Обсуждение: Downsides of liberally using CREATE TEMP TABLE ... ON COMMIT DROP
Hello,
I heard at a PostgreSQL talk that you should not liberally create temp tables in the course of frequently-used functions etc, because (roughly) you're using up some of the same resources that you for your regular tables.
On Sunday, January 28, 2018, Ryan Murphy <ryanfmurphy@gmail.com> wrote:
Is this true? Is there an important reason not to have e.g. a plpgsql function that uses a temp table? What are the specific problems if I do this? Is the problem ameliorated if I add ON COMMIT DROP?Hello,I heard at a PostgreSQL talk that you should not liberally create temp tables in the course of frequently-used functions etc, because (roughly) you're using up some of the same resources that you for your regular tables.
I believe the main, and maybe only, concern is the bloating of the system catalog tables since you are constantly adding and removing records. Yes, they will be vacuumed but vacuuming and bloat on catalog tables slows every single query down to some, degree since every query has to lookup its objects is those catalogs. Though caching probably alleviates some of that.
The way most temp tables are used on commit drop likely has little impact on this, but the specific usage pattern matters a great deal in answering the question.
David J.
I believe the main, and maybe only, concern is the bloating of the system catalog tables since you are constantly adding and removing records. Yes, they will be vacuumed but vacuuming and bloat on catalog tables slows every single query down to some, degree since every query has to lookup its objects is those catalogs. Though caching probably alleviates some of that
Yes, that's exactly the concern I heard, thanks for reminding me.
If I want to e.g. temporarily store a "setof records" or a "table" result in a variable as part of a calculation in a plpgsql function, do I have any other option than CREATE TEMPORARY TABLE? It didn't seem to work when I DECLAREd a variable of type "setof table_name" or "setof table_name%rowtype", and then SELECT INTO it.
On 01/28/2018 08:46 AM, Ryan Murphy wrote: > I believe the main, and maybe only, concern is the bloating of the system catalog tables since you are constantly addingand removing records. Yes, they will be vacuumed but vacuuming and bloat on catalog tables slows every single querydown to some, degree since every query has to lookup its objects is those catalogs. Though caching probably alleviatessome of that > > > Yes, that's exactly the concern I heard, thanks for reminding me. > > If I want to e.g. temporarily store a "setof records" or a "table" result in a variable as part of a calculation in a plpgsqlfunction, do I have any other option than CREATE TEMPORARY TABLE? It didn't seem to work when I DECLAREd a variableof type "setof table_name" or "setof table_name%rowtype", and then SELECT INTO it. > You may not need temp tables at all. You can use subselects, derived tables, and cte's: select sum(a+b) as total from ( select a, b+1 from detail ) as tmpx; This does the same thing as a temp table, with no temp table. -Andy
“On commit drop” is generally good practice for several reasons: if your function is called in a loop, or the create statementis in a loop or you want to call your function more than once in a given session (with out explicitly dropping yourtemp table). The commit in question is the function btw. > On Jan 28, 2018, at 8:53 AM, Andy Colson <andy@squeakycode.net> wrote: > >> On 01/28/2018 08:46 AM, Ryan Murphy wrote: >> I believe the main, and maybe only, concern is the bloating of the system catalog tables since you are constantly addingand removing records. Yes, they will be vacuumed but vacuuming and bloat on catalog tables slows every single querydown to some, degree since every query has to lookup its objects is those catalogs. Though caching probably alleviatessome of that >> Yes, that's exactly the concern I heard, thanks for reminding me. >> If I want to e.g. temporarily store a "setof records" or a "table" result in a variable as part of a calculation in aplpgsql function, do I have any other option than CREATE TEMPORARY TABLE? It didn't seem to work when I DECLAREd a variableof type "setof table_name" or "setof table_name%rowtype", and then SELECT INTO it. > > You may not need temp tables at all. You can use subselects, derived tables, and cte's: > > select sum(a+b) as total > from ( > select a, b+1 > from detail > ) as tmpx; > > > This does the same thing as a temp table, with no temp table. > > -Andy >
On Sunday, January 28, 2018, Rob Sargent <robjsargent@gmail.com> wrote:
The commit in question is the function btw.
That doesn't make sense - functions can't commit.
David J.
I think the table is dropped by end of function
On Sunday, January 28, 2018, Rob Sargent <robjsargent@gmail.com> wrote:The commit in question is the function btw.That doesn't make sense - functions can't commit.David J.