Обсуждение: Remaining 'needs review' patchs in July commitfest

21 patches remain in Needs Review state, in the July commitfest. Some of 
them have a reviewer signed up. I have highlighted some of them below 
that worry me the most. What are we going to do about these? For each of 
them, I'd like the authors to have some idea on what they need to do to 
get the patch into committable state (or if the whole approach is going 
to be rejected), but I don't know what that advise should be.

> pgbench - allow backslash-continuations in custom scripts

Everyone wants the feature, using multi-line SELECTs in pgbench scripts, 
but we don't seem to be reaching a consensus on how it should work. I 
think we'll need to integrate the lexer, but it would be nice to still 
support multi-statements as well, with some syntax.

> multivariate statistics

This has been a long discussion. Are we getting close to a committable 
state?

> COPY RAW

No consensus on whether to add this to the server's COPY command, or as 
a new psql backslash option.

> Unique Joins

Kyotaro HORIGUCHI commented on this back in March, but no-one's reviewed 
the latest version. It's a fairly big patch. I guess I'll review it if 
no-one else does, but it's going to take me some time to really dig into 
it...

> checkpoint continuous flushing

This does a big memory allocation at checkpoint, which Tom vehemently 
objects to. I don't much like it either, although I would be OK with a 
more moderately-sized allocation. It's not clear on what criteria this 
should be accepted or rejected. What workloads need to be tested?

> plpgsql raise statement with context

Impasse. Everyone wants this feature in some form, but no consensus on 
whether to do this client-side or server-side.

> Configurable location for extension .control files

Do we want this? In its current form? I feel that this isn't ready as it 
is, but I'm not sure what to suggest instead.

> dblink: add polymorphic result functions

Seems pretty ugly to me to add a dummy argument to functions, just so 
that you can specify the result type. The problem it's trying to solve 
is real, though. Should we take it as it is, or wait for some cleaner 
approach?

> Improving test coverage of extensions with pg_dump

Do we want to have this in src/test/modules or src/bin/pg_dump/t?

> Asynchronous execution on postgres-fdw

If we're going to have some sort of general-purpose Parallel node 
support soon, should we just forget about this? Or is it still useful? 
This adds a fair amount of new infrastructure, for a fairly niche feature..

- Heikki

-- 

- Heikki

On 07/28/2015 12:51 PM, Heikki Linnakangas wrote:
> Everyone wants the feature, using multi-line SELECTs in pgbench scripts,
> but we don't seem to be reaching a consensus on how it should work. I
> think we'll need to integrate the lexer, but it would be nice to still
> support multi-statements as well, with some syntax.

Seems like a perfect candidate for "returned with feedback".  We'll
straighten it out by the next commitfest.

Also note that this has actually spiralled out into 2 or 3 separate
features.


-- 
Josh Berkus
PostgreSQL Experts Inc.
http://pgexperts.com

On 2015-07-28 22:51:55 +0300, Heikki Linnakangas wrote:
> >checkpoint continuous flushing
> 
> This does a big memory allocation at checkpoint, which Tom vehemently
> objects to.

Uh. Didn't he just object to failing in that case? IIRC he even
indicated tentative assent, a year or so back, with my idea of just
pre-allocating all the required memory in shared memory, so that we
don't need to allocate anything at some point.

I've not read the last version of the patch, but in my old version the
allocation wasn't actually that large in comparison to the size of
shared buffers itself.


> I don't much like it either, although I would be OK with a more
> moderately-sized allocation.

That'll disallow some mighty fine optimizations like e.g. being able to
do the fsyncs of files early, directly after we wrote out all their
buffer, thereby reducing how much dirty data (written out by backend)
that needs to be flushed to disk.


Greetings,

Andres Freund

Heikki Linnakangas wrote:
> 21 patches remain in Needs Review state, in the July commitfest. Some of
> them have a reviewer signed up. I have highlighted some of them below that
> worry me the most. What are we going to do about these? For each of them,
> I'd like the authors to have some idea on what they need to do to get the
> patch into committable state (or if the whole approach is going to be
> rejected), but I don't know what that advise should be.
> 
> >pgbench - allow backslash-continuations in custom scripts
> 
> Everyone wants the feature, using multi-line SELECTs in pgbench scripts, but
> we don't seem to be reaching a consensus on how it should work. I think
> we'll need to integrate the lexer, but it would be nice to still support
> multi-statements as well, with some syntax.

Excuse me -- what's a multi-statement?  I thought this effort was all
about multi-line single statements only.  I think the proposed patch to
integrate psql's lexer in pgbench is a reasonable step forward, but we
need it to use our standard technique of symlinking the .l file in place
via some additional lines in pgbench's Makefile rather than copying it.

> >dblink: add polymorphic result functions
> 
> Seems pretty ugly to me to add a dummy argument to functions, just so that
> you can specify the result type. The problem it's trying to solve is real,
> though. Should we take it as it is, or wait for some cleaner approach?

Put like that, it does sound quite ugly.  I take it we have no better
alternative proposed?

> >Improving test coverage of extensions with pg_dump
> 
> Do we want to have this in src/test/modules or src/bin/pg_dump/t?

Are we testing pg_dump here, or are we testing extensions?  If the
former, src/bin/pg_dump/t seems best.

> >Asynchronous execution on postgres-fdw
> 
> If we're going to have some sort of general-purpose Parallel node support
> soon, should we just forget about this? Or is it still useful? This adds a
> fair amount of new infrastructure, for a fairly niche feature..

-0.1 on this one.

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

On 07/28/2015 11:01 PM, Alvaro Herrera wrote:
> Heikki Linnakangas wrote:
>>> pgbench - allow backslash-continuations in custom scripts
>>
>> Everyone wants the feature, using multi-line SELECTs in pgbench scripts, but
>> we don't seem to be reaching a consensus on how it should work. I think
>> we'll need to integrate the lexer, but it would be nice to still support
>> multi-statements as well, with some syntax.
>
> Excuse me -- what's a multi-statement?

Sending "SELECT 1; SELECT 2;" to the server in one Query message. I.e. 
PQexec(conn, "SELECT 1; SELECT 2;").

If you put that on a single line in pgbench today, it will send it as 
one message to the server. If we start using a lexer, and split that 
into two, that's a change in behaviour. Not the end of the world, but it 
would still be nice to be able to use multi-statements in pgbench, too.

>>> dblink: add polymorphic result functions
>>
>> Seems pretty ugly to me to add a dummy argument to functions, just so that
>> you can specify the result type. The problem it's trying to solve is real,
>> though. Should we take it as it is, or wait for some cleaner approach?
>
> Put like that, it does sound quite ugly.  I take it we have no better
> alternative proposed?

Joe Conway suggested a more generic approach here: 
http://www.postgresql.org/message-id/559A9643.9070409@joeconway.com. I'm 
not sure why that was not pursued. It certainly seems better to me.

>>> Improving test coverage of extensions with pg_dump
>>
>> Do we want to have this in src/test/modules or src/bin/pg_dump/t?
>
> Are we testing pg_dump here, or are we testing extensions?  If the
> former, src/bin/pg_dump/t seems best.

I think that's the crux of the disagreement :-).

- Heikki

On 07/28/2015 11:30 PM, Simon Riggs wrote:
> I've added my name as committer to a few things, but won't be able to work
> on them until at least next week when I've finished 9.5 stuff. Happy to
> step back if anyone else wants to claim those.

Thanks, every little helps!

- Heikki

Andres Freund <andres@anarazel.de> writes:
> On 2015-07-28 22:51:55 +0300, Heikki Linnakangas wrote:
>>> checkpoint continuous flushing

>> This does a big memory allocation at checkpoint, which Tom vehemently
>> objects to.

> Uh. Didn't he just object to failing in that case?

Right.  If it can fall back to "stupid" mode when there's not spare
memory, I'd be ok with that.
        regards, tom lane

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 07/28/2015 01:22 PM, Heikki Linnakangas wrote:
>>>> dblink: add polymorphic result functions
>>> 
>>> Seems pretty ugly to me to add a dummy argument to functions,
>>> just so that you can specify the result type. The problem it's
>>> trying to solve is real, though. Should we take it as it is, or
>>> wait for some cleaner approach?
>> 
>> Put like that, it does sound quite ugly.  I take it we have no
>> better alternative proposed?
> 
> Joe Conway suggested a more generic approach here: 
> http://www.postgresql.org/message-id/559A9643.9070409@joeconway.com.
> I'm not sure why that was not pursued. It certainly seems better to
> me.

I really think the second approach in that email is the right way to
go. Is there any benefit we get from the dblink specific hack that we
don't get via the cast?

I guess it is just be a matter of someone getting around to making it
happen. I can put it on my list of things to do, but probably will not
get to looking at it for at least a few more weeks.

- -- 
Joe Conway
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.22 (GNU/Linux)

iQIcBAEBAgAGBQJVt+xpAAoJEDfy90M199hlIM8P/iRq3nZo6Q/u14D9Vqvd8qZ1
cPaBgGoHvCCuM3rS/z0DobBMjKll0oEerjXeuBKTjWBwilPrdHMZahueFZ1896Mh
atcj7kz8O3nBj27sBAHI1tXCTPRkjOIpt6nNu8BEsG8mUmX8wC2odfyGFZIKzw2j
dtLGsczSMwrXnE1yTzgCYWVl0ej8y/CtAmL6uEQOIA5XDHmYe6EqqPd3rcNBNg+I
endHrAXGm/VVsKGUJfisWItVbPJy4VBx1y1iXfApxePceHjW6/U0Ivds5Qu+K3wd
0pwsBmqlHNBGzwril1jBpm9gecIYY6916Hbru7Fafae4bbFHRTDw0ExURybrlhGo
MjOhWmfpsCutGmBktgWxqiOY9UDnG1+aXREKIv7bIavPhBYbBRkvqT7qrAKureex
y2XZz86tMV2bUZxRyp6tFo8aL5hheV7UoaPKGlOxfgUNFv3U2uRUopVPH6uR1Z80
+JQoC2hDadI3wQ6/2egL62pQR3l/Ts8zF0sCFEGAZVrFklEneHsUS7PAbqIfY+TI
Ffc5q8Vm0QplWXr5vvxL1JKepXGgKSidMEnylAwIxKAnPtXyosPe1sZ+ocA+3gbP
Uj/bElWemX8b8YF/F1mecHqoLAvJ+3NKUdwS805VmdCLO1nxictKEOHAJFeQomif
OQquxdkcGtA9dmCt8JJ9
=SaUZ
-----END PGP SIGNATURE-----

Hi,

On 07/28/2015 09:51 PM, Heikki Linnakangas wrote:
>
>> multivariate statistics
>
> This has been a long discussion. Are we getting close to a committable
> state?

Certainly not - the discussion may seem long, but it only deals with 
some aspects of the patch so far. There was very little discussion of 
low-level implementation details (e.g. how exactly the histograms are 
built, data structures etc.) - we haven't got that far yet.


--
Tomas Vondra                  http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

On Wed, Jul 29, 2015 at 5:22 AM, Heikki Linnakangas wrote:
> On 07/28/2015 11:01 PM, Alvaro Herrera wrote:
>>>> Improving test coverage of extensions with pg_dump
>>>
>>>
>>> Do we want to have this in src/test/modules or src/bin/pg_dump/t?
>>
>>
>> Are we testing pg_dump here, or are we testing extensions?  If the
>> former, src/bin/pg_dump/t seems best.

All the tests are using pg_dump, but it is testing dumpable tables in
an extension. At this point I am not sure which one is better honestly
X/. Putting it in pg_dump/t will require two lines in the make target
prove_check such as modules in this path are installed as well, and
the argument against having it in src/test/modules is that it would
bloat it in the future if we do the same for all binaries,
particularly if we have multiple modules for each one.

> I think that's the crux of the disagreement :-).

Yep.
-- 
Michael

Michael Paquier <michael.paquier@gmail.com> writes:
> On Wed, Jul 29, 2015 at 5:22 AM, Heikki Linnakangas wrote:
>> On 07/28/2015 11:01 PM, Alvaro Herrera wrote:
>>> Do we want to have this in src/test/modules or src/bin/pg_dump/t?

>> Are we testing pg_dump here, or are we testing extensions?  If the
>> former, src/bin/pg_dump/t seems best.

> All the tests are using pg_dump, but it is testing dumpable tables in
> an extension. At this point I am not sure which one is better honestly
> X/.

ISTM we're testing pg_dump, but I don't especially want to clutter
the pg_dump directory with the code for a dummy extension, so I'd
vote for putting the extension code under src/test/modules.
(Or you could hide it inside src/bin/pg_dump/t/, but that seems
pretty weird too.)

In the end though, this is the sort of thing we generally leave to the
discretion of the responsible committer.  I won't whinge too hard
about whatever the decision is ...
        regards, tom lane

Hello, Heikki.

You wrote:

HL> 21 patches remain in Needs Review state, in the July commitfest. Some of
HL> them have a reviewer signed up. I have highlighted some of them below 
HL> that worry me the most. What are we going to do about these? For each of
HL> them, I'd like the authors to have some idea on what they need to do to
HL> get the patch into committable state (or if the whole approach is going
HL> to be rejected), but I don't know what that advise should be.

>> COPY RAW

HL> No consensus on whether to add this to the server's COPY command, or as
HL> a new psql backslash option.

I did quick review for this, however I need some more time for tests. May be I will
do it next week. There is a consensus, only detailed review needed.


HL> -- 

HL> - Heikki





-- 
With best wishes,Pavel                          mailto:pavel@gf.microolap.com

Hello Heikki,

About two patches I submitted:


>> pgbench - allow backslash-continuations in custom scripts
>
> Everyone wants the feature, using multi-line SELECTs in pgbench scripts, 
> but we don't seem to be reaching a consensus on how it should work. I 
> think we'll need to integrate the lexer, but it would be nice to still 
> support multi-statements as well, with some syntax.

I was willing to do a small job of that one, but it drifted to solving the 
meta problem of pgbench lexing for doing it "the right way".  Kyotaro 
HORIGUCHI has taken up the challenge and submitted a 3-part patch to 
modify psql lexer, then reuse it "as is" in pgbench.  I'm impressed:-)

I think that the approach is a little overkill for the simple targetted 
feature, and I'm not sure that it allows for multi-statements, but I have 
not checked.  Anyway I may try to review it, although not in the short 
term.  I'm not too optimistic because if passed, it means that the psql 
lexer would be used in 2 contexts, so changes in any of them would require 
ensuring that it does not break anything in the other one, and I'm not 
sure that it is such constraint is desirable.  Duplicating the lexer is 
not a realistic option either, I do not think that pgbench is worth it.

Anyway, in the mean time, the patch may be switched to "returned with 
feedback" or "rejected".



>> checkpoint continuous flushing
>
> This does a big memory allocation at checkpoint, which Tom vehemently 
> objects to. I don't much like it either, although I would be OK with a 
> more moderately-sized allocation.

AFAICS Tom has not expressed any view on the patch in its current form 
(there is no message from Tom on the thread).


ISTM that Tom complained about a year ago about OOM risks on a 2007 
version of the sorting part by Takahiro ITAGAKI which was dynamically 
allocating and freeing about 24 bytes per buffers on each checkpoint.

The current v5 version uses 4 bytes per buffer at the first run and reuse 
the memory, it is allocated once and thus there is no risk of returning it 
and failing to get it back, so no significant OOM risk. Maybe the 
allocation should be moved earlier when starting the checkpointer process, 
though.

A second objection a year ago from Tom was about proof of performance 
gains. I've spent quite some time to collect a lot of data to measure 
benefits under different loads, representing X00 hours of pgbench runs, as 
can be seen in the thread. ISTM that the performance (tps & latency) 
figures, for instance:

http://www.postgresql.org/message-id/raw/alpine.DEB.2.10.1506170803210.9794@sto

are overwhelmingly in favor of the patch.


If the memory requirement of 4 bytes per buffer is still too much, it is 
eventually possible to reduce it with a guc to specify the allowed amount 
of memory and some shuffling in the code to do things by chunk.

My opinion is that 4 bytes per buffer is reasonable enough given the 
measured benefits. Also, there is more benefits if the whole checkpoint is 
sorted instead of just part of it.

I'm really willing to improve the write stall issues which freezes 
postgresql when checkpointing on HDD, so if it has to be chunked because 
this is a blocker I'll make the effort, but I do not think that it is 
useful as such.


> It's not clear on what criteria this should be accepted or rejected.

Given the overall performance gains and reduction in latency, where a lot 
of write stalls are avoided or at least greatly reduced, I would be sad 
for pg to get it rejected. That does not mean that it cannot be improved.

> What workloads need to be tested?

If you tell me, and provide the matching dedicated host for testing, I can 
run tests...

-- 
Fabien.

On Tue, Jul 28, 2015 at 3:51 PM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:
>> plpgsql raise statement with context
>
> Impasse. Everyone wants this feature in some form, but no consensus on
> whether to do this client-side or server-side.

+1 for server-side.  Does anyone other than you even think that the
client side is a reasonable way to go?

>> dblink: add polymorphic result functions
>
>
> Seems pretty ugly to me to add a dummy argument to functions, just so that
> you can specify the result type. The problem it's trying to solve is real,
> though. Should we take it as it is, or wait for some cleaner approach?

+1 for take it.  If we wait for something better, we may be waiting a long time.

>> Asynchronous execution on postgres-fdw
>
> If we're going to have some sort of general-purpose Parallel node support
> soon, should we just forget about this? Or is it still useful? This adds a
> fair amount of new infrastructure, for a fairly niche feature..

IMHO, that's an extremely important feature.  I believe that it will
not be obsoleted by other parallelism work.  Actually, I think that
some of the infrastructure that it introduces may be quite helpful for
parallelism as well, but I haven't looked at in detail yet.  The core
design concern that I have is what to do about this:

Append
-> Scan
-> Scan
-> Scan

Right now, we start each scan when the previous scan finishes.  But
what if - either through parallelism or through this patch - we could
launch scans 2, 3, etc. before scan 1 completes?  If there's a Limit
node above this somewhere we may regret our decision bitterly.  But if
there isn't, we may do an enormous amount of good by starting that
plan early on the speculation that we will in fact need the results
eventually.

I haven't applied a whole lot of brainpower to this problem yet; I'm
sort of hoping someone else will have a good idea.  Kyotaro
Horiguchi's approach seems to be to say that we don't need to solve
this problem at this stage and that we should just not worry about the
possible waste of resources on the remote machine for now; fix it
later, as part of a future patch.   I'm not terribly confident in that
approach, but I don't currently have anything better to propose.

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

Robert Haas <robertmhaas@gmail.com> writes:
> On Tue, Jul 28, 2015 at 3:51 PM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:
>>> plpgsql raise statement with context
>> Impasse. Everyone wants this feature in some form, but no consensus on
>> whether to do this client-side or server-side.

> +1 for server-side.  Does anyone other than you even think that the
> client side is a reasonable way to go?

Yes.  This is presupposing on the server side what the client will want
to display.

>>> dblink: add polymorphic result functions
>> Seems pretty ugly to me to add a dummy argument to functions, just so that
>> you can specify the result type. The problem it's trying to solve is real,
>> though. Should we take it as it is, or wait for some cleaner approach?

> +1 for take it.  If we wait for something better, we may be waiting a long time.

-1.  We have a clear design spec for a solution that fixes this for much
more than just dblink.  I don't feel a need to paste on an ugly, single-use
band-aid in such a hurry.  If we get close to 9.6 and there is no better
fix, we could reconsider; but now is not the time to give up on pursuing
the better solution.

>>> Asynchronous execution on postgres-fdw
>> If we're going to have some sort of general-purpose Parallel node support
>> soon, should we just forget about this? Or is it still useful? This adds a
>> fair amount of new infrastructure, for a fairly niche feature..

> IMHO, that's an extremely important feature.  I believe that it will
> not be obsoleted by other parallelism work.

Again, this seems like rushing through a single-use band-aid while work
is still active on more general solutions.  I think we should put this
off and see whether it makes sense *after* the general parallel query
stuff is (more nearly) functional.  Even if it still makes sense then,
it may well need a rewrite.

> Actually, I think that
> some of the infrastructure that it introduces may be quite helpful for
> parallelism as well, but I haven't looked at in detail yet.

Then steal that stuff as and when you need it.
        regards, tom lane

On Wed, Jul 29, 2015 at 5:08 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Robert Haas <robertmhaas@gmail.com> writes:
>> On Tue, Jul 28, 2015 at 3:51 PM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:
>>>> plpgsql raise statement with context
>>> Impasse. Everyone wants this feature in some form, but no consensus on
>>> whether to do this client-side or server-side.
>
>> +1 for server-side.  Does anyone other than you even think that the
>> client side is a reasonable way to go?
>
> Yes.  This is presupposing on the server side what the client will want
> to display.

Fair enough.  I'm still not convinced we're doing anything other than
complicating what ought to be a simple matter.  It is just a fact that
logging tracing messages in PL/pgsql functions is a pain in the butt
right now in some situations because you get a huge number of CONTEXT
lines that you don't want.  Can we agree on some solution to that
problem without over-engineering this to infinity?

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