bug#29992: [PATCH] services: postgresql: Use pg_ctl to start and stop postgres.

[Clément Lassieur]

Clément Lassieur <address@hidden> writes:

> Ludovic Courtès <address@hidden> writes:
>
>> Clément Lassieur <address@hidden> skribis:
>>
>>> Fixes <https://bugs.gnu.org/29992>.
>>>
>>> * gnu/services/databases.scm (postgresql-shepherd-service): Replace
>>> make-forkexec-constructor and make-kill-destructor with pg_ctl.
>>
>> [...]
>>
>>> +     (let* ((pg_ctl-wrapper
>>> +             ;; Wrapper script that switches to the 'postgres' user before
>>> +             ;; launching daemon.
>>> +             (program-file
>>> +              "pg_ctl-wrapper"
>>> +              #~(begin
>>> +                  (use-modules (ice-9 match)
>>> +                               (ice-9 format))
>>> +                  (match (command-line)
>>> +                    ((_ mode)
>>> +                     (let ((user (getpwnam "postgres"))
>>> +                           (pg_ctl #$(file-append postgresql 
>>> "/bin/pg_ctl")))
>>> +                       (setgid (passwd:gid user))
>>> +                       (setuid (passwd:uid user))
>>> +                       (system
>>> +                        (format #f "~a -D ~a -o '--config-file=~a -p ~d' 
>>> ~a"
>>> +                                pg_ctl #$data-directory #$config-file 
>>> #$port
>>> +                                mode))))))))
>>
>> I think we should use ‘execl’ here instead of ‘system’ so that (1) the
>> exit code is correct, and (2) we don’t go through /bin/sh.
>
> Hi Ludovic, thank you for the review.
>
> How do you pass single quotes as an ‘execl’ argument?
>
> https://www.postgresql.org/docs/9.3/static/app-pg-ctl.html says:
> --8<---------------cut here---------------start------------->8---
> -o options
>     Specifies options to be passed directly to the postgres command.
>     The options should usually be surrounded by single or double quotes
>     to ensure that they are passed through as a group.
> --8<---------------cut here---------------end--------------->8---

Oh I understood this.  I'll do a new patch.

> Also, I don't understand how the exit code matters since Guile scripts
> seem to always return 0, no matter if the last return value is true or
> false.
>
> Clément