Re: [PATCH] Add log_transaction setting

   


   


  10.08.2024, 16:40, "Kirill Reshke" <reshkekirill@gmail.com>:


  

    On Thu, 4 Jul 2024 at 21:46, Sergey Solovev

    <sergey.soloviev@tantorlabs.ru> wrote:
  
  
    

     Hi.

    

     We encountered a problem with excessive logging when transaction is

     sampled.

     When it is getting sampled every statement is logged, event SELECT. This

     can

     lead to performance degradation and log polluting.

     I have added new setting to filter statements when transaction is

     sampled - log_transaction.

    

     Overview

     ===================

     This parameter works very similar to log_statement, but it filters out

     statements only

     in sampled transaction. It has same values as log_statement, access

     rights (only superuser),

     setup in postgresql.conf or by superuser with SET.

     But by default it has value "all" in compliance with existing behaviour

     (to log every statement).

    

     Example usage

     ==================

     Log every DDL, but only subset of other statements.

    

     postgresql.conf

      > log_transaction = ddl

      > log_statement = all

      > log_transaction_sample_rate = 1

      > log_statement_sample_rate = 0.3

    

     backend:

      > BEGIN;

      > CREATE TABLE t1(value text);

      > INSERT INTO t1(value) VALUES ('hello'), ('world');

      > SELECT * FROM t1;

      > DROP TABLE t1;

      > COMMIT;

    

     logfile:

      > LOG: duration: 6.465 ms statement: create table t1(value text);

      > LOG: statement: insert into t1(value) values ('hello'), ('world');

      > LOG: duration: 6.457 ms statement: drop table t1;

    

     As you can see CREATE and DROP were logged with duration (because it is

     DDL), but

     only INSERT was logged (without duration) - not SELECT.

    

     Testing

     ===================

     All tests are passed - default configuration is compatible with existing

     behaviour.

     Honestly, I could not find any TAP/regress tests that would test logging

     - some

     tests only use logs to ensure test results.

     I did not find suitable place for such tests, so no tests provided

    

     Implementation details

     ===================

     I have modified signature of check_log_duration function - this accepts

     List of

     statements that were executed. This is to decide whether we should log

     current statement if transaction is sampled.

     But this list can be empty, in that case we decide to log. NIL is passed

     only

     in fast path, PARSE and BIND functions - by default these are logged

     if transaction is sampled, so we are compliant with existing behaviour

     again.

     In first implementation version, there was boolean flag (instead of

     List), but

     it was replaced by List to defer determining (function call) whether it

     is worth logging.

    

     -------------

     Best regards, Sergey Solovev
  
  

    

    Hi!

    

    As I understand, the need of this GUC variable comes from fact, that

    is the transaction is sampled for logging, all statements within this

    tx are logged and this is not configurable now, correct?

    Well, if this is the case, why should we add a new GUC? Maybe we

    should use `log_statement` in this case too, so there is a bug, that

    log_statement is honored not during tx sampling?

    

    

    Also, tests are failing[1]

    

    

    [1] https://cirrus-ci.com/task/5645711230894080

     
  --

  Best regards,

  Kirill Reshke


  Hi, Kirill!


   


  Thanks for review.


  
    
      Also, tests are failing[1]
    
  


  That is my mistake: I have tested patch on REL_17BETA2, but set 18 version, and


  it seems that I run tests without proper ./configure flags. I reproduced error and fixed


  it in new patch.


  
    As I understand, the need of this GUC variable comes from fact, that

    is the transaction is sampled for logging, all statements within this

    tx are logged and this is not configurable now, correct?
  


  Yes. That's the point - all or nothing.


  
    Well, if this is the case, why should we add a new GUC? Maybe we

    should use `log_statement` in this case too, so there is a bug, that

    log_statement is honored not during tx sampling?
  


  
    That can be good solution, but I proceed from possibility of flexible logging setup.
  
  
    Shown example (when all DDL statements were logged) is just one of such use cases.
  
  
    Another example is security policy: when we must log every data modification -
  
  
    set `log_transaction = mod`.