PostgreSQL 12: Detach postmaster process from pg_ctl’s session at server startup

Recently a commit landed in the PostgreSQL development tree that made me aware of something I did not know so far: When you start PostgreSQL with a script using pg_ctl and that script was cancelled the postmaster was killed as well before PostgreSQL 12. Sounds weird? Lets do a little demo.

The little demo script is quite simple: Print the version of pg_ctl, startup PostgreSQL and then sleep for 10 seconds:

postgres@pgbox:/home/postgres/ [PG10] cat start.sh 
#!/bin/bash
/u01/app/postgres/product/10/db_3/bin/pg_ctl --version
/u01/app/postgres/product/10/db_3/bin/pg_ctl -D /u02/pgdata/10/PG103 start
sleep 10

When you execute that against PostgreSQL before version 12 and then CRTL-C the script while it is sleeping the postmaster will be killed as well:

postgres@pgbox:/home/postgres/ [PG10] ./start.sh 
pg_ctl (PostgreSQL) 10.5
waiting for server to start....2019-01-25 13:00:07.258 CET - 1 - 6853 -  - @ LOG:  listening on IPv4 address "0.0.0.0", port 5432
2019-01-25 13:00:07.259 CET - 2 - 6853 -  - @ LOG:  listening on IPv6 address "::", port 5432
2019-01-25 13:00:07.263 CET - 3 - 6853 -  - @ LOG:  listening on Unix socket "/tmp/.s.PGSQL.5432"
2019-01-25 13:00:07.282 CET - 4 - 6853 -  - @ LOG:  redirecting log output to logging collector process
2019-01-25 13:00:07.282 CET - 5 - 6853 -  - @ HINT:  Future log output will appear in directory "pg_log".
 done
server started
^C
postgres@pgbox:/home/postgres/ [PG10] ps -ef | grep postgres

Starting with PostgreSQL 12 this will not happen anymore:

postgres@pgbox:/home/postgres/ [PG10] cat start.sh 
#!/bin/bash
/u01/app/postgres/product/DEV/db_1/bin/pg_ctl --version
/u01/app/postgres/product/DEV/db_1/bin/pg_ctl -D /u02/pgdata/DEV start
sleep 10
postgres@pgbox:/home/postgres/ [PGDEV] ./start.sh 
pg_ctl (PostgreSQL) 12devel
waiting for server to start....2019-01-25 13:02:51.690 CET - 1 - 9408 -  - @ LOG:  listening on IPv6 address "::1", port 5433
2019-01-25 13:02:51.690 CET - 2 - 9408 -  - @ LOG:  listening on IPv4 address "127.0.0.1", port 5433
2019-01-25 13:02:51.700 CET - 3 - 9408 -  - @ LOG:  listening on Unix socket "/tmp/.s.PGSQL.5433"
2019-01-25 13:02:51.734 CET - 4 - 9408 -  - @ LOG:  redirecting log output to logging collector process
2019-01-25 13:02:51.734 CET - 5 - 9408 -  - @ HINT:  Future log output will appear in directory "pg_log".
 done
server started
^C
postgres@pgbox:/home/postgres/ [PGDEV] ps -ef | grep postgres
postgres  9408     1  0 13:02 ?        00:00:00 /u01/app/postgres/product/DEV/db_1/bin/postgres -D /u02/pgdata/DEV
postgres  9409  9408  0 13:02 ?        00:00:00 postgres: logger   
postgres  9411  9408  0 13:02 ?        00:00:00 postgres: checkpointer   
postgres  9412  9408  0 13:02 ?        00:00:00 postgres: background writer   
postgres  9413  9408  0 13:02 ?        00:00:00 postgres: walwriter   
postgres  9414  9408  0 13:02 ?        00:00:00 postgres: autovacuum launcher   
postgres  9415  9408  0 13:02 ?        00:00:00 postgres: stats collector   
postgres  9416  9408  0 13:02 ?        00:00:00 postgres: logical replication launcher   
postgres  9422 29009  0 13:03 pts/2    00:00:00 ps -ef
postgres  9423 29009  0 13:03 pts/2    00:00:00 grep --color=auto postgres
root     29005  2910  0 12:53 ?        00:00:00 sshd: postgres [priv]
postgres 29008 29005  0 12:53 ?        00:00:01 sshd: postgres@pts/2
postgres 29009 29008  0 12:53 pts/2    00:00:00 -bash
root     29463  2910  0 12:54 ?        00:00:00 sshd: postgres [priv]
postgres 29466 29463  0 12:54 ?        00:00:00 sshd: postgres@pts/1
postgres 29467 29466  0 12:54 pts/1    00:00:00 -bash

Learned something new again.