You are browsing documentation for an older version. See the latest documentation here.

Service Health Probes

Kong Mesh is able to track the status of the Envoy proxy and the underlying app.

Compared to HealthCheck policies, Health Probes have the following advantages:

• knowledge about health is propagated back to kuma-cp and is visible in both in Kong Mesh GUI and Kong Mesh CLI
• scalable with thousands of data plane proxies

Unlike HealthCheck policies, Health Probes:

• only updates when kuma-cp is up and running and the proxy can connect to it.
• doesn’t check connectivity between data plane proxies.

Every inbound in the Dataplane model has a health section:

type: Dataplane
mesh: default
name: web-01
networking:
  address: 127.0.0.1
  inbound:
    - port: 11011
      servicePort: 11012
      health:
        ready: true
      tags:
        kuma.io/service: backend
        kuma.io/protocol: http

This health.ready status is intended to show the status of the endpoint itself. It is set differently depending on the environment (Kubernetes or Universal). It’s treated the same way regardless of the environment:

• if health.ready is true or health section is missing - Kong Mesh considers the inbound as healthy and includes it into load balancing set.
• if health.ready is false - Kong Mesh doesn’t include the inbound into the load balancing set.

Also, health.ready is used to compute the status of the data plane proxy and service. You can see these statuses both in Kong Mesh GUI and Kong Mesh CLI:

Data plane proxy health follows these rules:

• if proxy status is Offline, then data plane proxy is Offline.
• if proxy status is Online:
  • if all inbounds have health.ready=true or no health then data plane proxy is Online
  • if all inbounds have health.ready=false then data plane proxy is Offline
  • if at least one of the inbounds has health.ready=false then data plane proxy is Partially degraded

Service health is computed by aggregating the health of all data plane proxies that have one inbound with a given kuma.io/service and is set this way:

• if all data plane proxies are Online then the service is Online
• if no data plane proxy is Online then the service is Offline
• if at least one of the data plane proxies is Offline are Partially degraded then the service is Partially degraded

Kubernetes

Kong Mesh leverages container statuses within the pod to evaluate dataplane.inbound.health and will follow these rules:

• If the sidecar container is not ready then all inbound will have health=false.
• If the side container is ready then the health of the inbound will be whatever is the readiness of the container which exposes the port used by the inbound.

Kubernetes Probes

Native probe functionalities are not supported until 2.9 release. HTTPGet and gRPC probes always fail when mTLS enabled and TCPSocket probes always succeed regardless of mTLS.

Virtual Probes

For better lifecycle management Kubernetes has readiness, liveness and startup probes.

When using mTLS the container ports are not available outside the pod and Kubernetes will fail to check the probes. To work around this Kong Mesh generates a special non-mTLs listener and overrides the probe definitions in the Pod to proxy them through this sidecar listener. These are called Virtual probes.

This feature currently only supports httpGet probes.

For example, if we specify the following probe:

livenessProbe:
  httpGet:
    path: /metrics
    port: 3001
  initialDelaySeconds: 3
  periodSeconds: 3

Kong Mesh will replace it with:

livenessProbe:
  httpGet:
    path: /3001/metrics
    port: 9000
  initialDelaySeconds: 3
  periodSeconds: 3

Where 9000 is a default virtual probe port, this is configurable:

runtime:
  kubernetes:
    injector:
      virtualProbesPort: 19001

annotations:
  kuma.io/virtual-probes-port: 19001

You can also disable Kong Mesh’s probe virtualization:

runtime:
  kubernetes:
    injector:
      virtualProbesEnabled: false

annotations:
  kuma.io/virtual-probes: disabled

Universal probes

On Universal there is no single standard for probing the service. For health checking, the Dataplane status on Universal Kong Mesh uses Envoy’s Health Discovery Service (HDS). Envoy does health checks and reports the status back to the Kong Mesh Control Plane.

In order to configure health checks for your Dataplane you have to update inbound config with serviceProbe:

type: Dataplane
mesh: default
name: web-01
networking:
  address: 127.0.0.1
  inbound:
    - port: 11011
      servicePort: 11012
      serviceProbe:
        timeout: 2s # optional (default value is taken from KUMA_DP_SERVER_HDS_CHECK_TIMEOUT)
        interval: 1s # optional (default value is taken from KUMA_DP_SERVER_HDS_CHECK_INTERVAL)
        healthyThreshold: 1 # optional (default value is taken from KUMA_DP_SERVER_HDS_CHECK_HEALTHY_THRESHOLD)
        unhealthyThreshold: 1 # optional (default value is taken from KUMA_DP_SERVER_HDS_CHECK_UNHEALTHY_THRESHOLD)
        tcp: {}
      tags:
        kuma.io/service: backend
        kuma.io/protocol: http

If there is a serviceProbe configured for the inbound, Kong Mesh will automatically fill the inbound.health section and update it with the interval equal to KUMA_DP_SERVER_HDS_REFRESH_INTERVAL. Alternatively, it’s possible to omit a serviceProbe section and develop custom automation that periodically updates the health of the inbound.

If the grpc stream with Envoy is disconnected then Kong Mesh considers this proxy offline. It will, however, still advertise inbounds using the final update on their health before disconnection. This is to avoid connection issues between kuma-cp and kuma-dp blocking data plane traffic.

Additionally, when serviceProbe is defined, probes takes into account a health of Envoy. When kuma-dp receives the first shutdown signal, it goes into draining state and all inbounds are considered unhealthy.