Thread: 2018-03 Commitfest starts tomorrow

Hackers!

I'll be starting the Commitfest at midnight AoE (07:00 ET, 13:00 CET) so
please get your patches in before then.

Please remember that if you drop a new and large (or invasive patch)
into this CF it may be moved to the next CF.

This last CF for PG11 should generally be restricted to patches that
have gone through review in prior CFs or are modest in scope.

Thanks,
-- 
-David
david@pgmasters.net

David Steele <david@pgmasters.net> writes:
> I'll be starting the Commitfest at midnight AoE (07:00 ET, 13:00 CET) so
> please get your patches in before then.
> Please remember that if you drop a new and large (or invasive patch)
> into this CF it may be moved to the next CF.
> This last CF for PG11 should generally be restricted to patches that
> have gone through review in prior CFs or are modest in scope.

As of right now, there are 229 entries in this commitfest (not counting
the items already committed or RWF).

It's hard for me to tell for sure, because the activity log page doesn't
stretch back far enough, but I think about 40 of those have been created
in the last 24 hours.  It's certainly well over 20, because the log does
go back far enough to show 21 patch records created since about noon
Wednesday UTC.

This is NOT how it is supposed to work.  This is gaming the system.

I think that we should summarily bounce to the September 'fest anything
submitted in the last two days; certainly anything that's nontrivial.

There is no way that we can possibly handle 200+ CF entries in a month.
A large fraction of these are going to end up pushed to September no
matter what, and I think simple fairness demands that we spend our
time on the ones that are not Johnny-come-latelies.

We also need to be pretty hard-nosed about quickly bouncing anything
that's not realistically going to be able to get committed this month.
I've not gone through the list in detail, but there are at least several
waiting-on-author items that have seen no activity since the last 'fest.
I'd recommend marking those RWF pretty soon.

            regards, tom lane

On Thu, Mar 01, 2018 at 12:42:25AM -0500, Tom Lane wrote:
> I think that we should summarily bounce to the September 'fest anything
> submitted in the last two days; certainly anything that's nontrivial.

I don't have any issues with simple patches posted at the last minute.
There is for example a small one for pg_rewind I was looking at a bit
today which is IMO fine to deal with for this CF.

> that are not Johnny-come-latelies.

Reference? :)

> We also need to be pretty hard-nosed about quickly bouncing anything
> that's not realistically going to be able to get committed this month.
> I've not gone through the list in detail, but there are at least several
> waiting-on-author items that have seen no activity since the last 'fest.
> I'd recommend marking those RWF pretty soon.

Agreed.
--
Michael

Hi,

On 2018-03-01 10:52:12 +0100, Magnus Hagander wrote:
>  2018-02-28 17:21:13.384903+00 | JIT compiling expressions & tuple deforming

Note this is an old thread, just a new CF entry.

> > There is no way that we can possibly handle 200+ CF entries in a month.
> > A large fraction of these are going to end up pushed to September no
> > matter what, and I think simple fairness demands that we spend our
> > time on the ones that are not Johnny-come-latelies.
> >
> 
> +1.

Same.  I'm making a pass through the CF right now (and will finish it
tomorrow).

Greetings,

Andres Freund

On 3/1/18 4:52 AM, Magnus Hagander wrote:
> On Thu, Mar 1, 2018 at 6:42 AM, Tom Lane <tgl@sss.pgh.pa.us
> <mailto:tgl@sss.pgh.pa.us>> wrote:
> 
>     I think that we should summarily bounce to the September 'fest anything
>     submitted in the last two days; certainly anything that's nontrivial.
> 
> 
> So FYI,  it's now 250. Of those, 52 entered into this commitfest on or
> after 2018-02-27. 4 of those have already been committed, 2 have been
> bumped to the next CF already,and 1 have been returned with feedback.
> The remaining ones are:
> 
> The remaining 45 are, ordered by time created:

I'll run through these this morning and bump anything that looks
non-trivial.

-- 
-David
david@pgmasters.net

Michael Paquier wrote:
> On Thu, Mar 01, 2018 at 12:42:25AM -0500, Tom Lane wrote:

> > that are not Johnny-come-latelies.
> 
> Reference? :)

This is a relatively recent term, dating back only two centuries ago
according to M-W:
https://www.merriam-webster.com/dictionary/Johnny-come-lately


Also, it's already been used in the lists a few times.  One example:
https://www.postgresql.org/message-id/18196.1303965884@sss.pgh.pa.us

(There's also an Eagles song, of course.)

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

On 3/1/18 4:52 AM, Magnus Hagander wrote:
> 
> The remaining 45 are, ordered by time created:

OK, here's my break down after a brief review of all the patches in the
list.  Many of the patches were trivial/small, bug fixes, had history,
or related to builds, tests, or docs.

There are still 13 patches that are medium/large (in terms of impact,
not size) from my analysis, and new.

Medium/Large and New:

Verify Checksums during Basebackups
CALL optional in PL/pgSQL
chained transactions
INOUT parameters in procedures
SET TRANSACTION in PL/pgSQL
PL/pgSQL nested CALL with transactions
committing inside cursor loop
ON CONFLICT DO UPDATE for partitioned tables
handling of heap rewrites in logical decoding
Changing the autovacuum launcher scheduling; oldest table first algorithm
Let's remove DSM_INPL_NONE?
[PROPOSAL] Nepali Snowball dictionary?
Opclass parameters (recommend RWF or push by me)
Parallel Dump to /dev/null (recommend push by Andres)
Reduce amount of WAL generated by CREATE INDEX for gist, gin and sp-gist
(already moved to next CF by me)

This one is large, has history, but had been idle for a long time:

kNN for SP-GiST (last submitted for 2017-03 CF, RWF, then idle for one
year until this CF)

Was already rejected:
Direct converting numeric types to bool

The following patches seem OK to me, but I'm happy for feedback.  Note
that one of the small patches is mine -- I think it's small because it
shares a lot of code with a patch that is already marked RFC, but I
won't object if the consensus is to push it.

Trivial / Build / Tests / Docs:

Unlogged tables re-initialization tests
Comment of formdesc() forgets pg_subscription
Minor clean-up in dshash.{c,h}
Align isolation tester output
Updating parallel.sgml's treatment of parallel joins
symlink installs

Bugs:

Fix "unsupported type" with custom numeric types in convert_to_scalar
Standby cannot catch up primary after pg_rewind
line_perp() (?-|) is broken.
Correctly inhibit indexonly when multiple occurrence of an index column
ActivePerl 5.24.3 breaks PG compilation in Windows
ConvertRowtypeExpr reference errors from partition-wise join
Incorrect flag passed to initial_cost_hashjoin()
fixing more format truncation issues (this is under refactoring but
reads a bit more like bugs fixes to me)

Active history and review, just newly added to CF:

Cast jsonb to numeric, int, float, bool
Predicate locking in gin index
JIT compiling expressions & tuple deforming
Transactions involving multiple postgres foreign servers
Advanced partition matching for partition-wise join
Minor clean-up in dshash.{c,h}

Small and New:

Nested ConvertRowtypeExpr optimization
Reopen logfile on SIGHUP
Function to track shmem reinit time
Removing shm_mq.c's volatile qualifiers
Exclude temp relations from base backup (This is my patch so I prefer
somebody else decide its fate)
remove pg_class.relhaspkey

Ready for Committer:

pg_proc.prokind column
Increase initdb's fallback value of max_connection to 20

Feedback?

Thanks,
-- 
-David
david@pgmasters.net

David Steele <david@pgmasters.net> writes:
> OK, here's my break down after a brief review of all the patches in the
> list.  Many of the patches were trivial/small, bug fixes, had history,
> or related to builds, tests, or docs.

> There are still 13 patches that are medium/large (in terms of impact,
> not size) from my analysis, and new.

FWIW, several of these are part of Peter's continuing work on procedures,
and thus arguably not totally new:

> chained transactions
> INOUT parameters in procedures
> SET TRANSACTION in PL/pgSQL
> PL/pgSQL nested CALL with transactions
> committing inside cursor loop

So I'd be inclined to cut these a little slack in hopes of having a more
complete/coherent procedure feature in v11.  Still, they are showing up
late in the dev cycle, so they shouldn't get a whole lot of slack.

> Changing the autovacuum launcher scheduling; oldest table first algorithm

Definitely too late for this cycle.

> Let's remove DSM_INPL_NONE?

This discussion has been ongoing for awhile, and it does have some of the
nature of a bug fix since if we don't do this we'll instead have to deal
with bugs in the places that fail to cope.

> [PROPOSAL] Nepali Snowball dictionary?
> Opclass parameters (recommend RWF or push by me)
> Parallel Dump to /dev/null (recommend push by Andres)

Push or RWF all of these, IMO.

> The following patches seem OK to me, but I'm happy for feedback.

I'm okay with leaving these alive for now.  Some of 'em perhaps should
fail on merit, but that is a case-by-case issue rather than something
to handle through this discussion.

            regards, tom lane

Hi,

On 2018-03-01 00:42:25 -0500, Tom Lane wrote:
> As of right now, there are 229 entries in this commitfest (not counting
> the items already committed or RWF).
> 
> It's hard for me to tell for sure, because the activity log page doesn't
> stretch back far enough, but I think about 40 of those have been created
> in the last 24 hours.  It's certainly well over 20, because the log does
> go back far enough to show 21 patch records created since about noon
> Wednesday UTC.
> 
> This is NOT how it is supposed to work.  This is gaming the system.

Strongly agreed. There's a lot of stuff in here that has really no place
being there.


> I think that we should summarily bounce to the September 'fest anything
> submitted in the last two days; certainly anything that's nontrivial.

I've now gone through all the patches and commented on each of them in a
summary http://archives.postgresql.org/message-id/20180302075242.yfqkcgzbrmjysboa%40alap3.anarazel.de
I've sent comments to many of the ones that were submitted very
late.

I personally don't see much of a difference between submitting a brand
new patch 2018-02-15 and 2018-02-25. There's rarely a meaningful
difference in the amount of review features have gotten.


> There is no way that we can possibly handle 200+ CF entries in a month.
> A large fraction of these are going to end up pushed to September no
> matter what, and I think simple fairness demands that we spend our
> time on the ones that are not Johnny-come-latelies.

A surprisingly large part of those are ones that haven't meaningfully
evolved between the last few CFs.  I think we need to used 'moved to
next CF' a lot less frequently.


> We also need to be pretty hard-nosed about quickly bouncing anything
> that's not realistically going to be able to get committed this month.
> I've not gone through the list in detail, but there are at least several
> waiting-on-author items that have seen no activity since the last 'fest.
> I'd recommend marking those RWF pretty soon.

I've done so, or proposed doing so, for a lot of them.


Having gone through all patches and having at least opened each of the
~230 open items, one thing that makes me extremely unhappy is that
there's a *massive* imbalance around reviews.  There's a few folks that
do a lot of reviews in a lot of different areas, and there's others that
have lot and lots of huge patches open but don't do much to help others.
I don't think that's ok.


Greetings,

Andres Freund