Re: object versions in core Bill Frantz (frantz@netcom.com)
Mon, 11 Sep 1995 15:38:29 -0800

At 13:22 9/9/95 -0400, Jonathan Shapiro wrote:
>Norm commented the other day that there could be up to three versions
>of an object in memory...
>
>One case I've thought of arises in nodes, where you can have:
>
> 1. The current version
> 2. The last checkpoint version
> 3. The home location version (nodes only)
>
>The last can be in memory briefly because it's node pot has been
>loaded in order for the migrator to overwrite it, but it won't ever be
>in the object hash table, so it might as well not be present for
>object search purposes.

But, the node pot itself may be in memory as a result of refering to some other node.

>Are there funny cases that arise due to journaling?

There certainly are. Particulary partially written pages or pages which have had one copy written and not the other(s). See the transaction processing paper.

Bill