Re: PG 13 release notes, first draft - Mailing list pgsql-hackers

On Tue, May  5, 2020 at 12:50:01PM -0500, Justin Pryzby wrote:
> On Tue, May 05, 2020 at 01:18:09PM -0400, Bruce Momjian wrote:
> > > |Release date: 2020-05-03
> > > => Should say 2020-XX-XX, before someone like me goes and installs it everywhere in sight.
> > 
> > Agreed!
> > 
> > > |These triggers cannot change the destination partition. 
> > > => Maybe say "cannot change which partition is the destination"
> 
> Looks like you copied my quote mark :(

I kind of liked it, but OK, removed.  ;-)

> > > | Allow hash aggregation to use disk storage for large aggregation result sets (Jeff Davis) 
> > > |  Previously, hash aggregation was not used if it was expected to use more than work_mem memory. This is
controlledby enable_hashagg_disk. 
 
> > > => enable_hashagg_disk doesn't behave like other enable_* parameters.
> > > As I understand, disabling it only "opportunisitically" avoids plans which are
> > > *expected* to overflow work_mem.  I think we should specifically say that, and
> > > maybe suggest recalibrating work_mem.
> > 
> > I went with "avoided":
> > 
> >     Previously, hash aggregation was avoided if it was expected to use more
> >     than work_mem memory.  This is controlled by enable_hashagg_disk.
> 
> I think we should expand on this:
> 
> |Previously, hash aggregation was avoided if it was expected to use more than
> |work_mem memory, but it wasn't enforced, and hashing could still exceed
> |work_mem.  To get back the old behavior, increasing work_mem.

I think work_mem has too many other effects to recommend just changing
it for this.

> |The parameter enable_hashagg_disk controls whether a plan which is *expected*
> |to spill to disk will be considered.  During execution, an aggregate node which
> |exceeding work_mem will spill to disk regardless of this parameter.
> 
> I wrote something similar here:
> https://www.postgresql.org/message-id/20200407223900.GT2228%40telsasoft.com

I think this kind of information should be in our docs, not really the
release notes.

> > > | This is controlled by GUC wal_skip_threshold. 
> > > I think you should say that's a size threshold which determines which strategy
> > > to use (WAL or fsync).
> > 
> > I went with:
> >     The WAL write amount where this happens is controlled by wal_skip_threshold.
> >
> > They can use the doc link if they want more detail.
> 
> I guess I would say "relations larger than wal_skip_threshold will be fsynced
> rather than copied to WAL"

How is this?

    Relations larger than wal_skip_threshold will have their files fynsced
    rather than writing their WAL records.

-- 
  Bruce Momjian  <bruce@momjian.us>        https://momjian.us
  EnterpriseDB                             https://enterprisedb.com

+ As you are, so once was I.  As I am, so you will be. +
+                      Ancient Roman grave inscription +