Re: Rethinking MemoryContext creation

On 2017-12-12 14:50:37 -0500, Robert Haas wrote:
> On Tue, Dec 12, 2017 at 2:30 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> > 379 CurTransactionContext was never used
> > 24 CurTransactionContext was used
> > 66978 ExprContext was never used
> > 17364 ExprContext was used
> > 11139 SPI Exec was never used
> > 2421 SPI Exec was used
> > 38386 ginPlaceToPage temporary context was never used
> > 3348 ginPlaceToPage temporary context was used
>
> It strikes me that a way to optimize these cases even more would be to
> postpone creating the context until it's actually needed. That might
> not always be a reasonable plan -- in particular, it occurs to me to
> think that CurTransactionContext is probably so widely used that
> changing anything about how it works would probably be really painful
> -- but it might be possible in some cases.

That's not generally easy without slowing things down though - e.g. we
don't want to check for ExprContext's existence before every use, there
can be billions of usages in a single analytics query. The branches (yea
yea ;)) would show up as being noticeable.

There are a few places where could probably reliably *detect* that
they're not needed however. E.g. plenty executor nodes only need a
ExprContext when either a qual or projection is needed, both are pretty
cheap to detect at ExecInitNode() time.

> Another idea I have is that perhaps we could arrange to reuse contexts
> instead of destroying them and recreating them. For example, when
> asked to delete a context, we could instead push it on a linked list
> of old contexts, or only if the list isn't too long already, and when
> asked to create one, we could pop from the list. Or we could keep
> around an array of, say, 1024 contexts that are never freed and only
> allocated dynamically when we run out.

I'm a bit doubtful that's going to help, maintaining that list isn't
going to be free, and the lifetime and number of those contexts aren't
always going to match up.

I think you're somewhat on to something however: I do think that
especially executor startup would have a good chance of avoiding
noticeable overhead by batching the allocation of all these tiny
contexts together - I just don't quite know how given how our current
initialization works. The best thing to do would probably to do two
walks during executor initialization, one to compute memory sizes, and a
second to initialize pre-requested memory that's laid out
serially... But obviously that's not a small change.

Greetings,

Andres Freund