Implementation:Kserve Kserve LLMIsvc Controller

Knowledge Sources	KServe KServe LLM Serving
Domains	Kubernetes, Operator_Pattern, LLM_Serving
Last Updated	2026-02-13 00:00 GMT

Overview

Concrete Go controller entry point that registers the LLMISVCReconciler, webhooks, and API conversion handlers.

Description

The LLMIsvc controller is a standalone binary at cmd/llmisvc/main.go. It registers the LLMISVCReconciler, validation webhooks for v1alpha1 and v1alpha2, LLMInferenceServiceConfig webhooks, and conversion webhooks for API version conversion. The cache selector filters resources by the app.kubernetes.io/part-of: llminferenceservice label.

Usage

This binary is deployed as a Deployment in the kserve namespace by the LLMIsvc Helm chart. It runs independently of the main KServe controller manager.

Code Reference

Source Location

Repository: kserve
File: cmd/llmisvc/main.go, Lines 99-262

Signature

// LLMISVCReconciler reconciles LLMInferenceService resources
// Registered at main.go:L202-212
type LLMISVCReconciler struct {
    Client        client.Client
    Clientset     kubernetes.Interface
    EventRecorder record.EventRecorder
    Validator     *webhook.Validator
}

// Webhooks registered:
// L186-190: v1alpha2.LLMInferenceServiceValidator
// L193-197: v1alpha1.LLMInferenceServiceValidator
// L214-221: v1alpha1.LLMInferenceServiceConfigValidator
// L223-230: v1alpha2.LLMInferenceServiceConfigValidator
// L234-246: Conversion webhooks (v1alpha1 ↔ v1alpha2)

// Cache selector: app.kubernetes.io/part-of: llminferenceservice
// Leader election ID: llminferenceservice-kserve-controller-manager

Import

import "github.com/kserve/kserve/cmd/llmisvc"
// This is a standalone binary entry point

I/O Contract

Inputs

Name	Type	Required	Description
LLMInferenceService events	Watch events	Yes	Create/update/delete triggers
LLMInferenceServiceConfig	ConfigMap	Yes	Pod templates and scheduler config

Outputs

Name	Type	Description
vLLM Pods	Pods/Deployment	Model serving containers
InferencePool	CRD	Endpoint pool for Gateway inference extension
HTTPRoute	CRD	Gateway API route to model endpoint
Scheduler	Deployment	Endpoint picker pod
Status	LLMInferenceServiceStatus	Ready conditions and endpoint URL

Usage Examples

Verify Controller Running

# Check controller pod
kubectl get pods -n kserve -l control-plane=llminferenceservice-kserve-controller-manager

# Check CRDs registered
kubectl get crd llminferenceservices.serving.kserve.io
kubectl get crd llminferenceserviceconfigs.serving.kserve.io

# Check leader election
kubectl get lease llminferenceservice-kserve-controller-manager -n kserve

Related Pages

Implements Principle

Principle:Kserve_Kserve_LLMIsvc_Controller_Reconciliation

Requires Environment

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment