GCP HealthCheck customizations lost upgrading to GLBC 0.9.3

[abstrctn]

Our Google Kubernetes Engine cluster masters were recently upgraded to 1.6.4 from 1.6.2 during an automatic update, and we later noticed that some of the backend services created by the GLBC began to report as unhealthy. This seems to correspond with an upgrade in the GLBC to v0.9.3.

We had customized several of the default health checks, changing the request path and / or the Host header. These changes were stable in the past as expected, as the GLBC does not try to update backends that already exist, in order to accommodate the types of overrides we had to make. After the master upgrade, some of these changes disappeared.

Looking through our Stackdriver logs, I saw that a number of httpHealthCheck resources had been deleted right after the master upgrade, and seemingly replaced with HealthCheck resources. I'm wondering if that could have been caused by this section of the implementation of #519? It seems like the new health checks may have been reset to defaults based on the current readiness probe, and not initialized based on the contents of the legacy httpHealthCheck resource.

Is that expected behavior for replacing the legacy health check with the newer default?