kubeflow / trainer / 33003377893

70%

push

github

fix(controller): apply configured health probe endpoint names (#3886)

* fix(operator): apply configured health probe endpoint names

The config API, the base manifests, and the Helm chart all expose
readinessEndpointName and livenessEndpointName, but the manager never received
them, so the probes stayed on controller-runtime's own /readyz and /healthz. A
deployment that renamed an endpoint and pointed its probe at the new path got a
404 back and a container that never turned ready.

Every place these names are configured spells them without a leading slash,
while controller-runtime registers them as http.ServeMux patterns, which panic
on a pattern that has none. They are rooted on the way through so the shipped
defaults keep serving where the probes already look, and the startup log now
reports the paths that were registered rather than the two it assumed.

Signed-off-by: rjgoyln <pt40419@gmail.com>

* fix(operator): report probe paths from the manager options

The paths the probes are registered on come from the options the manager is
built with, so the startup log reads them from there rather than from strings
threaded separately through the call.

Signed-off-by: rjgoyln <pt40419@gmail.com>

---------

Signed-off-by: rjgoyln <pt40419@gmail.com>

10 of 10 new or added lines in 1 file covered. (100.0%)

2625 of 3831 relevant lines covered (68.52%)

0.8 hits per line