Обсуждение: partitioning using dblink

Scara Maccai wrote:

> I started thinking that using dblink I could "easily" get some kind of
> read only multi-server partitioning, if only VIEWs could be declared
> with "INHERITS"...

I think you can do pretty much the same thing with PL/Proxy; see
https://developer.skype.com/SkypeGarage/DbProjects/PlProxy

--
Alvaro Herrera                                http://www.CommandPrompt.com/
The PostgreSQL Company - Command Prompt, Inc.

Alvaro Herrera wrote:

> I think you can do pretty much the same thing with
PL/Proxy; see
>
https://developer.skype.com/SkypeGarage/DbProjects/PlProxy

Mmmh, I actually looked into that but I thought it
only worked with user functions...
am I wrong?

What I'd like to have is an almost-transparent
horizontal partitioning system, and
I think that everything is there: postgresql
partitioning (which even has partition pruning)
 + dblink should be enough...

it's only that you can't use them together, because
with dblink you should use
VIEWs but partitioning can't work with those (which I
find correct, BTW).
What I would like is for Postgresql to know that a
table is actually a remote table...

BUT!!!
since "Views in PostgreSQL are implemented using the
rule system", I could do
(pseudo-sql)

CREATE TABLE mypartion HINERITS blabla CHECK
CONSTR..[...] ;

CREATE RULE "_RETURN" AS ON SELECT TO myview DO
INSTEAD
    SELECT * FROM dblink;

Can't I??? That would be horizontal partitioning using
dblink+postgresql!!!

Am I wrong???


      ___________________________________
L'email della prossima generazione? Puoi averla con la nuova Yahoo! Mail: http://it.docs.yahoo.com/nowyoucan.html

I can't get views to participate in the hierarchy...

create table outings1 as select * from outings_root limit 0;

alter table outings1 inherit outings_root;
SELECT *
   FROM dblink('host=myhost dbname=tacche port=5433 user=postgres password=postgres'::text,
        'SELECT * from outings1'::text)  as
    (id integer, date date, spot_id integer, notes text);


Selects to outings_root won't show data from host "myhost". Selects from outings1 will work as expected.

This is not what I got from the docs:

"the information about a view in the     PostgreSQL
system catalogs is exactly the same as it is for a table. So for the
parser, there is absolutely no difference between a table and a view"



Am I doing something wrong?

If it worked I would have a very basic (but working!) form of horizontal partitioning....





      ___________________________________
L'email della prossima generazione? Puoi averla con la nuova Yahoo! Mail: http://it.docs.yahoo.com/nowyoucan.html

On 2/29/08, Scara Maccai <m_lists@yahoo.it> wrote:
> I can't get views to participate in the hierarchy...

The partition exclusion _may_ work if you do something like:

create view as
  select * from dblink/plproxy-from-part1 where part1 constraint
  union all
  select * from dblink/plproxy-from-part2 where part2 constraint

So if you do 'select * from view where constraint;' the postgres
will skip partitions which do not match.  You may need to
define the setof function immutable or something... i'm not sure.

But any contraint exclusion wont change the fact you are
doing select * from tbl; in remote db, which makes the
exercise quite pointess IMHO.

I obviously would recommend pl/proxy for such task, but that
would expect you are able to write your queries.  If you are
in situation where you don't control the queries, then plproxy
quite likely is not use.

--
marko

I'm sorry, I didn't understand you post...

1) Why does my current implementation is not working? Hierarchy doesn't work with views in general, not only with
dblink
2) Why am I supposed to use unions in the view?
3) I know that I am doing select * from tbl in the remote db; that is something I can work on later.
At least I would like to see it working, since there is nothing in the docs that says it shouldn't be working...
4) I am not able to rewrite my queries.

----- Messaggio originale -----
Da: Marko Kreen <markokr@gmail.com>
A: Scara Maccai <m_lists@yahoo.it>
Cc: pgsql-general@postgresql.org
Inviato: Venerdì 29 febbraio 2008, 10:46:09
Oggetto: Re: [GENERAL] partitioning using dblink

On 2/29/08, Scara Maccai <m_lists@yahoo.it> wrote:
> I can't get views to participate in the hierarchy...

The partition exclusion _may_ work if you do something like:

create view as
  select * from dblink/plproxy-from-part1 where part1 constraint
  union all
  select * from dblink/plproxy-from-part2 where part2 constraint

So if you do 'select * from view where constraint;' the postgres
will skip partitions which do not match.  You may need to
define the setof function immutable or something... i'm not sure.

But any contraint exclusion wont change the fact you are
doing select * from tbl; in remote db, which makes the
exercise quite pointess IMHO.

I obviously would recommend pl/proxy for such task, but that
would expect you are able to write your queries.  If you are
in situation where you don't control the queries, then plproxy
quite likely is not use.

--
marko





      ___________________________________
L'email della prossima generazione? Puoi averla con la nuova Yahoo! Mail:
http://it.docs.yahoo.com/nowyoucan.html

On 2/29/08, Scara Maccai <m_lists@yahoo.it> wrote:
> I'm sorry, I didn't understand you post...
>
>  1) Why does my current implementation is not working? Hierarchy doesn't work with views in general, not only with
dblink

Exactly, because inheritance/constraint exclusion wont work with views.

>  2) Why am I supposed to use unions in the view?

So that query evaluator can exclude unnecessary partitions.

Given view:

  create view.. as
    select * from blah() where id >= 0 and id < 10000
    union all
    select * from blah() where id >= 10000 and id < 20000
    ...

Then running query

  select * from view where id = 10;

can skip partitions by simply examining where expression.

Quite likely you need to tune it for your case.

>  3) I know that I am doing select * from tbl in the remote db; that is something I can work on later.
>  At least I would like to see it working, since there is nothing in the docs that says it shouldn't be working...
>  4) I am not able to rewrite my queries.

Have fun then.

--
marko

> Exactly, because inheritance/constraint exclusion wont work with views.

Ok, so there should be something written in the docs about it...
From:

"the information about a view in the     PostgreSQL
system catalogs is exactly the same as it is for a table. So for the
parser, there is absolutely no difference between a table and a view"

I got that there should be no difference... plus, I don't get any errors, it's only that data in the view doesn't show
upwhen I query the master table. 

> >  2) Why am I supposed to use unions in the view?
>
> So that query evaluator can exclude unnecessary partitions.

Ok: that would be another way of having partitions, right?

> >  3) I know that I am doing select * from tbl in the remote db; that is
> something I can work on later.
> >  At least I would like to see it working, since there is nothing in the docs
> that says it shouldn't be working...
> >  4) I am not able to rewrite my queries.
>
> Have fun then.


Yeah I know...


Thank you






      ___________________________________
L'email della prossima generazione? Puoi averla con la nuova Yahoo! Mail:
http://it.docs.yahoo.com/nowyoucan.html

On 2/29/08, Scara Maccai <m_lists@yahoo.it> wrote:
> > Exactly, because inheritance/constraint exclusion wont work with views.
>
> Ok, so there should be something written in the docs about it...
>  From:
>
>  "the information about a view in the     PostgreSQL
>  system catalogs is exactly the same as it is for a table. So for the
>  parser, there is absolutely no difference between a table and a view"
>
>
> I got that there should be no difference... plus, I don't get any errors, it's only that data in the view doesn't
showup when I query the master table. 

Seems like bug in docs/code.  But I think the paragraph is written
with "in queries, views can be used anywhere tables can" in mind,
not that you can administer them the same way.

>  > >  2) Why am I supposed to use unions in the view?
>  >
>  > So that query evaluator can exclude unnecessary partitions.
>
> Ok: that would be another way of having partitions, right?

Yes, effect should be same.

--
marko

Scara Maccai <m_lists@yahoo.it> writes:
> I got that there should be no difference... plus, I don't get any
> errors,

You should have.  The system enforces (or tries to) that a view can't be
part of an inheritance hierarchy, but you seem to have managed to find a
sequence of operations that avoids those checks.  Turning a table into a
view with a manual CREATE RULE operation has always been a kluge, and
it's missing a check that the table isn't part of an inheritance tree.

            regards, tom lane

Tom Lane wrote:
> Scara Maccai <m_lists@yahoo.it> writes:
> > I got that there should be no difference... plus, I don't get any
> > errors,
>
> You should have.  The system enforces (or tries to) that a view can't be
> part of an inheritance hierarchy, but you seem to have managed to find a
> sequence of operations that avoids those checks.  Turning a table into a
> view with a manual CREATE RULE operation has always been a kluge, and
> it's missing a check that the table isn't part of an inheritance tree.

Is this a TODO?  Seems so.

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

  + If your life is a hard drive, Christ can be your backup. +

Bruce Momjian <bruce@momjian.us> writes:
> Tom Lane wrote:
>> Scara Maccai <m_lists@yahoo.it> writes:
>>> I got that there should be no difference... plus, I don't get any
>>> errors,
>>
>> You should have.  The system enforces (or tries to) that a view can't be
>> part of an inheritance hierarchy, but you seem to have managed to find a
>> sequence of operations that avoids those checks.  Turning a table into a
>> view with a manual CREATE RULE operation has always been a kluge, and
>> it's missing a check that the table isn't part of an inheritance tree.

> Is this a TODO?  Seems so.

I think it's just a minor bugfix, but if you want to put it in TODO for
a day or two, go ahead...

            regards, tom lane

Tom Lane wrote:
> Bruce Momjian <bruce@momjian.us> writes:
> > Tom Lane wrote:
> >> Scara Maccai <m_lists@yahoo.it> writes:
> >>> I got that there should be no difference... plus, I don't get any
> >>> errors,
> >>
> >> You should have.  The system enforces (or tries to) that a view can't be
> >> part of an inheritance hierarchy, but you seem to have managed to find a
> >> sequence of operations that avoids those checks.  Turning a table into a
> >> view with a manual CREATE RULE operation has always been a kluge, and
> >> it's missing a check that the table isn't part of an inheritance tree.
>
> > Is this a TODO?  Seems so.
>
> I think it's just a minor bugfix, but if you want to put it in TODO for
> a day or two, go ahead...

That's fine --- I will just push it to the patches queue so we know it
is a live issue.

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

  + If your life is a hard drive, Christ can be your backup. +

> > >> You should have.  The system enforces (or tries to) that a view can't be
> > >> part of an inheritance hierarchy, but you seem to have managed to find a
> > >> sequence of operations that avoids those checks.  Turning a table into a
> > >> view with a manual CREATE RULE operation has always been a kluge, and
> > >> it's missing a check that the table isn't part of an inheritance tree.


Is there a "specific" reason why views can't be part of an inheritance tree?
I mean: it's that we "don't want" it or it would be just difficult to implement?







      ___________________________________
L'email della prossima generazione? Puoi averla con la nuova Yahoo! Mail: http://it.docs.yahoo.com/nowyoucan.html

Scara Maccai <m_lists@yahoo.it> writes:
> Is there a "specific" reason why views can't be part of an inheritance tree?
> I mean: it's that we "don't want" it or it would be just difficult to implement?

It would certainly require a lot of rethinking of assumptions, in the
planner and elsewhere.  I have no good idea of how large the actual
patch might end up being if it were attempted.  But it's not something
that's high on anyone's wish-list, and there's a chance that it could
interfere with development of inheritance behaviors that people *do*
care about (like partitioning).

            regards, tom lane

Added to TODO:

        o Add checks to prevent a CREATE RULE views on inherited tables

          http://archives.postgresql.org/pgsql-general/2008-02/msg01420.php


---------------------------------------------------------------------------

Tom Lane wrote:
> Scara Maccai <m_lists@yahoo.it> writes:
> > I got that there should be no difference... plus, I don't get any
> > errors,
>
> You should have.  The system enforces (or tries to) that a view can't be
> part of an inheritance hierarchy, but you seem to have managed to find a
> sequence of operations that avoids those checks.  Turning a table into a
> view with a manual CREATE RULE operation has always been a kluge, and
> it's missing a check that the table isn't part of an inheritance tree.
>
>             regards, tom lane
>
> ---------------------------(end of broadcast)---------------------------
> TIP 2: Don't 'kill -9' the postmaster

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

  + If your life is a hard drive, Christ can be your backup. +