Operator sets HTTPS on the Engine's liveness and ready checks

[adriangonz]

Describe the bug

When the user chooses to deploy their models with the Java engine, the operator sets the container's liveness and ready checks to be called with HTTPS. Since the engine gets exposed through HTTP, this means that it never becomes "ready" and the model never gets deployed.

In other words, the generated container for the engine, ends up with this config:

seldon-container-engine:
	Liveness:   http-get https://:admin/live delay=20s timeout=60s period=5s #success=1 #failure=3
	Readiness:  http-get https://:admin/ready delay=20s timeout=60s period=5s #success=1 #failure=3

This seems to be happening in these lines:

seldon-core/operator/controllers/seldondeployment_engine.go

Lines 347 to 358 in 5b1b2b1

	ReadinessProbe: &corev1.Probe{Handler: corev1.Handler{HTTPGet: &corev1.HTTPGetAction{Port: intstr.FromString("admin"), Path: "/ready", Scheme: corev1.URISchemeHTTPS}},
	InitialDelaySeconds: 20,
	PeriodSeconds: 5,
	FailureThreshold: 3,
	SuccessThreshold: 1,
	TimeoutSeconds: 60},
	LivenessProbe: &corev1.Probe{Handler: corev1.Handler{HTTPGet: &corev1.HTTPGetAction{Port: intstr.FromString("admin"), Path: "/live", Scheme: corev1.URISchemeHTTPS}},
	InitialDelaySeconds: 20,
	PeriodSeconds: 5,
	FailureThreshold: 3,
	SuccessThreshold: 1,
	TimeoutSeconds: 60},

To reproduce

1. Deploy a model using the Java engine
2. Check the generated pod spec

Expected behaviour

The liveness and readiness probes should be configured with HTTP.

Environment

• Seldon Core: 1.3.0 (not sure when it started happening)

[adriangonz]

/priority p0