Stackdriver logging doesn't always have to be done on Google Cloud Console

I have a relatively large number of services running under the Kubernetes / Google Container Engine, with a cluster created through the Google Cloud Console UI. In containers, I register Java services in JSON format.

Most of the time it just works, but lately more and more logs "stop" at a seemingly random point in time, with the last entry (at any log level) from hours or days ago.

I am not changing any parameters and have not been able to isolate the reason for this. What's going on here? What can I do to improve and / or fix this problem?