Re: Strategy for "one-off" tasks

[Jamie Wilkinson]

This one time, at band camp, Frank Smith wrote:
>
>
>--On Thursday, July 10, 2003 15:34:06 -0700 Darren Dunham <ddunham@taos.com> 
>wrote:
>
>>> 
>>> But to run another one-time-only later then you need another
>>> filename, and you can't ever remove them or it will run
>>> again.  The other idea of year.month.day.hr seems cleaner,
>>> but then there is no assurance that the command will run at
>>> all (if there is a network problem, for example, during that
>>> hourly run).  I get a couple of failed connection errors a day
>>> from remote hosts).
>> 
>> I think the idea is "why would you need to perform a one-time-only
>> task"?  Can you give an example of what this would be?  Why would it not
>> need to be performed if a machine were being installed?  (possibly for a
>> conversion of somthing)?
>>
>
>One example:
>   
>   We have a bunch of web servers that connect to a farm of database servers
>on the same private subnet, and the connection goes through a load balancer.
>The database servers need a static route set for each web server pointing to
>the load balancer, so the web server will get the reply back from the same
>IP it sent the request to (the load balancer).
>   When we add a new web server, we need to add a route to each database
>server.  Using "AppendIfNoSuchLine" works fine for adding to the init
>script, but we also need to run a shellcomand one-time-only to add the
>route now instead of on the next reboot (which is probably many months
>from now).

Adapt the shellcommands I just posted:

shellcommands:

    any::

        "/sbin/route | grep ${some_route}"
            useshell=true
            define=set_route
            inform=false

    set_route::

        "/sbin/route add ${some_route}"


-- 
jaq@spacepants.org                           http://spacepants.org/jaq.gpg