In the design doc that's under review at:
https://reviews.apache.org/r/18979/, there's a section talking about cron.
When we were doing the clientv2 design, we discussed the idea of separating
cron functionality on the command line. In the current client, to create a
job that runs on a schedule, you take a job whose configuration has a cron
schedule, and run it, using "aurora job create". To stop running it on a
schedule, you use "aurora job kill".
In the proposed change, running a job wouldn't set up a cron schedule: to
schedule a job with a cron field in its configuration, you would run
"aurora cron schedule". To remove it's schedule, you would run "aurora cron
deschedule".
In the current system, it's not just the client that makes the assumption
that a schedule entry is created by running a job with a schedule field;
it's wired into the scheduler itself. That means that in order to make this
cron change, I would have to also make changes to the scheduler.
That's the point of this email. I'd like to propose adding new API methods
to schedule and deschedule cron jobs. For the moment, I'd prefer to leave
the existing API calls alone; we have enough users that are used to the
existing behavior that we need a transition period where the new method is
available, but the existing behaviors continue to work. Once this is
working, we'll do a gradual deprecation (first generating a polite warning,
then an error, and then removing the cron function from the non-cron
scheduler calls.).
I don't want to do this without an OK from the folks working on the
scheduler. What do you all think?
-Mark
