Re: SSI patch version 14

On 25.01.2011 22:53, Kevin Grittner wrote:
> Heikki Linnakangas<heikki(dot)linnakangas(at)enterprisedb(dot)com> wrote:
>> On 25.01.2011 05:30, Kevin Grittner wrote:
>
>> The readme says this:
>>> 4. PostgreSQL supports subtransactions -- an issue not mentioned
>>> in the papers.
>>
>> But I don't see any mention anywhere else on how subtransactions
>> are handled. If a subtransaction aborts, are its predicate locks
>> immediately released?
>
> No. Here's the reasoning. Within a top level transaction, you
> might start a subtransaction, read some data, and then decide based
> on what you read that the subtransaction should be rolled back. If
> the decision as to what is part of the top level transaction can
> depend on what is read in the subtransaction, predicate locks taken
> by the subtransaction must survive rollback of the subtransaction.
>
> Does that make sense to you?

Yes, that's what I suspected. And I gather that all the data structures
in predicate.c work with top-level xids, not subxids. When looking at an
xid that comes from a tuple's xmin or xmax, for example, you always call
SubTransGetTopmostTransaction() before doing much else with it.

> Is there somewhere you would like to
> see that argument documented?

README-SSI .

--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com