Re: Instance Recovery Question

AJG (noemail_at_noemail.noemail) wrote:
: Instance crashes
: - Change in log buffer not written to disk yet

As I understand it...

The Redo Log buffer is ALWAYS written first. If this data has not been written to disk, then the transactions outstanding are lost. A "commit" does not return until the log file (all copies of it) have been written successfully by LGWR.

The redo logs are essential for database recovery, which is why you should have multiple copies, and why (if you switch log files too quickly) processing can freeze until a log file becomes free; all transactions are written here :-) If you have archive logging on, then the archive logs are just(!) historical copies of the redo logs, and using these (and the current log file) you can roll forwarded restored-from-tape datafiles to the current database position. This is how the technique of "archive log shipping" works.

Also note that during a "checkpoint", all modified datablocks (even if the transaction is not committed yet) are written to disk.

Instance recovery involves rolling forwarded all the data in the redo logs, and then rolling back the uncomitted transactions (maybe delayed in oracle 8).

--

rgds
Stephen Received on Sat Dec 05 1998 - 14:51:16 CST