Transaction Logs and Automatic Recovery

The Oracle RDBMS performs automatic recovery each time it is started. It verifies that the contents of the tablespace files are coordinated with the contents of the online redo log files. If they are not, Oracle applies the contents of the online redo log files to the tablespace files (roll forward), and then removes any uncommitted transactions that are found in the rollback segments (roll back). If Oracle cannot obtain the information it requires from the online redo log files, it consults the archived redo log files.

Microsoft SQL Server 7.0 also performs automatic data recovery by checking each database in the system each time it is started. It first checks the master database and then launches threads to recover all of the other databases in the system. For each SQL Server database, the automatic recovery mechanism checks the transaction log. If the transaction log contains any uncommitted transactions, the transactions are rolled back. The recovery mechanism then checks the transaction log for committed transactions that have not yet been written out to the database. If it finds any, it performs those transactions again, rolling forward.

Each SQL Server transaction log has the combined functionality of an Oracle rollback segment and an Oracle online redo log. Each database has its own transaction log that records all changes to the database and is shared by all users of that database. When a transaction begins and a data modification occurs, a BEGIN TRANSACTION event (as well as the modification event) is recorded in the log. This event is used during automatic recovery to determine the starting point of a transaction. As each data modification statement is received, the changes are written to the transaction log prior to being written to the database itself. For more information, see the “Transactions, Locking, and Concurrency” section later in this chapter.

SQL Server has an automatic checkpoint mechanism that ensures completed transactions are regularly written from the SQL Server disk cache to the transaction log file. A checkpoint writes any cached page that has been modified since the last checkpoint to the database. Checkpointing these cached pages, known as dirty pages, onto the database, ensures that all completed transactions are written out to disk. This process shortens the time that it takes to recover from a system failure, such as a power outage. This setting can be changed by modifying the recovery interval setting by using SQL Server Enterprise Manager or with Transact-SQL (sp_configure system stored procedure).