In-memory checkpoint simulation

[adrienmaillard]

• Tickets addressed: Closes #1189, Closes #1190
• Review: By commit
• Merge strategy: Merge (no squash)

Description

The first part of this branch has been done by @mattdailis, it

• enables to duplicate simulation engines
• then designs a specific simulate procedure that is able to use the best cached engine to simulate a given plan

Then, I took this work and integrated it into the scheduling agent. Here are some elements of this work:

• Separation between simulation and the computation of simulation results. E.g. rootfinding does not need simulation results (used only in scheduling)
• Partial computation of simulation results for only a subset of the resources (used only in scheduling)
• SimulationFacade has been completely overhauled to account for checkpoint simulation. It now takes care of updating the plan to simulate with child activities and activity durations.
• The scheduler worker now maintain a cache of engines with a configurable size.

Other stuff:

• A little bit of cleaning in the scheduler (DurationExpression mainly)
• In the priority solver, goals and activity expressions now explicitely compute resource dependencies and we use the partial sim computation above + we mock the simulation results when no resource is need, that is, we compute fake simulation results from grounding the schedule into a SimulationResults object.
• The interval of the last profile segment of constraint simulation results become closed-closed to be consistent with what the final state of the simulation engine means. When the engine is finishing at time t, everything at this time has been performed, meaning that the resource value at this time has been computed.

This is not visible in the tests because the metric for simulation performance in scheduling has changed but all of this is greatly decreasing the required simulation time (for the scheduling tests at least...).

Verification

Few tests were added.

Documentation

Needs some documentation, at least for the configuration of cache size.

Future work

• Smarter checkpoint policy
• Incremental simulation

[adrienmaillard]

Matt, I have not rewritten your part of the history. I think there is some stuff to squash, let me know what you want to do.

[adrienmaillard]

Just added breaking change tag because the mission models need to be updated. Custom tasks need to implement the duplicate() method.

[adrienmaillard]

I think I have made all the changes you have asked for. I am wondering about the commit history of this branch, it's pretty horrible at this point, maybe I should merge the whole thing together to avoid polluting Aerie's commit history ?

[Mythicaeda]

Is it feasible to fix the history up at all? If not, then given how big this is I think I'd prefer a bunch of commits than it all just flattened into one mega "Checkpoint Simulation" commit.

[adrienmaillard]

Hey @Mythicaeda I have done all the required changes.

[adrienmaillard]

I adressed 3 of the 4 comments. For the last one, I don't want to change the name of GlobalConstraintWithIntrospection.