[OPEN-ILS-DEV] On the merits of init scripts...

[Mike Rylander]

On Sun, Jul 27, 2008 at 9:05 AM, Dan Scott <denials at gmail.com> wrote:
> 2008/7/16 Scott McKellar <mck9 at swbell.net>:
>>
>> --- Aaron Joyner <aaron at joyner.ws> wrote:
>>
>> <snip -- about a system init script, at least for simple cases>
>>
>> Any init script should respect the distinction between OpenSRF (a
>> generic infrastructure layer) and Evergreen (a specific application
>> built on top of OpenSRF).
>>
>> In principle it should be possible, and preferably easy, to run
>> OpenSRF for other things besides Evergreen, or even without Evergreen
>> at all.  The separation between those layers is still not completely
>> clean, but hopefully it will be some day, and in the meanwhile we
>> should at least maintain the pretense of independence.
>>
>> So I would suggest two init scripts.  The first one would initialize
>> OpenSRF.  The second one would either call the first one or verify
>> that OpenSRF was already up and running, and then initialize the
>> pieces specific to Evergreen.
>>
>>
>
> I've been thinking about this a little bit.
>
> Given that Evergreen is an OpenSRF application, I'm not sure what it
> would mean to initialize OpenSRF without initializing the pieces
> specific to Evergreen; opensrf.xml defines the OpenSRF services that
> should be started up as part of start_perl / start_c.
>
> I suppose you could have a "opensrf" init script that just performed
> the equivalent of stop/start/restart[router&&perl&&c] (along with
> dependency checking to ensure postgresql / ejabberd is running), and
> an "evergreen" init script to stop/start/restart  clark-kent.pl / SIP
> server / assorted other daemons, and make "opensrf" and apache as
> dependencies. That separation would buy us the ability to restart the
> daemons without restarting the router/c/perl services.
>
> I definitely like the idea of a universal init script. It's going to
> be most useful for single-server systems, of course; if you've moved
> Apache and PostgreSQL and ejabber and memcached off to separate
> servers for scaling purposes, the scripts will obviously have to be
> tweaked. But at that point you're in a different realm than just "I
> have a spare server and want to try out Evergreen in the simplest
> possible fashion" :)

Another thing to consider (in the "pros" column for separation) is
that we will eventually be able to address each service independently
for start/stop/restart.  I was hoping this would happen for 1.4, but
2.0 is more likely.  Still, separating now will make things easier
down the road.

-- 
Mike Rylander
 | VP, Research and Design
 | Equinox Software, Inc. / The Evergreen Experts
 | phone: 1-877-OPEN-ILS (673-6457)
 | email: miker at esilibrary.com
 | web: http://www.esilibrary.com