Subject Read Committed Transaction
Author Jim Starkey
Ann W. Harrison wrote:

> Dmitry Yemanov wrote:
>> 1) Backport some changes into B1_5_Release branch to release v1.5.3
>> RC1 next
>> week. The list of fixes to be proposed and discussed here.
> I'd like to commit the change to the lock data for read-committed
> transactions so a commit retaining will move the oldest active
> transaction.

I believe that if a program has a valid blob id, advancing the oldest
active transaction will allow the blob to be garbage collected, meaning
that an attempt to use the blob id may fail, report a corrupt database,
or silently fetch the wrong blob. A check in blob open to check for the
case and return an appropriate error may be a suitable solution. In
general, however, removing the floor from underneath a transaction is a
pretty dangerous thing to do.


Jim Starkey
Netfrastructure, Inc.
978 526-1376