>
> On Aug 12, 2015, at 7:18 AM, Bernd Helmle <mailings@oopsware.de> wrote:
>
> Hi,
>
> Looks like postgresql9[45]-setup is careless when someone tries to use them
> in customized
> systemd environments (e.g. via drop-in configurations[1]). Currently we do
> this:
>
> # this parsing technique fails for PGDATA pathnames containing spaces,
> # but there's not much I can do about it given systemctl's output format...
> PGDATA=`systemctl show -p Environment "${SERVICE_NAME}.service" |
> sed 's/^Environment=//' | tr ' ' '\n' |
> sed -n 's/^PGDATA=//p' | tail -n 1`
>
> [...some more code later...]
>
> # Get data directory from the service file
> PGDATA=`sed -n 's/Environment=PGDATA=//p' "${SERVICE_FILE}"`
>
> So we obviously overwrite any PGDATA setting returned by 'systemctl show'
> earlier. If someone uses service files only, this doesn't heavily matter,
> since SERVICE_FILE is tested against multiple locations. However, drop-in
> configurations are ignored/overwritten with this method. I don't
> understand, why we do the sed approach anyways, since 'systemctl show'
> already covers all cases, afaics. So i suggest to get rid of handling the
> SERVICE_FILE directly and leave 'systemctl show' alone, patch attached.
>
> Opinions?
>
> [1] <http://www.freedesktop.org/software/systemd/man/systemd.unit.html>
Unfortunately, I’m not that familiar with systemd yet and Devrim is out for a few weeks, so we might not get to this
veryquickly.
