Initiating group crash recovery
Group crash recovery is almost always performed automatically, with no user action required. You will only initiate group crash recovery manually in situations where the cluster manager has repeatedly failed in its attempt to perform group crash recovery, or it is not present (perhaps because cluster management was disabled), or when the autorestart configuration parameter is set to OFF.
About this task
Only one member can perform group crash recovery on a database at a given time. If two members initiate a group crash recovery of the same database at the same time, the second member's command will wait until group crash recovery completes on the first member.
In the case where there are indoubt transactions on the group crash recovery member at the conclusion of the crash recovery operation, the member that performed the group crash recovery is left active and is able to accept new work; however, any data associated with the indoubt transactions is not accessible because the rows accessed by the indoubt transaction must be protected (that is, they are locked) until its eventual resolution. The fact that indoubt transactions exist on the group crash recovery member means that if the member fails before the indoubt transactions are resolved, a subsequent member crash recovery must occur when the database is next activated on the member.
In the case where a group crash recovery is issued on member A and it is found, in the course of the group crash recovery, that member B has associated with it, one or more indoubt transactions, member B remains inconsistent after the group crash recovery. This implies that before member B can be started after the group crash recovery completes, a member crash recovery must occur on member B. After the member crash recovery completes, member B is available to accept new connections. It should be noted that this subsequent member crash recovery completes very quickly because there is no redo or undo work required (the crash recovery is only required to load the indoubt transactions into the member's transaction table), and all of the locks required to protect the data affected by the indoubt transactions are already reserved for the member in the cluster caching facility. Additionally, this subsequent RESTART operation is, in most cases, hidden from the user because when AUTORESTART is enabled, the member crash recovery is performed automatically when the first connection is made to the database on the given member.
Procedure
Results
When the group crash recovery completes, the member on which the command was run will be able to accept new connections and all data will be available with the exception of data that remains locked by the unresolved indoubt transactions. After the RESTART DATABASE command completes, a connection will be maintained to the database if the user has CONNECT privilege for the database. The database will not be activated on any member other than the member on which group crash recovery was run.