Hi, this one is bugging me for quite some time...

I read that when a checkpoint occurs, "all" the dirty buffers are flushed into datafiles. And control files and datafile headers get updated with latest SCN.
But I believe "dirty buffers" include even data blocks modified in on-going transactions (uncommitted). Does this mean uncommitted transaction get saved in datafiles during a checkpoint ?

I haven't seen any doco suggesting that only commited records are written to the disk during a check point ...

But from various other features, like LOG_CHECKPOINT_INTERVEL, a logfile not being reused until checkpoint completion, crash recovery reading logfiles from the last checkpoint etc makes me assume that uncommited data from the data buffer is written to the disk.

Cheers

Sathyaram

Hi odba,
Yes, at checkpoint, the dirty buffers get flushed to disk - they are still recorded on disk as uncommitted, therefore they can still be rolled back.
Rgs,
Breen.

Thanks for your comments! Come to think of it, rollback segs get flushed too. So oracle should be possible to read old values from there...