Kubernetes Events for Scaling Decisions

[ev-shindin]

Emit Kubernetes events on VariantAutoscaling resources for key scaling decisions and errors. Currently only ServiceMonitorDeleted warning events are emitted.

Events:

Event	Type	Reason	When
Scale-up decision	Normal	ScaledUp	Target replicas increased
Scale-down decision	Normal	ScaledDown	Target replicas decreased
Limiter constrained	Warning	ResourceConstrained	Decision was limited by GPU availability
Metrics unavailable	Warning	MetricsUnavailable	Prometheus query failed or returned no data
Scale-to-zero applied	Normal	ScaledToZero	Enforcer set replicas to 0 (no active requests)
Optimization error	Warning	OptimizationFailed	Analysis/optimizer error for this VA

Implementation:

• Use r.Recorder.Eventf() in applySaturationDecisions() and the reconciler
• Rate-limit: at most 1 event per VA per optimization cycle to avoid flooding the API server
• Include relevant context in event message (e.g., "Scaled up from 2 to 4 replicas, saturation=0.85")

Acceptance Criteria:

• 6 event types emitted on appropriate conditions
• Events visible via kubectl describe va <name>
• Events are rate-limited (no flooding on rapid cycles)
• Unit/integration tests verify event emission

[lionelvillard]

@shuynh2017 FYI (as discussed).

[shuynh2017]

@lionelvillard @ev-shindin, can we consider also having these events to be sent to OTEL events if OTEL endpoint is specified?