Re: large database maintenance

In a previous article, kaz_at_solo.bms.com () says:

>During an OS backup, some redolog entries will occur prior to the OS backup
>of the corresponding block in the database file, some will occur after.
>If the entire block is not written to the redo log, those prior changes
>will be lost.

   Every redo-entry contains the appropriate System Change Number which is also stored in the database block header. During recovery, Oracle compares the SCN of the redo entry with that of the database block, and knows whether it is appropriate to apply the entry to the block (i.e. whether the change described by the redo entry was applied to the block before of after it was backed up). This ensures that only the necessary changes are applied to the block, and that those changes are applied in the correct order.

>
>My understanding is that the only situation in which you should consider
>setting LOG_BLOCKS_DURING_BACKUP=FALSE is if your Oracle blocksize is the
>same as your OS's disk block size.

Indeed...from my "Oracle System Administration on SunOS" manual...

   "This may be set to *false* if the ORACLE block size equals the O/S     block size, or if it is otherwise ensured that (the) hot backup reads     consistent versions of blocks even if those blocks are being written     at the time."

   Considering the scarcity of information on this issue, and the danger involved if you get it wrong, I think this is one of those parameters best left alone in the interest of being able to sleep at night.

-- 
   - Doug Harris
     Database Administrator,
     System Development Division,
     Statistics Canada             ### Standard Disclaimer Applies ###