Redo log

In the Oracle RDBMS environment, redo logs comprise files in a proprietary format which log a history of all changes made to the database. Each redo log file consists of redo records. A redo record, also called a redo entry, holds a group of change vectors, each of which describes or represents a change made to a single block in the database.

For example, if a user UPDATEs a salary-value in a table containing employee-related data, the DBMS generates a redo record containing change-vectors that describe changes to the data segment block for the table. And if the user then COMMITs the update, Oracle generates another redo record and assigns the change a "system change number" (SCN).

Whenever something changes in a datafile, Oracle records the change in the redo log. The name redo log indicates its purpose: If the database crashes, the RDBMS can redo (re-process) all changes on datafiles which will take the database data back to the state it was when the last redo record was written. DBAs use the views V$LOG, V$LOGFILE, V$LOG_HISTORY and V$THREAD to find information about the redo log of the database. Each redo log file belongs to exactly one group (of which at least two must exist). Exactly one of these groups is the CURRENT group (can be queried using the column status of v$log). Oracle uses that current group to write the redo log entries. When the group is full, a log switch occurs, making another group the current one. Each log switch causes checkpoint, however, the converse is not true: a checkpoint does not cause a redo log switch. One can also manually cause a redo-log switch using the ALTER SYSTEM SWITCH LOGFILE command.

Redo log files occur in two types:

Before a user receives a "Commit complete" message, the system must first successfully write the new or changed data to a redo log file.

The RDBMS first writes all changes included in the transaction into the log buffer in the System Global Area (SGA). Using memory in this way for the initial capture aims to reduce disk IO. Of course, when a transaction commits, the redo log buffer must be flushed to disk, because otherwise the recovery for that commit could not be guaranteed. The LGWR (Log Writer) process does that flushing.

Having a redo log makes it possible to replay SQL statements. Before an Oracle database changes data in a datafile it writes changes to the redo log. If something happens to one of the datafiles, a recovery procedure can restore a backed-up datafile and then replay the redo written since backup-time; this brings the datafile to the state it had before it became unavailable. Standby databases in an Oracle Data Guard environment use the same technique: one database (the primary database) records all changes and sends them to the standby database(s). Each standby database applies (replays) the arrived redo, resulting in synchronization with the primary database.

...
Wikipedia