On Sat, Apr 30, 2016 at 4:54 AM, David Rowley
<david.row...@2ndquadrant.com> wrote:
>> Right, but they're probably not doing the SAME work.  You can look at
>> EXPLAIN (ANALYZE, VERBOSE, BUFFERS) to see.  Of course, all the work
>> above the Gather node is being done by the leader, but the stuff below
>> the Gather node often has a bit of participation from the leader, but
>> is mostly the workers.
> Robert, I'd imagine that most of your tests to make you think what you
> do would have come from testing parallel seq scan, where perhaps
> Daniel's comes from testing something like parallel aggregates, or at
> least something that gives the workers a decent amount of work per
> tuple returned.

Sure, if you're doing parallel aggregate with a small number of output
groups, the work distribution will be much more nearly equal.  I
wasn't trying to dispute that.  There are many cases where it doesn't
work out that way, though.

> So I'd say this very much depends on how busy the main process is
> pulling rows from each worker.


> It would also be quite nice if we could see at a glance how much the
> main process did, without having to go subtracting what all the
> workers managed to do.

I agree, but that looked considerably more invasive, because of the
way that the worker results get aggregated into the leader's result.
There may be a clever way to avoid that, but I didn't see it.  Or we
may want to do it anyway despite it requiring more invasive changes.
I just wasn't in a hurry to rush into all that while trying to get
parallel query v1 out the door.

Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:

Reply via email to