Обсуждение: [GENERAL] Catching errors inside a LOOP is causing performance issues

Hi all,

I had a function with a performance issue:

totalCharge := 0;

FOR myRecord IN ... LOOP

               ......

               IF severalConditionsAreMet THEN

                              BEGIN

                                             SELECT t1.charge INTO STRICT recordCharge

                                             FROM t1

                                             WHERE t1.id = myRecord.id AND otherComplexConditionsHere;

                                             totalCharge := totalCharge + recordCharge;

                                             ...........

                              EXCEPTION

                                             WHEN OTHERS THEN

                                                            NULL;

                              END;

               END IF;

END LOOP;

The function was being called 232 times (not counting the number of times the code from the FOR was accessed).

The IF from the FOR LOOP ended up being accessed 4466 times and was taking 561 seconds to complete all 4466 iterations.

For the particular data set that I had, the IF was always accessed, the SELECT from above never return data and the code was reaching the EXCEPTION branch each and every time.

I have changed the code to:

totalCharge := 0;

FOR myRecord IN ... LOOP

               ......

               IF severalConditionsAreMet THEN

                              SELECT t1.charge INTO recordCharge

                              FROM t1

                              WHERE t1.id = myRecord.id AND otherComplexConditionsHere;

                              IF (recordCharge IS NULL) THEN

                                             CONTINUE;

                              END IF;

                              totalCharge := totalCharge + recordCharge;

                              ...........

               END IF;

END LOOP;

Please note that for the table t1, the t1.charge column has a NOT NULL condition defined on it.

This time, the code from the IF takes 1-2 seconds to complete all 4466 iterations.

Basically, all I did was replace the

BEGIN

…

EXCEPTION

….

END;

With

               IF conditionIsNotMet THEN

                              CONTINUE;        

END IF;

Can someone please explain to me why this worked?

What happened behind the scenes?

I suspect that when you catch exceptions inside of a LOOP and the code ends up generating an exception, Postgres can’t use cached plans to optimize that code so it ends up planning the code at each iteration and this causes performance issues.

Is my assumption correct?

Thanks a lot,

Denisa Cîrstescu

I am adding an example that is runnable in order to demonstrate my point.

CREATE OR REPLACE FUNCTION initialVersion()

RETURNS VOID AS $$

declare

  testDate DATE;

begin

  for i in 1..999999 loop

    begin

    select now() into strict testDate where 1=0;

    exception when others

    then null;

    end;

  end loop;

end;

$$ Language plpgsql;

CREATE OR REPLACE FUNCTION secondVersion()

RETURNS VOID AS $$

declare

    testDate DATE;

begin

  for i in 1..999999 loop

    select now() into testDate where 1=0;

    if testDate is null then

      continue;

    end if;

  end loop;

end;

$$ Language plpgsql;

select initialVersion(); -- 19.7 seconds

select secondVersion(); -- 5.2

As you can see there is a difference of almost 15 seconds.

In the example that I have provided initially, the difference is bigger because the SELECT FROM t1 runs against complex data and takes more time to execute that the simple SELECT provided in this second example.

Thanks,

Denisa

Hi all,

I had a function with a performance issue:

totalCharge := 0;

FOR myRecord IN ... LOOP

               ......

               IF severalConditionsAreMet THEN

                              BEGIN

                                             SELECT t1.charge INTO STRICT recordCharge

                                             FROM t1

                                             WHERE t1.id = myRecord.id AND otherComplexConditionsHere;

                                             totalCharge := totalCharge + recordCharge;

                                             ...........

                              EXCEPTION

                                             WHEN OTHERS THEN

                                                            NULL;

                              END;

               END IF;

END LOOP;

The function was being called 232 times (not counting the number of times the code from the FOR was accessed).

The IF from the FOR LOOP ended up being accessed 4466 times and was taking 561 seconds to complete all 4466 iterations.

For the particular data set that I had, the IF was always accessed, the SELECT from above never return data and the code was reaching the EXCEPTION branch each and every time.

I have changed the code to:

totalCharge := 0;

FOR myRecord IN ... LOOP

               ......

               IF severalConditionsAreMet THEN

                              SELECT t1.charge INTO recordCharge

                              FROM t1

                              WHERE t1.id = myRecord.id AND otherComplexConditionsHere;

                              IF (recordCharge IS NULL) THEN

                                             CONTINUE;

                              END IF;

                              totalCharge := totalCharge + recordCharge;

                              ...........

               END IF;

END LOOP;

Please note that for the table t1, the t1.charge column has a NOT NULL condition defined on it.

This time, the code from the IF takes 1-2 seconds to complete all 4466 iterations.

Basically, all I did was replace the

BEGIN

…

EXCEPTION

….

END;

With

               IF conditionIsNotMet THEN

                              CONTINUE;        

END IF;

Can someone please explain to me why this worked?

What happened behind the scenes?

I suspect that when you catch exceptions inside of a LOOP and the code ends up generating an exception, Postgres can’t use cached plans to optimize that code so it ends up planning the code at each iteration and this causes performance issues.

Is my assumption correct?

Thanks a lot,

Denisa Cîrstescu

Hi Tom,

You said that trapping an arbitrary exception is a “fairly expensive mechanism”.

What if the:

begin

    ….

exception when others

    then null;

end;

would be replaced with

begin

    ….

exception when NO_DATA_FOUND

    then null;

end;

When the code is catching a certain exception: NO_DATA_FOUND does this make any difference?

Or it’s all about the process of setting up and ending a subtransaction?

Thanks,

Denisa Cîrstescu

-----Original Message-----
From: Tom Lane [mailto:tgl@sss.pgh.pa.us]
Sent: Wednesday, September 27, 2017 9:00 PM
To: David G. Johnston <david.g.johnston@gmail.com>
Cc: Denisa Cirstescu <Denisa.Cirstescu@tangoe.com>; pgsql-general@postgresql.org
Subject: Re: [GENERAL] Catching errors inside a LOOP is causing performance issues

"David G. Johnston" <david.g.johnston@gmail.com> writes:

> ​Not sure how much detail you are looking for but the docs say this:

> "​Tip: A block containing an EXCEPTION clause is significantly more

> expensive to enter and exit than a block without one. Therefore, don't

> use EXCEPTION without need."

> https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.p

> ostgresql.org%2Fdocs%2Fcurrent%2Fstatic%2Fplpgsql-control-structures.h

> tml%23PLPGSQL-ERROR-TRAPPING&data=01%7C01%7CDenisa.Cirstescu%40tangoe.

> com%7C6243898de8ae4141290a08d505d194e6%7C3ba137049b66408a9fb9db51aba57

> 9e4%7C0&sdata=iTBlh1PpcvJQiBZNPjDxsu7ExT%2BP%2BAirqr9Upz9sbJQ%3D&reser

> ved=0

> I'm somewhat doubting "plan caching" has anything to do with this; I

> suspect its basically that there is high memory and runtime overhead

> to deal with the possibilities of needing to convert a exception into

> a branch instead of allowing it to be fatal.

Yeah, it's about the overhead of setting up and ending a subtransaction.

That's a fairly expensive mechanism, but we don't have anything cheaper that is able to recover from arbitrary errors.

                                             regards, tom lane