Re: init.d scripts: LSB and Debian policy

[Thomas Hood <jdthood@yahoo.co.uk>]

Tobias Burnus wrote:
> [DP] "force-reload -- cause the configuration to be reloaded if the
>       service supports this, otherwise restart the service."
> [LSB 1.9-pre snapshot]
>      "force-reload -- cause the configuration to be reloaded if the
>       service supports this, otherwise restart the service
>       if it is running"
> This solves the reload/restart ambiguity if the service is not running.

FYI: There is a wish (#203239) for policy to endorse a name for the
action of restarting a service iff it is running.  Names proposed
have been "restart-if-running" and "restart" with second argument
"running".

It is interesting that DP and LSB differ in what they require of
force-reload.  The LSB requirement makes more sense whereas the DP
requirement is easier to implement in a distro whose initscripts
fail to implement restart-if-running.  In present Debian initscripts
force-reload is either synonymous with reload or synonymous with
restart.  If an initscript already has restart-if-running then
implementing LSB-style force-reload is easy: in each case it is
either synonymous with reload or synonymous with restart-if-running.

> The exit status codes in the DP are a bit unclear. Especially regarding
> . running "start" on a service already running
> . running "stop" on a service already stopped or not running

The initscript should do nothing and exit with status 0 in these
cases.

> . running "reload" on a service already stopped or not running
> The LSB says that these are successful while the DP says nothing about
> this.

You misquote the LSB (1.3).  It says that running "restart" on a service
already stopped or not running should be counted as a success.

> Some use Debian init.d scripts use start-stop-daemon with --nodook,
       "--oknodo"
> others (a bit more) without.
> Adding to the DP that they should/shall return a "0" might make sense.

Trying to reload a service that is not running should result in
a nonzero status, I believe.

> (The [DP]   test -f program-executed-later-in-script || exit 0
>      [LSB]  test -f program-executed-later-in-file || exit 5
>  difference it also unfortunate (0=success, 5=program is not installed))

Becoming LSB compliant on this point wouldn't be trivially easy.
A lot of scripts run with "set -e" would have to be changed from
doing

    /etc/init.d/foo action

which would fail on the nonzero exit status, to doing

    if /etc/init.d/foo action ; then
        case $? in
            ...
        esac
    fi

Making this change would take a long time to accomplish.  Sounds
like something that should happen post-sarge.

--
Thomas Hood