RAC Redo Recovery

According to 1-23 of the Concepts Manual, “After an instance failure, Oracle automatically performs instance recovery. If one instance in a RAC environment fails, then another instance recovers the redo for the failed instance.”

For this to work, I guess that each instance must keep its redo logs on the shared storage (no use of locally-attached storage, which I believe is possible in 11g). Each instance must also know which log files belong to which other instance.

As far as I can see, there also must be a way of distinguishing how much of the logged changes have already been stored “properly”, and how many remain to be written. This would be true in a RAC or a single-node environment. The first possibility that occurs is that the process of writing changes back to disk is also logged.