Health entities, in the context of Service Fabric, can be considered as logical entities used by the runtime. These entities form a logical hierarchy based on the dependencies and interactions among them.

Service Fabric uses a health store to persist the hierarchy of its health entities and their health information. To ensure high availability, health store is implemented as a stateful service within the Service Fabric cluster and is instantiated with when the cluster starts.

The health subsystem leverages the health entities and hierarchy to effectively report, debugg, and monitor. The following diagram illustrates the hierarchy of health entities:

You may notice that the health entities match the Service Fabric deployment entities discussed earlier in this book. Let's dive deeper to understand each of these entities and the parent-child relationships illustrated in the preceding figure.

The root node, Cluster represents the health of Service Fabric cluster. This is an aggregation of conditions that affects the entire cluster and its components. The first child of the cluster, Node represents the health of a Service Fabric node. Health information such as resource usages and connections for a node is captured within this health entity. Applications, the second child of the Cluster, represents a Service Fabric application. The health of all running instances of services in the scope of an application is aggregated into this entity. The Services entity represents all partitions of an application service deployed on the cluster. This entity can be further drilled down to Partitions which represents service partitions and Replicas which represents the replicas of each partition. The Deployed Application and Deployed Service Packages are entities which capture the application and services in scope of a specific node.

This granularity in the hierarchy makes it easy to identify a fault and fix is easily. For example, the cluster will appear unhealthy if one of its application is unhealthy. We can then drill down into the Applications sub tree to detect the service which is unhealthy and then to the partition and so on.

Service Fabric may use internal or external watchdog services to monitor a specific component. At any given point, the state of the health entities will represent near-real-time state of the cluster. A property called health state associated with each entity is used to identify the health of that entity.