Zero Down Time Deployment in Kubernetes
Generally speaking, a deployment consists of 3 steps: 1. Stop 2. Deploy 3. Start When it comes to zero-down-time deployment, we need to fulfill more details.
Stop Stop routing new requests Consume ongoing requests Transfer state for stateful application Deploy Deploy new version of application Start Warm up (prepare data, transfer state back, JVM warm-up etc) Start to receive requests
In the Kubernetes environment, things are relatively easy. readinessProbe is the major feature we need.
Stop
Your application should handle the exit signal. When app starts to stop, fail the readinessProbe. Therefore k8s knows an instance is not functioning, no more request will be routed to your instance that gonna upgrade.
Existing requests are able to continue until app completes the whole task. Be aware of the grace period, if request take too long to complete, the instance will be force killed, which will lead to service outage.
Deploy
Update your deployment or run a helm upgrade
Start
Set readiness to true when everything is ready.
