Why oracle LGWR process should write redo log file before DBW0 process flush buffer cache?

[稀里哗啦]

I don't know what is the benifit from this action.

[Tom Geudens · Accepted solution]

Hi,
Imagine a database crash. You want to recover your database up to the last committed data. You restore your last fullbackup and start playing back ... the redologs. And that's the reason. Funny as it may sound, but for Oracle it's more important (relatively speaking off course :-) that the redolog gets written than that the actual database page gets written.

Hope this explains,
Tom

[Brian Crabtree]

Well, hopefully I wont get to deep into this, but I will try to answer generally. In this case, you have it backwards. When oracle triggers a log switch (called a checkpoint BTW), it forces the buffer cache to write all dirty buffers to the disk.

While oracle is running, it allows processes to update data in the memory structure rather than update data on the disk due to the cost increase. While this is helpful on the surface by increasing the speed at which a process will come back to you, it does mean that what the database on disk is and what is true in memory is not in sync.

The reason the the checkpoint is flushed at that point, is so that the log in question is concurrent with the database, and can be used in a recovery situation (if that log were the last log used in a point-in-time recovery). Checkpoints can be forced either manually, or by setting the "LOG_CHECKPOINT_TIMEOUT" or "LOG_CHECKPOINT_INTERVAL" in the init.ora.

I hope this answers some of your questions.

Thanks,

Brian

[Thierry Poels_1]

Hi,

next to data integrity, which already has been mentioned, think about SPEED!

A data block might get updated in memory several times before effectively being written to disk.

regards,
Thierry.

[Simeon Fox]

This is one the key features of Oracle. Redo logs record all changes made to the datbase. The redo logs are a straight dump of the O/S blocks and can be written instantly to disk. The more complex operation of writing to the oracle datafiles can be safely deferred - the so-called "lazy write". With performance in mind, Oracle never does more disk writing than it needs to - it is the continuous flushing of the redo logs to disk that guarantees recoverability.

When a user issues a COMMIT, LGWR puts a commit record in the redo log buffer and flushes the entire buffer to disk. The so called "fast commit" mechanism. Oracle can record a successful commit even though the data blocks have not yet been written to disk.

[allen_7]

Data integration - benefit