Re: The curious case of two inserts, a shrinking xmax, and a ShareLock on transaction

On Tue, Sep 22, 2015 at 10:44 PM, Alvaro Herrera <alvherre(at)2ndquadrant(dot)com>
wrote:

> Jeff Dik wrote:
>
> > I'd really love to learn:
> >
> > 1. Why the xmax for foo_id1 goes from 696 to 1 and what does that
> > mean?
>
> When two transactions want to lock the same row, the xmax field is a
> multixact, no longer a bare transaction ID. This is an object that
> resolves to multiple transaction IDs.
>
> > 2. How does transaction A know it needs to take a ShareLock on
> > transaction B?
>
> Because it reads the two transaction ID values from pg_multixact.
>
> > 3. What is a virtualtransaction and what do its numerator and denominator
> > mean?
>
> It's not a division operation (so no numerator/denominator). The part
> before the / is a backend ID and the part after the / is a local
> transaction counter. It's just an identifier for the transaction,
> useful for the time before the transaction acquires a transaction ID.
> This optimizes that a transaction that doesn't modify tuples does not
> need to acquire a transaction ID (and thus keeps transaction ID
> consumption rate low.)
>

Thanks so much for these explanations!

Is there any way to inspect a multixact via psql to see what transaction ID
values it has? I wasn't able to find anything while searching for an hour
or so.

Thanks again!
Jeff

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