Skip to main content

Scheduling

Procedural scheduling goals are very similar to procedural constraints, except that where constraints report problems, goals fix them. You do that by implement the Goal interface and interacting with the provided EditablePlan object.

EditablePlan

The EditablePlan gives you the same interface as Plan, but also allows you to add new activities. You do this with plan.create(...). You can either provide the minimal amount of information (type, arguments, and start time), or provide a NewDirective object that is a little more configurable.

The create function will tentatively add your directive to the plan in an uncommitted state, and return an ActivityDirectiveId object for the activity you just added. This way you can anchor another activity to the one you just created.

danger

Activity directive IDs are only accurate for activities that already exist in the plan. For newly created activities, plan.create returns an estimated placeholder ID, which will almost certainly change when it is uploaded to the database. So if your goal has any other side effects, do not rely on the accuracy of these IDs.

If you then call plan.directives(...), you'll see the activity you just created now included in the plan, but you're not quite done. These new activities are in an uncommitted state, and you must either call plan.commit() or plan.rollback() before the goal returns. rollback will reset the plan to the last time you called commit, or to the beginning of the goal. We use this system so you can add some activities to the plan, test it out with simulation, and then choose to keep them or not depending on the simulation results.

Simulation

To try out your changes, you can call simResults = plan.simulate(). This will simulate the whole plan from start to finish. This returns the same SimulationResults object used in constraints.

tip

If you run simulation multiple times, you might have multiple SimulationResults objects floating around; it is up to you to keep those straight and manage that state. Most goals that simulate will just have a single mutable simResults variable that you update with each simulation, ensuring that the old results are lost and won't mess up your state.

You can configure the simulation by passing a SimulateOptions object into plan.simulate(...). Currently, the only option is to set when the simulation will pause. If you pause partway through the plan, then make a change after the pause time, if you simulate again it will attempt to pick up where it left off rather than start over.

Stale sim results

Some goals will want to get simulation results cheaply without simulating, at the risk of unsoundness. You can do this by calling plan.latestResults() instead of .simulate(). This will return whatever was simulated last, even if it didn't simulate the whole plan, even if modifications were made later. There may be mismatches between the plan state and what was actually simulated; it is up to you to either accept this risk, or check its accuracy.

It is possible for latestResults to return null, if the plan has never been simulated. So it is recommended to start such goals with:

val simResults = plan.latestResults() ?: plan.simulate()

Simulate After

All goals have a flag stored outside the goal definition called "Simulate After". If enabled, a new simulation will be run when the goal finishes. This exists to enable some simulation control in eDSL goals, but is applicable to all goals because procedural and eDSL goal may need to be interleaved in a scheduling specification.

danger

Even if you write a procedural goal that does not simulate and uses stale results, simulation may still be run after your goal anyway if the "Simulate After" flag is checked in the UI.

Additionally, if "Simulate After" is enabled on the last goal in the spec, the scheduler will simulate one last time and upload those results to the database. If not, it will just upload the latest stale results.

Idempotency

Most scheduling goals will be written to detect a problem in the plan that can be solved by adding activities, and then add them. But it's important to keep in mind that you have to actually check for the problem. For example, if I want a MyActivity directive at every hour on the plan, I could easily write it like this:

@SchedulingProcedure
class MyActivityEveryHour: Goal {
override fun run(plan: EditablePlan) {
for (time in plan.totalBounds() step Duration.HOUR) {
plan.create(
"MyActivity",
DirectiveStart.Absolute(time),
mapOf() // assuming it takes no arguments
)
}

plan.commit()
}
}

This is essentially quick-and-dirty procedural version of the eDSL's Recurrence Goal.

However, every time this goal is run, it will unconditionally create a new series of activities, so you'd have to be careful to disable the goal from your scheduling spec afterward, or otherwise ensure that scheduling will only ever be run once. This is not a requirement that most missions can impose on operations.

So instead, you have to first check if the goal is already satisfied. A rudimentary version of this could look like:

@SchedulingProcedure
class MyActivityEveryHour: Goal {
override fun run(plan: EditablePlan) {
// This produces a Booleans profile that is true at the instant of a MyActivity directive.
val existingActivities = plan.directives("MyActivity").active().cache()

for (time in plan.totalBounds() step Duration.HOUR) {
if (!existingActivities.sample(time)) plan.create(
"MyActivity",
DirectiveStart.Absolute(time),
mapOf() // assuming it takes no arguments
)
}

plan.commit()
}
}

This will check only for MyActivity directives that happen at the exact time it wants to place a new one. This is better than nothing, but still not very smart. If a planner manually inserted a MyActivity at a time that didn't exactly line up, this check would ignore it.

Maybe you would want the goal to ensure that there is at most one hour between activities, reacting to existing activities by detecting directives that don't land exactly on the hour and proceeding to a hour after that directive. This is a pretty common pattern that can be accomplished with a while loop, repeatedly popping off the first directive in the list of existing MyActivitys, and is for now left as an exercise for the user.

In the future we plan to provide helper functions that accomplish common patterns like this.