Consider the following two-pass strategy for crash recovery using a sharp checkpoint and physical logging: (1) The first pass is a backward pass in which active transactions are rolled back. Active transactions are identified as described in Section 22.2. The pass extends at least as far as the begin record of the oldest active transaction or the most recent checkpoint record, whichever is earlier in the log. As update records for these transactions are encountered in the scan, their before-images are applied to the database. (2) The second pass is a forward pass from the most recent checkpoint record to roll forward, using after-images, all changes made by transactions that completed since the checkpoint record was written (compensation log records are processed in the same way as

ordinary update records so that aborted transactions are handled properly). Does the procedure work?

What will be an ideal response?

Yes. The first pass rolls back all the updates made by transactions that were active when the crash occurred. The second pass does (or redoes) all the changes made by transactions that committed since the last checkpoint. (those changes might or might not have yet been stored in the database.)