Last week I gave an internal 2-days dba course. All three students had good experience of filesystem backup and unix, but hardly any oracle experience.
I figured out recovery is more complex than I thought! By explaining to other, how often do you realise do you do not know the answer yourself? It has been very educative to me too…
First inconsistency I discovered :
I tried to simulate a disaster by removing all controlfiles, redologs, datafiles. I expected the students to notice “something”. But, O Surprise, the db kept running, nothing written in alert logs, it was possible to create and drop tables, nothing went wrong (but they were no files). Probably the OS has not notified the file were no more there, and Oracle opened the files already. Strange. Even SHUTDOWN has been successful! But of course STARTUP did not work.
Next time I prepare a B&R workshop, I must invest time to check that a disaster is effective!
Report unrecoverable datafiles and validate backups. I wrote this in the rman technology forum this morning: REPORT NEED BACKUP is not suffisant to have consistent backups, nor RESTORE VALIDATE DATABASE!
More inconsistencies :
somehow, when users are doing backups with nocatalog and try to restore with the recovery manager catalog, it is not working. Well, it is not that surprising, but it really increases the difficulty of recovery procedures!
However, those “inconsitencies” were good in this informal training, feedback from students was positive, we did “survive” all those problems and they realised RMAN recovery is not as simple as RMAN backup…