Re: reorderbuffer: memory overconsumption with medium-size subxacts

Hi,

On 2018-12-16 17:30:30 -0300, Alvaro Herrera wrote:
> On 2018-Dec-16, Andres Freund wrote:
> 
> > > I think there's a one-line fix, attached: just add the number of changes
> > > in a subxact to nentries_mem when the transaction is assigned to the
> > > parent.
> > 
> > Isn't this going to cause significant breakage, because we rely on
> > nentries_mem to be accurate?
> > 
> >     /* try to load changes from disk */
> >     if (entry->txn->nentries != entry->txn->nentries_mem)
> 
> Bahh.
> 
> Are you suggesting I should try a more thorough rewrite of
> reorderbuffer.c?

I'm saying you can't just randomly poke at one variable without actually
checking how it's used.  I kinda doubt you're going to find something
that's immediately backpatchable here. Perhaps something for master
that's then backpatched after it matured for a while.

Greetings,

Andres Freund