel. One of the PCM threads on each instance performs the recovery task. Other threads on the coordinator instance, as well as other instances, pursue other activities.
If trace flag 3483 is on, distributed failover recovery is disabled, and user databases are recovered in serial order by the recovery event handler on the coordinator instance.
Processes trying to access data modified by the failed instance are blocked until recovery is complete.
The Cluster Edition recovers user databases in the order specified by their database IDs (dbid), and does not use any user-defined recovery order during instance failover recovery.
