[DOCKER] 20190803_Java & k8s on Azure The story of going to the festival

I've been free for a while. (I forgot to put it together)

20190803 Java & k8s on Azure Festival

Morning session

Container usage scene learned from Java utilization

• Containers have a history --It's getting k8s more than Docker.

• What is microservice for? ――To be competitive with competitors —— Acceptable and unacceptable --Repeated improvement --Minimization of impact --Canary release

• Docker can be used from Windows Server 2016 TP3.

• Dev/Ops --If you want to touch the container, you need to know both

• Container ――You should try to make the image as small as possible.

• open-jdk --Azul Systems zulu can be used.

• Not enough features with docker alone --Not suitable for redundant configurations (difficult to express in complex configurations)

• docker security

• Vulnerability --Use the official image.
  • Aqua Security/Twistlock --It seems that vulnerabilities can be checked

• Is docker-compose unnecessary? ――If you have a complicated system (such as having multiple services), docker-compose will be annoying. ――So, if it's simple, docker-compose is fine, but I'm pressing k8s.

• Do I have to consider the operating environment during production operation? ――I have to do it. --You have to design the container so that it doesn't eat up all your resources.

• Not everything is suitable for containers ――It is suitable for things that scale out. --Not suitable for items that require scale-up. ――After all, it's better to contact directly because you're just borrowing.

• For Azure

• Web App --It seems to be recommended over k8s when there are few services. (The running cost is low.)
• Container Instance --Web App is charged once it is running and it keeps running. --There is a difference that Container Instance is charged only when it is running.

Afternoon

Hands-on materials k8s

• How to install k8s

• GUI --az command --Use resource group templates

• k8s --po is an abbreviation for pod. --The container is not running. ――The reason for this is that you are trying to put together things that have strong dependencies. (Performance is important) --If it falls, the IP will change.

--Can you filter by the labels attribute of the yaml file? --You can operate the pod.

--p.52 apply requires -f --Running on p.58 is 12-Ingress.yaml

• Notes on containers --Do not use the latest tag. --Add build-id to the tag. ――Try to create a small image. --Alpine is the lightest in openjdk. --Set user permissions. ――I don't believe the image of Docker Hub unnecessarily. --Check security before using! --Create a docker file in a way that cache works.

• k8s

  • ReplicationConroller -> Replica Set -> Deployment --Resource usage: Specify properly. (If you do not set it, all will be used up) --You can see it with kubectl top pod. --Understanding Selector / Label is important --You can switch safely --Do not use LB for Service ――It is basically Ingress to publish to the outside ――When in trouble
    • describe, log, exec, get events ――See English for information. --Especially if you apply it to a production environment. (If you want to try it, you can refer to the information in Japanese.) --You don't have to use all the features of k8s --Fast change -Do not make complicated configurations. ――Do not force k8s to do something that can be done by changing the operation configuration. ――Is k8s portable? ――There are some points and some points that are not. --Both LB and PV are implementation-dependent ――Which version are you using? --Large differences between versions (settings, etc.) --Do not configure a large k8s cluster (do not create many nodes) --It's best to keep it at most 20-30 nodes (it takes time to rebuild) --It is recommended to build a new cluster to upgrade k8s. --Do not use PV as much as possible. --DB should use managed services. ――Because there is entanglement with PV?