This change adds the AlertActionManager and AlertActionEntry. The old AlertActionManager has become the AlertActionRegistry.
This means that now the results of Alerts are queued up in a job queue and executed in separate threads.
The AlertActionManager is a composite member of the AlertManager.
Change the BasicTest to just run on a single node to fix the action registration if the action happens on a different node.
Threads are not directly constructed but now the threadpool is used.
The ClusterStateListener in AlertManager is responsible now for starting the job queue.
Original commit: elastic/x-pack-elasticsearch@a73c6b60f8
* Using cluster state listener to clear alerts when .alerts index is removed. (when running on multiple nodes the .alerts index is scattered so indices listener doesn't work).
* Remove the starting / loading thread and move the initial loading to cluster state listener.
Original commit: elastic/x-pack-elasticsearch@b8f41db2ea
* Made use of IndicesLifecycle, to catch when the .alerts index gets deleted, so we can clear the alerts
* Moved to concurrent hashmap in favour over normal hashmap with synchronized blocks
Original commit: elastic/x-pack-elasticsearch@5599d01c78
This commit renames the alerting package to alerts and will create the
alerts index on addAlert if needed.
Original commit: elastic/x-pack-elasticsearch@7cd691bd9b
