OpenShift Service Concepts
External
- https://kubernetes.io/docs/concepts/services-networking/service/
- https://docs.openshift.com/container-platform/latest/architecture/core_concepts/pods_and_services.html#services
Internal
Overview
A service is a Kubernetes primitive representing functionality provided by the Kubernetes cluster, exposed at a permanent IP, hostname and port, for other applications to use. There is no actual process running on any node that can be identified as a service; a service is rather state in the API server. In the vast majority of the cases, the functionality is provided by a group of pods managed by the Kubernetes environment. The pods provide equivalent, replicated functionality, and they may individually, and arbitrarily, come and go. This situation provides the main use case for services, as described in the "Need for Services" section. The relationship between a service and a pod it represents is defined by a label selector declared by the service:
apiVersion: v1
kind: Service
spec:
[...]
selector:
key1: value1
key2: value2
A pod will be identified as being part of the service "cluster" if the label selector declared by the service matches the pod. More details are available in the Service defintion. A service may expose one or more ports.
In some special cases, the service can represent a set of pods running outside the current project, or processes running outside OpenShift altogether. This is known as an external service.
The services constitute the service layer, which is how how applications communicate with one another. All pods that are part of a projects get injected with the address of all other services available in the project as <SVCNAME>_SERVICE_HOST, <SVCNAME>_SERVICE_PORT pairs of environment variables, which facilitates consumption of those services at the published addresses.
Services of a project can be displayed with:
oc get all oc get svc
Service Definition
Need for Services
Pods are transient, they may be started, and more importantly, die at any time, and the IP addresses their endpoints are accessible at are dynamically allocated from the cluster network. This dynamic behavior is hidden behind the service abstraction, that provides the appearance of a stable, consistent service backed by a permanent IP address. A client that needs the functionality provided by the service will contact the well-known address - the service's address - and the service will proxy the call, via a service proxy, to one of the actual pods providing the service. From this perspective, the service acts as an internal load balancer: it identifies a set of replicated pods and then proxies the connections it receives to those pods.
Service Address
Each service has a service IP address and a port, which is allocated from the services subnet. The service IP address is sometimes referred to as the "Cluster IP", which is consistent with the abstraction provided by the service: a cluster of OpenShift-managed pods providing the service:
apiVersion: v1 kind: Service spec: [...] clusterIP: 172.30.73.18 ports: - name: 80-tcp port: 80 protocol: TCP targetPort: 80
This is somewhat confusing, as the pod IPs, which belong to a completely different category of addresses, are said to allocated from the cluster network. The service IP is only available internally in the cluster and it is not routable (in the IP sense) externally. The service can be exposed externally via a route.
Service Endpoints
The EndpointsController component of the API server evaluates its services' selectors continuously, and the information about pods that match those selectors are published as part of the state of an API primitive named Endpoints. The name of the Endpoints object is the same as the name of the corresponding Service:
oc get -o yaml endpoints/serviceA
apiVersion: v1
kind: Endpoints
metadata:
name: serviceA
subsets:
- addresses:
- ip: 10.130.1.0
nodeName: node1
targetRef:
kind: Pod
name: serviceA-1-gr7rh
ports:
- name: 80-tcp
port: 80
protocol: TCP
The <pod IP address>:<port> pairs correspond to processes running inside of a container, on a pod, on a node. The same information can be obtained from the service:
oc describe service <service-name>
Name: logging-kibana ... Endpoints: 10.129.2.17:3000 ...
EndpointsController
The EndpointsController system component associates a service with the endpoints of pods that match the selector. Once the association is done, the actual physical network streaming is performed by a service proxy. More about endpoints is available here: service endpoints.
Service Proxy
The service proxy is a simple network proxy that represents the services defined in the API on the node. Each node runs a service proxy instance. The service proxy does simple TCP and UDP stream forwarding across a set of backends.
SessionAffinity
Ports
Multi-Port Service
External Service
In some special cases, the service can represent a set of pods running outside the current project, or an instance running outside OpenShift altogether. Services representing external resources do not require associated pods, hence do not need a selector. They are called external services. If the selector is not set, the EndpointsController ignores the services, and endpoints can be specified manually. The procedure to declare services from another projects or external service is available here:
Service Dependencies
TO PROCESS: https://blog.giantswarm.io/wait-for-it-using-readiness-probes-for-service-dependencies-in-kubernetes/
Note that "service.alpha.openshift.io/dependencies", sometimes present in Service declarations, do not refer to service dependencies, but rather to GUI details:
apiVersion: v1
kind: Service
metadata:
name: jenkins
annotations:
service.alpha.openshift.io/dependencies: '[{"name": "jenkins-jnlp", "namespace": "", "kind": "Service"}]'
...
See: https://github.com/openshift/origin-web-console/issues/2319