During a typical database operation, crash recovery occurs whenever the database is started in single or multi-user mode. The process is completed in three stages—physical redo, physical undo and logical undo. At the end of this process, the database is presumed to have been made durable; all outstanding transactions have been committed or rolled back.
JTA transactions require that transactions must be able to be restored to a state pending the end of the two-phase commit. This means crash recovery logs what types of locks are acquired during runtime. The lock manager generates a new note indicating a table lock has been taken and identifies the lock strength. As record operations occur, recovery notes capture information on additional table, record or schema locks applied.
After the first physical redo phase of crash recovery has completed, the lock acquisition phase identifies JTA transactions that need transactional and locking information applied.
