Kubernetes three dozen probes and probe mode

One 、 Introduction to probe background

1、 Orchestration tool runtime , although pod Hang up will restart under the scheduling of the controller , appear pod When restarting , however pod Status is running, Can't really react at that time pod A healthy state , We can go through Kubernetes Probe monitoring of pod In real time .

Two 、Kubernetes Three probe types

Start the probe （Startup Probe）： Judge whether the application in the container is started and completed （ Before starting the probe to judge success , Ready probes and survival probes will not execute ）

Ready probe （Readiness Probe）： Determine whether the container is ready , If not ready , The container will be in the not ready state , Containers that are not ready , No traffic scheduling .

Survival probe （Liveness Probe）： Determine whether the application in the container is normal , If it's not normal ,K8s Will restart the container .

3、 ... and 、 Three ways of probe

exec： By executing the specified command in the container , To determine the status code returned when the command exits , If 0 Is normal .

httpGet： Through the of the container IP Address 、 Port and URL Path to send GET request ; If the status code of the response is 200 ~ 399 between , Is normal .

tcpSocket： Through the of the container IP Address and specified port , Conduct TCP Check , If the port is open , Is normal .

Four 、 Configuration item

initialDelaySeconds： Wait for the time we define After that, start the probe inspection ;

periodSeconds： The probe's Time interval between ;

timeoutSeconds： The probe's Timeout time , When we exceed the time we define , It will be regarded as a failure ;

successThreshold： The probe's Minimum number of consecutive successes ;

failureThreshold： The probe's Minimum number of consecutive failures ;

Now we use different probe methods for these three probes , Mainly for the convenience of giving you a simple introduction , It's not fixed writing ; meanwhile , These three probes can be used together ; The more common is Ready probe and survival probe are used together （ Start the probe at K8s 1.6 Added after version

5、 ... and 、 Introduction to the use of probe

1） Start the probe

apiVersion: v1
kind: Pod
metadata:
  name: nginx-start-up
  namespace: probe
spec:
  containers:
  - name: nginx-start-up
    image: nginx:latest
    ports:
    - containerPort: 80
    startupProbe:
      failureThreshold: 3  # Failed three times, the probe failed 
      exec:
        command: ['/bin/sh','-c','echo Hello World']
      initialDelaySeconds: 20  # Delay 20s After that, conduct the first probe 
      periodSeconds: 3 # interval 3s Conduct a probe 
      successThreshold: 1 # Calculate the probe successfully once ok
      timeoutSeconds: 2 # Overtime 2s It's a failure 

2） Ready probe

apiVersion: v1
kind: Pod
metadata:
  name: nginx-ready
  namespace: probe
  labels:
    app: nginx-ready   # Verify the key parameters of the ready probe 
spec:
  containers:
  - name: nginx-ready
    image: nginx:latest
    ports:
    - containerPort: 80
    readinessProbe:
      failureThreshold: 3
      tcpSocket:
        port: 80
      initialDelaySeconds: 20
      periodSeconds: 3
      successThreshold: 1
      timeoutSeconds: 2

Port detection is normal ：

Write a nodeport Type for flow verification

apiVersion: v1
kind: Service
metadata:
  name: ready-nodeport
  labels:
    name: ready-nodeport
  namespace: probe
spec:
  type: NodePort
  ports:
  - port: 88
    protocol: TCP
    targetPort: 80
    nodePort: 30880
  selector:
    app: nginx-ready

The result of the simulated flow is ok Of

Now modify the ready probe tcpSocket The port of the detection mode is 81 Failed to simulate the probe , Will the flow reach this pod

Show 81 Port failure ,ready Status as 0, but pod yes running Of , The result of the request is as follows , Traffic injection failed . Ready probe failed , I won't give it pod Input flow .

3） Survival probe

apiVersion: v1
kind: Pod
metadata:
  name: nginx-live
  namespace: probe
  labels:
    app: nginx-live
spec:
  containers:
  - name: nginx-live
    image: nginx:latest
    ports:
    - containerPort: 80
    livenessProbe:
      failureThreshold: 3
      httpGet:
        path: /
        port: 80
        scheme: HTTP
      initialDelaySeconds: 20
      periodSeconds: 3
      successThreshold: 1
      timeoutSeconds: 2

Generate a nodeport type service

apiVersion: v1
kind: Service
metadata:
  name: live-nodeport
  labels:
    name: live-nodeport
  namespace: probe
spec:
  type: NodePort
  ports:
  - port: 89
    protocol: TCP
    targetPort: 80
    nodePort: 30810
  selector:
    app: nginx-live

curl Conduct ip Port verification

modify httpget in port Parameter values for 81, Simulated survival probe failed , Probe failed , Will make pod To restart .

6、 ... and 、 Probe recovery

Correct probe detection port , Reload yaml file , Normal flow .