pg.ddx.io  pgsql-sql@postgresql.org mailing list archive  
help / color / mirror / Atom feed
Update problem I cannot work out
4+ messages / 3 participants
[nested] [flat]

* Update problem I cannot work out
@ 1999-05-02 02:25  M Simms <grim@argh.demon.co.uk>
  0 siblings, 0 replies; 4+ messages in thread

From: M Simms @ 1999-05-02 02:25 UTC (permalink / raw)
  To: pgsql-sql

I cannot work out how to do the following query

testdb=> select * from t1;
var1|var2
----+----
   1|   2
   2|   2
   2|   3
(3 rows)

testdb=> select * from t2;
var1|var2
----+----
   5|   2
   9|   3
   4|   4
(3 rows)

I need to add to the value of each var1 in t2 the sum of all var1's in t1
that have the same value corresponding var2's

testdb=> update t2 set var1=t2.var1+sum(t1.var1) where t1.var2=t2.var2;
ERROR:  parser: illegal use of aggregates or non-group column in target list

I imagine that this is failing because it isnt sure which values to
sum(), but I cannot for the life of me figure out how to solve this.

My second idea was to do

testdb=> update t2 set var1=t2.var1+t1.var1 where t2.var2=t1.var1;
NOTICE:  Non-functional update, only first update is performed
UPDATE 1

This will only, obviously, add one of the values I need to add.

The values I need to end up with, just so you know you are solving the
right problem, are:

testdb=> select * from t1;
var1|var2
----+----
   1|   2
   2|   2
   2|   3
(3 rows)

testdb=> select * from t2;
var1|var2
----+----
   8|   2
  11|   3
   4|   4
(3 rows)

Any help on this would be appreciated. I am a bit stuck here, and I am
sure I am just missing something blindingly obvious.
I am loathe to use a cursor to do this, because my actual dataset is
upwards of 10,000 records, and I do not wish to do 10,000 different
queries.



^ permalink  raw  reply  [nested|flat] 4+ messages in thread

* RE: [SQL] Update problem I cannot work out
@ 1999-05-02 03:13  Michael J Davis <michael.j.davis@tvguide.com>
  0 siblings, 1 reply; 4+ messages in thread

From: Michael J Davis @ 1999-05-02 03:13 UTC (permalink / raw)
  To: 'M Simms' <grim@argh.demon.co.uk>; pgsql-sql

	Try something like:

	Create function new_function(value_data_type)
		'Select sum(t1.var1) from t1 where t1.var2 = $1'  LANGUAGE
'sql';

	update t2 set var1=t2.var1+new_function(t2.var2);


	-----Original Message-----
	From:	M Simms [SMTP:grim@argh.demon.co.uk]
	Sent:	Saturday, May 01, 1999 8:26 PM
	To:	pgsql-sql@postgreSQL.org
	Subject:	[SQL] Update problem I cannot work out

	I cannot work out how to do the following query

	testdb=> select * from t1;
	var1|var2
	----+----
	   1|   2
	   2|   2
	   2|   3
	(3 rows)

	testdb=> select * from t2;
	var1|var2
	----+----
	   5|   2
	   9|   3
	   4|   4
	(3 rows)

	I need to add to the value of each var1 in t2 the sum of all var1's
in t1
	that have the same value corresponding var2's

	testdb=> update t2 set var1=t2.var1+sum(t1.var1) where
t1.var2=t2.var2;
	ERROR:  parser: illegal use of aggregates or non-group column in
target list

	I imagine that this is failing because it isnt sure which values to
	sum(), but I cannot for the life of me figure out how to solve this.

	My second idea was to do

	testdb=> update t2 set var1=t2.var1+t1.var1 where t2.var2=t1.var1;
	NOTICE:  Non-functional update, only first update is performed
	UPDATE 1

	This will only, obviously, add one of the values I need to add.

	The values I need to end up with, just so you know you are solving
the
	right problem, are:

	testdb=> select * from t1;
	var1|var2
	----+----
	   1|   2
	   2|   2
	   2|   3
	(3 rows)

	testdb=> select * from t2;
	var1|var2
	----+----
	   8|   2
	  11|   3
	   4|   4
	(3 rows)

	Any help on this would be appreciated. I am a bit stuck here, and I
am
	sure I am just missing something blindingly obvious.
	I am loathe to use a cursor to do this, because my actual dataset is
	upwards of 10,000 records, and I do not wish to do 10,000 different
	queries.



^ permalink  raw  reply  [nested|flat] 4+ messages in thread

* Re: [SQL] Update problem I cannot work out
@ 1999-05-02 12:24  M Simms <grim@argh.demon.co.uk>
  parent: Michael J Davis <michael.j.davis@tvguide.com>
  0 siblings, 1 reply; 4+ messages in thread

From: M Simms @ 1999-05-02 12:24 UTC (permalink / raw)
  To: michael.j.davis@tvguide.com; +Cc: pgsql-sql

> 
> 	Try something like:
> 
> 	Create function new_function(value_data_type)
> 		'Select sum(t1.var1) from t1 where t1.var2 = $1'  LANGUAGE
> 'sql';
> 
> 	update t2 set var1=t2.var1+new_function(t2.var2);
> 

*snip*

Hi, thanks for replying.

My reply to this is the question, that as each function contains a
select, is this technically a single pass.

I get the feeling there will be tremendous overhead on this theory with a
large dataset, in comparison to a single query that does the
update. Can anyone comment on how much slower this could be (although
obviously if there is no way to do it in a single query then there is
no choice)?

Thanks

						M Simms



^ permalink  raw  reply  [nested|flat] 4+ messages in thread

* Re: [SQL] Update problem I cannot work out
@ 1999-05-02 16:36  Tom Lane <tgl@sss.pgh.pa.us>
  parent: M Simms <grim@argh.demon.co.uk>
  0 siblings, 0 replies; 4+ messages in thread

From: Tom Lane @ 1999-05-02 16:36 UTC (permalink / raw)
  To: M Simms <grim@argh.demon.co.uk>; +Cc: pgsql-sql

M Simms <grim@argh.demon.co.uk> writes:
>> Create function new_function(value_data_type)
>> 'Select sum(t1.var1) from t1 where t1.var2 = $1'  LANGUAGE
>> 'sql';
>> 
>> update t2 set var1=t2.var1+new_function(t2.var2);

> My reply to this is the question, that as each function contains a
> select, is this technically a single pass.

I think you are right: this way will repeat the sub-select for each
tuple in t2.  Now if you have an index on t1.var2, the system could
exploit it to pull out just the desired tuples in each sub-select, so
the total work wouldn't really be O(M*N).  But it still looks slow.

If I knew t1 would be large, I think I'd do it with a temporary table:

	SELECT var2, sum(var1) FROM t1 GROUP BY var2 INTO temptable;
	UPDATE t2 SET var1 = t2.var1 + temptable.sum
		WHERE t2.var2 = temptable.var2;
	DROP TABLE temptable;

(In 6.5 you can use a temporary table to avoid needing explicit DROP,
but that's just a convenience.)  Perhaps there's a way to accomplish
the same thing with a subselect, rather than explicitly creating a
table, but I don't know how.

This is more like O(N log N) in the size of each table, assuming that
the join is done with mergejoin.  Actually, if there's not a huge
number of distinct values of var2, then temptable should be small enough
for a hashjoin, which'd mean that there's no need to sort t2 at all.

BTW, I think your original try

update t2 set var1=t2.var1+t1.var1 where t2.var2=t1.var1;

should have worked; it's probably a bug that it doesn't work.  IMHO
the semantics ought to be that the addition is repeated for each t1
tuple that matches the current t2 tuple ... but I haven't read the
SQL spec.

			regards, tom lane



^ permalink  raw  reply  [nested|flat] 4+ messages in thread


end of thread, other threads:[~1999-05-02 16:36 UTC | newest]

Thread overview: 4+ messages (download: mbox mbox.gz follow: Atom feed)
-- links below jump to the message on this page --
1999-05-02 02:25 Update problem I cannot work out M Simms <grim@argh.demon.co.uk>
1999-05-02 03:13 RE: [SQL] Update problem I cannot work out Michael J Davis <michael.j.davis@tvguide.com>
1999-05-02 12:24 ` M Simms <grim@argh.demon.co.uk>
1999-05-02 16:36   ` Tom Lane <tgl@sss.pgh.pa.us>

This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox