[systemd-devel] environment variable questions

[Brian Kroth]

Mantas Mikulėnas <grawity at gmail.com> 2016-06-17 07:47:
>On Fri, Jun 17, 2016 at 6:01 AM, Brian Kroth <bpkroth at gmail.com> wrote:
>
>> Hi again, related to my inetd conversion example, in my .service unit I
>> have something like this:
>>
>> # nagios-nrpe-server at .service:
>> [Service]
>> Environment=NICENESS=0
>> EnvironmentFile=-/etc/default/nagios-nrpe-server
>> ExecStart=/usr/sbin/nrpe -i $DAEMON_OPTIONS
>> Nice=$NICENESS
>>
>> # /etc/default/nagios-nrpe-server:
>> DAEMON_OPTIONS="--no-ssl"
>> #NICENESS=5
>> INETD=1
>>
>>
>> With that I get this sort of error message:
>> [/etc/systemd/system/nagios-nrpe-server at .service] Failed to parse nice
>> priority, ignoring: $NICENESS
>>
>> I added the leading "Environment=NICENESS=0" directive to try and make
>> sure it wasn't just an empty variable kinda thing, but it didn't seem to
>> help.
>>
>>
>> It's somewhat unclear from the man pages as to whether or not $VAR
>> expansion is done outside of the Exec* directives, and I couldn't find a
>> definitive answer online, but based on the above, I'm guessing it's not,
>> correct?
>>
>
>No, they're not expanded anywhere else.
>
>Generally it's best to entirely avoid indirection via /etc/default, and
>just configure daemons directly via their .service units. (Unlike init
>scripts, they're freely editable by sysadmins.) In certain cases, a
>variable for command-line arguments might make sense, but $NICENESS? Why?

It was mostly an exercise in understanding as I went through the process 
of trying to convert a legacy init script over for myself.

I think I agree with you that it's easy enough to add the 
/etc/systemd/system/$service.conf.d/override.conf style overrides for 
individual parameters like that.

The distaste I guess I'm left with is that, at least as things 
transition, we end up having to look in several places to figure out 
where all the configuration sources are coming from (eg: /etc/default, 
/{etc,run,lib}/systemd, etc.).  It's not always clear on first glance 
which parts are respected by other parts.

>> Also, so long as I'm asking questions, I've got one about Condition*
>> directives.  There doesn't seem to be one for Environment variable sort of
>> parsing or some sort of shell command evaluation (eg: returns 0 kinda
>> thing).
>>
>
>Nope. (I think this was rejected once already.) Ask #debian-systemd about
>how they're dealing with the existing /etc/default files with "ENABLE=yes"
>during upgrades to native units...

Will do, thanks.

>Though It's slightly different for .service, where a failed ExecStartPre
>can cancel the start of the main daemon.
>
>There's also the "generator" concept, where external tools are used to
>create actual unit files in RAM (or do the equivalent of `systemctl
>enable`) before systemd even starts the boot process; this is how
>/etc/fstab is parsed into .mount units by fstab-generator, for example, and
>also how Debian's sysv-generator creates fake .service units for old
>/etc/init.d scripts.

Yeah, I get the reason for those, but I find them somewhat opaque when 
trying to analyze the configuration of a system, especially outside of a 
running system itself (eg: just on the config management end of things).

>Something like:
>>
>> # nagios-nrpe-server.socket:
>> [Unit]
>> ConditionReturnsZero=/bin/grep -q ^INETD=1$ /etc/default/nagios-nrpe-server
>>
>> I got to thinking about this for the INETD directive that comes in the
>> /etc/default/nagios-nrpe-server script.  It's somewhat irrelevant in this
>> case since the .socket unit already needs to be active before the @.service
>> pair is instantiated.
>>
>> Looking at some of the other examples the system ships with like
>> ssh.service vs ssh at .service/ssh.socket, they have Conflicts= directives
>> against each other, but there's not really a conf file that I see that
>> directs the system to use inetd-style vs daemon mode for those.
>>
>
>I'm not really sure how this could have worked even before systemd.. I
>mean, this would be selecting between two entirely separate systems, and
>even if the init.d scripts understood that INETD=yes means "do not start
>the daemon", I doubt [x]inetd itself cared about configuration /etc/default
>at all. (Did it?)

No, that's a good point.  I think you're probably right: there are inetd 
helpers for dpkg at least, but I think it was probably still up to the 
admin to make sure one was disabled and the other was enabled, so that 
notion at least doesn't change - just the mechanism.

>> So, since there's not really a way to read that from an EnvironmentFile
>> style conf file and use it in a Conditional* directive to disable one vs
>> the other right now, is it just expected that people use something like
>> "systemctl mask ssh.service" vs. "systemctl mask ssh.socket" to select
>> between them?
>>
>
>It's expected that people use `systemctl enable` and `systemctl disable` to
>select between them.

K, that's what I figured.  Good to hear it from another source though.

Thanks again.

Cheers,
Brian
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <https://lists.freedesktop.org/archives/systemd-devel/attachments/20160617/089b07da/attachment.sig>