Re: wither the checkpoint?

From: Steve Adams <steve.adams_at_ixora.com.au>
Date: Fri, 02 Apr 1999 23:10:52 GMT
Message-ID: <7e3ipn$9cl$1@nnrp1.dejanews.com>

Hi Jonathan,

The checkpoint message sent to DBWn specifies a redo block address (RBA) up to which blocks should be checkpointed, rather than a change (SCN) as such. An RBA consists of a log file sequence number, a log block number in that file, and an offset into that block in bytes. For each block in the database buffer cache, the buffer header contains the 3 RBAs - the low RBA, the recovery RBA and the high RBA. The low RBA is the RBA of the first change to a current mode block since it came into memory. The recovery RBA is the RBA needed to recovery the block on disk. The high RBA is the RBA of the most recent change to the block in memory. In a local or global checkpoint, DBWn scans the buffer headers for dirty buffers with low RBA < checkpoint RBA && recovery RBA < checkpoint RBA. The checkpoint RBA can in fact be half way through the redo entry for the most recent change, in which case that change is checkpointed. Alternately, the checkpoint RBA can be equal to the RBA for the most recent change, in which case that change is not checkpointed at this time. In a log switch checkpoint, the checkpoint RBA is beyond the most recent change, and so all changes are checkpointed. However, it does not really matter when changes are checkpointed. It only matters that they are checkpointed before the redo log file is reused. So as you say, the question is esoteric. Nevertheless, I hope you find my answer helpful. As to a reliable reference, I think you have already referred to the best one!

Regards,
Steve Adams

---

In article <37060596.14070120_at_netnews.worldnet.att.net>,   jonathan_at_gennick.com wrote:
> I have an esoteric question on checkpoints that I need
> answered. The Oracle8 Backup & Recovery manual, under the
> heading "Managing Checkpoints", states the following:
>
> "A checkpoint identifies a redo log record such that all
>
> changes prior to the change pointed to by the checkpoint
>
> have been written to the datafiles by DBWn. "
>
> The way this sentence is written, it sounds as if the
> specific change pointed to by the checkpoint has NOT been
> written to the datafiles. Is that really true? Does anyone
> know for sure. Can anyone point me to a reliable reference
> that clarifies this?
>
> Jonathan
>

-----------== Posted via Deja News, The Discussion Network ==---------- http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own Received on Fri Apr 02 1999 - 17:10:52 CST