feature: concrete handles to specific requested instances when a cluster of instances is created

[shadeofblue]

the most important pain point is that when a cluster is created and/or new instances are added to an existing cluster, the instances of the Service class are not created until the activities are started -> this makes it cumbersome for the client code to identify specific instances added for each request -> and the requirement here is that each newly-created instance or set of instances should be easily attributable to the specific Golem.run_service / Cluster.spawn_instances calls

one suggested solution is that:

• instances of the Service class are created immediately (with some new status -> e.g. pending)
• the call to run_service returns a tuple of Cluster + list of instances
• the call to spawn_instances returns the list of instances ...

CAVEAT

• how do we treat "restart before running" in this scenario? replace the Service inside the ServiceInstance? make ServiceInstance available to the external code and making the aforementioned replacement explicit (e.g. in that, the client code should always retrieve the instance of a Service through some property/helper method of ServiceInstance?) ... ... maybe ServiceInstance could serve a role similar to what a ServiceWrapper currently does in yapapi-service-manager?

possibly also connected: #544

[johny-b]

the call to run_service returns a tuple of Cluster + list of instances

I would rather have:

• run_service return a Cluster
• attribute Cluster.instances with all instances (not only those already starting)

the call to spawn_instances returns the list of instances ...

And here, I would prefer add_instance creating a single instance.

So putting this together, it would be:

cluster = golem.run_service(..., num_instances=2)
instance_0 = cluster.instances[0]
instance_1 = cluster.instances[1]
instance_2 = cluster.add_instance()
assert cluster.instances == [instance_0, instance_1, instance_2]

cluster.stop_instance(instance_0)
assert cluster.instances == [instance_1, instance_2]

[filipgolem]

the call to run_service returns a tuple of Cluster + list of instances

I would rather have:

• run_service return a Cluster
• attribute Cluster.instances with all instances (not only those already starting)

I agree. In any case, there is no need to return a tuple if it can be accessed using the instances() method.

cluster = golem.run_service(..., num_instances=2)
instance_0 = cluster.instances[0]
instance_1 = cluster.instances[1]
instance_2 = cluster.add_instance()
assert cluster.instances == [instance_0, instance_1, instance_2]

cluster.stop_instance(instance_0)
assert cluster.instances == [instance_1, instance_2]

As Service (i.e. an instance) already contains a reference to its Cluster, a better idea is to implement stop for the Service:

- cluster.stop_instance(instance_0)
+ instance_0.stop()

[johny-b]

CURRENT STATE:

• main refactoring is done, there are few things missing (-> check commit)
• restarting - new idea: maybe it could be done in ServiceRunner, but the condition when instance should be restarted would be passed to it? So restarting "logic" is in Cluster, but restarting implementation in ServiceRunner