Re: [HACKERS] Transaction control in procedures

On 15 November 2017 at 16:36, Peter Eisentraut
<peter(dot)eisentraut(at)2ndquadrant(dot)com> wrote:
> On 11/8/17 18:48, Simon Riggs wrote:
>> What would happen if some of the INSERTs failed? Where would control
>> go to? (Maybe this is just "no change" in this particular proc)
>
> An exception is raised and unless the exception is caught (depending on
> the PL), control leaves the procedure. What is already committed stays.
>
>> What happens if the procedure is updated during execution? Presumably
>> it keeps executing the original version as seen in the initial
>> snapshot?
>
> correct
>
>> Does the xmin of this session advance after each transaction, or do we
>> hold the snapshot used for the procedure body open, causing us to hold
>> back xmin and prevent vacuuming from being effective?
>>
>> What would happen if a procedure recursively called itself? And yet it
>> was updated half-way through? Would that throw an error (I think it
>> should).
>
> I don't think anything special happens here. The snapshot that is used
> to read the procedure source code and other meta information is released
> at a transaction boundary.

I think we need to document that, or at least note in README

It's quite important for VACUUM.

--
Simon Riggs http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services