Skip site navigation (1) Skip section navigation (2)

Re: Why does the sequence skip a number with generate_series?

From: Shane Ambler <pgsql(at)Sheeky(dot)Biz>
To: PostgreSQL Hackers <pgsql-hackers(at)postgresql(dot)org>
Cc: Stephan Szabo <sszabo(at)megazone(dot)bigpanda(dot)com>, Jeff Frost <jeff(at)frostconsultingllc(dot)com>, pgsql-sql(at)postgresql(dot)org
Subject: Re: Why does the sequence skip a number with generate_series?
Date: 2007-10-04 14:25:14
Message-ID: 4704F7CA.4070200@Sheeky.Biz (view raw or flat)
Thread:
Lists: pgsql-hackerspgsql-sql
Stephan Szabo wrote:
> On Tue, 2 Oct 2007, Jeff Frost wrote:
> 
>> I expected these numbers to be in sync, but was suprised to see that the
>> sequence skips a values after every generate series.
>>
>> CREATE TABLE jefftest ( id serial, num int );
>> INSERT INTO jefftest (num) values (generate_series(1,10));
>> INSERT INTO jefftest (num) values (generate_series(11,20));
>> INSERT INTO jefftest (num) values (generate_series(21,30));
> 
> It seems to do what you'd expect if you do
>  INSERT INTO jefftest(num) select a from generate_series(1,10) as foo(a);
>  INSERT INTO jefftest(num) select a from generate_series(11,20) as foo(a);
>  INSERT INTO jefftest(num) select a from generate_series(21,30) as foo(a);
> 
> I tried a function that raises a notice and called it as
>  select f1(1), generate_series(1,10);
> and got 11 notices so it looks like there's some kind of phantom involved.
> 

That's interesting - might need an answer from the core hackers.
I am posting this to pgsql-hackers to get their comments and feedback.
I wouldn't count it as a bug but it could be regarded as undesirable 
side effects.

My guess is that what appears to happen is that the sequence is created 
by incrementing as part of the insert steps and the test to check the 
end of the sequence is -
if last_inserted_number > end_sequence_number
	rollback_last_insert

This would explain the skip in sequence numbers.

My thoughts are that -
if last_inserted_number < end_sequence_number
	insert_again

would be a better way to approach this. Of course you would also need to 
check that the (last_insert + step_size) isn't greater than the 
end_sequence_number when the step_size is given.

I haven't looked at the code so I don't know if that fits easily into 
the flow of things.

The as foo(a) test would fit this as the sequence is generated into the 
equivalent of a temporary table the same as a subselect, then used as 
insert data. The rollback would be applied during the temporary table 
generation so won't show when the data is copied across to fulfill the 
insert.

Maybe the planner or the generate series function could use a temporary 
table to give the same results as select from generate_series()


-- 

Shane Ambler
pgSQL(at)Sheeky(dot)Biz

Get Sheeky @ http://Sheeky.Biz

In response to

Responses

pgsql-hackers by date

Next:From: Simon RiggsDate: 2007-10-04 14:27:52
Subject: Re: First steps with 8.3 and autovacuum launcher
Previous:From: Simon RiggsDate: 2007-10-04 14:11:22
Subject: Connection Pools and DISCARD ALL

pgsql-sql by date

Next:From: Alvaro HerreraDate: 2007-10-04 14:46:56
Subject: Re: [HACKERS] Why does the sequence skip a number withgenerate_series?
Previous:From: A. KretschmerDate: 2007-10-04 13:00:11
Subject: Re: Rule Error

Privacy Policy | About PostgreSQL
Copyright © 1996-2014 The PostgreSQL Global Development Group