How does GCP decides when a Kubernetes deployment is completed

[Luca Lupo]

Hi There,

I have a few Java microservices and when I redeploy them, one of them needs like 40 seconds after starting up (it's a dropwizard application) to load data and do some computation to actually be ready to serve requests. This extra time makes so GCP thinks the service is ready and keeps deploying other instances and when the service receives requests they fail.

How does GCP determins that a Kubernetes app is ready to serve request? Is there a way to prologue the deployment state or tell GCP that it's not finished until say, an API returns a 200?

Thanks

[Andrew Cooper]

You will find this documentation interesting as it advises how to implement and configure "readiness and startup" probes for your containers. The "liveness" probe can check a container or when an application has stalled and does not accept traffic. This approach can be customised to suit your requirements rather than relying on a general catch all signal.