3

I have several microservices on spring. One of them is config server. I try to start services altogether with docker-compose. But there is problem. Microservices try to get configs from config server before it had been initialized. I want to write script for microservice's startup preventing false start. I have to now how i can determinate moment when my config server is ready. Port listening isn't working. Docker hides info about his networking. I believe there is better way than standard output parsing.

Does somebody know how i can determine time of spring boot application starting?

xpmatteo
  • 11,156
  • 3
  • 26
  • 25
Artyom Karnov
  • 557
  • 9
  • 24

2 Answers2

10

Does somebody know how i can determine time of spring boot application starting?

Your application must be ready when it returns code 200 from /health endpoint with a payload like that

{
  "status": "UP"
}

It means that your application have no issues and is ready to work, anyway this endpoint is normally used by a monitor application to know about your application health that way it can maintain the cluster, this endpoint is not commonly used by others apps, this monitor commonly

  • Uses that endpoint to know when your application is ready to receive world requests, that way it can register your application to a public address for example
  • Try to restart that application expecting it can solve it's problem when /health returns DOWN status

Take a look at docker healthcheck, it uses the same concept as spring

To have that endpoint available you will need to add spring actuator dependency, here is a gradle sample

compile group: 'org.springframework.boot', name: 'spring-boot-starter-actuator', version: '1.5.10.RELEASE'

Microservices try to get configs from config server before it had been initialized ....

Here some important points to let clear

  • Docker compose don't grants startup order unless you are using depends_on clause, anyway docker will never wait to first container completely start up (event using healthcheck) to then start the second container
  • If your microservice A depends microservice B then A must be prepared to deal with B failure and unavailability, it's a microservice premise, because it will just happens, someday or even worse, in a unexpected moment when it is not supposed to. What about config-server restarts in some moment? What will happen with the dependent apps?

So my advise is to you let your application just fail when it tries to get information from config server app, if it fails you can do some things:

  • Do some kind of retry using spring-retry for example
  • If you can deal with some temporary config server unavailability then my suggestion is to add a custom check in your /health exposing this information to some monitor microservice
  • If your app can't work without config server then just call System.exit(-1) and let docker-compose restart your application again and again until config server answer something helpfull

Here some compact example simulating your case and solving that using docker

  • config server
  • app-1

config-server will ever take longer than app-1 to get ready, then app-1 will stay unhealthy until config-server responds correctly

version: '3.4'
services:
  mg-config-server:
    image: nginx:1.10
    healthcheck:
      test: ["CMD", "bash", "-c", "sleep 15; exit 0"]
      interval: 10s
      timeout: 17s
      retries: 3
      start_period: 10s
    networks:
      - my-net
    command: bash -c "echo starting; sleep 20; nginx -g 'daemon off;'"

  mg-app-1:
    image: alpine:3.7
    healthcheck:
      test: ["CMD", "curl", "http://mg-config-server"]
      interval: 5s
      timeout: 5s
      retries: 3
      start_period: 1s
    command: sh -c 'echo starting; apk add --update curl; tail -f /dev/null'
    restart: always
    networks: 
      - my-net
    dns:
    - 8.8.8.8

networks:
  my-net:
    driver: overlay

Then just start it

docker-compose up
docker ps | grep "mg"

Anyway it makes more sense to use docker swarm in that case, because it will check healthcheck endpoints and restart the containers if it's not healthy

docker swarm init --advertise-addr <your-machine-ip>
docker stack deploy --compose-file docker-compose.yml my-stack && docker ps | grep "my-stack"

Docker version: 18.02.0-ce

Sorry for the too long answer, hope it helps

deFreitas
  • 4,196
  • 2
  • 33
  • 43
1

If your issue is specific to Config server, You can also implement spring retry mechanism

spring:
   application:
      name: test-service
   cloud:
      config:
         enabled: true
         uri: ${CONFIG_SERVER_URL:http://127.0.0.1:8761} #where the config-service is running
         fail-fast: true #the service will not run if it can't reach the config-service
         name: common,test-service
         retry:
           max-attempts: 10
           max-interval: 5000

You need to have spring-retry dependency also 

// https://mvnrepository.com/artifact/org.springframework.retry/spring-retry
compile group: 'org.springframework.retry', name: 'spring-retry', version: '1.2.4.RELEASE'
Jebil
  • 1,144
  • 13
  • 25