Revisit parallelism in solution reconciler

[msftcoderdjw]

https://github.com/eclipse-symphony/symphony/blob/17392d8c1b1decc110ca3cef28a8675e4dd45c9f/api/pkg/apis/v1alpha1/managers/solution/solution-manager.go#L220-L222

Now global lock is used in reconcile function.

[Haishi2016]

Several thoughts: 1) we may want to consider using an Actor framework (such as Akka or Orleans) to allow individual instance/target to drive towards its own desired state in parallel; 2) for parallelism of job manager, we should use platform to scale it out - such as putting job vendor in a separate pod and scale the pod to multiple instances, which essentially implement the competing consumer pattern. 3) Also note that the solution manager is also used on agent side for reconciliation.

[msftcoderdjw]

Several thoughts: 1) we may want to consider using an Actor framework (such as Akka or Orleans) to allow individual instance/target to drive towards its own desired state in parallel; 2) for parallelism of job manager, we should use platform to scale it out - such as putting job vendor in a separate pod and scale the pod to multiple instances, which essentially implement the competing consumer pattern. 3) Also note that the solution manager is also used on agent side for reconciliation.

I agree. I think we can divide the fixes into several phases.

1. Short term (single pod): we need to use fine-granularity lock (lock per deployment id) to replace the global lock in solution-manager. Then we can make sure multi deployments can be reconciled concurrently. This is also applicable for agent side in standalone deployment.
2. Long term (multiple pods): consider evaluating actor framework, such as Akka or Orleans. (The alternative is to leverage distributed lock when scaling out the pod. I don't prefer to the distributed lock if the mature actor framework can avoid this.)