Recovery Manager

The Recovery Protocol for Minibase is based on ARIES. It uses WAL (write ahead logging), and it supports the use of the Steal/No-force policy by the Buffer Manager.

The Recovery Manager ensures durability to the database. It operates in two modes. It runs in the Normal Mode during normal database operation and in the Restart Mode during crash restart.

While running in the Normal Mode, the Recovery Manager supports transaction atomicity by supporting rollbacks for aborted transactions. It also performs checkpoints to assist in recovery from crashes. The Restart Mode is invoked by the system after a crash. The Minibase server always excutes the Restart operation at server startup, so as to ensure that the system is in a consistent state.

Design Decisions

The Recovery Manager provides methods for checkpointing, rollback and restart. It primarily interacts with the Log Manager. One of the functions of the Recovery Manager is to generate records to log different actions (abort, update, commit). The Recovery Manager creates the log records which are written to the log file by the Log Manager.

The Recovery Manager also interacts to a lesser degree with the Buffer Manager and the Transaction Manager. It is invoked by the Transaction Manager for transaction rollback. It requests the Buffer Manager for the Dirty Page list and the Transaction Manager for the Transaction table.

Click here for the public interface.
Click here for the implementation details.

Back to the List of Components
Back to the Minibase Home Page