- * When not using a directory, most resource names will hash to a new static
- * master nodeid and the resource will need to be remastered.
+ * All MSTCPY locks are purged and rebuilt, even if the master stayed the same.
+ * This is necessary because recovery can be started, aborted and restarted,
+ * causing the master nodeid to briefly change during the aborted recovery, and
+ * change back to the original value in the second recovery. The MSTCPY locks
+ * may or may not have been purged during the aborted recovery. Another node
+ * with an outstanding request in waiters list and a request reply saved in the
+ * requestqueue, cannot know whether it should ignore the reply and resend the
+ * request, or accept the reply and complete the request. It must do the
+ * former if the remote node purged MSTCPY locks, and it must do the later if
+ * the remote node did not. This is solved by always purging MSTCPY locks, in
+ * which case, the request reply would always be ignored and the request
+ * resent.