move inf model IsCritial func out of datastore (kubernetes-sigs#670)

nirrozenbaum · web-flow · commit 4761c71b91b3 · 2025-04-09T09:46:40.000-07:00
* move inf model IsCritial func out of datastore

Signed-off-by: Nir Rozenbaum &lt;nirro@il.ibm.com&gt;

* remove IsCritical function helper function

Signed-off-by: Nir Rozenbaum &lt;nirro@il.ibm.com&gt;

---------

Signed-off-by: Nir Rozenbaum &lt;nirro@il.ibm.com&gt;
diff --git a/pkg/epp/datastore/datastore.go b/pkg/epp/datastore/datastore.go
@@ -69,7 +69,7 @@ type Datastore interface {
 	Clear()
 }
 
-func NewDatastore(parentCtx context.Context, pmf *backendmetrics.PodMetricsFactory) *datastore {
+func NewDatastore(parentCtx context.Context, pmf *backendmetrics.PodMetricsFactory) Datastore {
 	store := &datastore{
 		parentCtx:       parentCtx,
 		poolAndModelsMu: sync.RWMutex{},
@@ -302,10 +302,3 @@ func stripLabelKeyAliasFromLabelMap(labels map[v1alpha2.LabelKey]v1alpha2.LabelV
 	}
 	return outMap
 }
-
-func IsCritical(model *v1alpha2.InferenceModel) bool {
-	if model.Spec.Criticality != nil && *model.Spec.Criticality == v1alpha2.Critical {
-		return true
-	}
-	return false
-}
diff --git a/pkg/epp/handlers/request.go b/pkg/epp/handlers/request.go
@@ -26,7 +26,7 @@ import (
 	extProcPb "github.com/envoyproxy/go-control-plane/envoy/service/ext_proc/v3"
 	"google.golang.org/protobuf/types/known/structpb"
 	"sigs.k8s.io/controller-runtime/pkg/log"
-	"sigs.k8s.io/gateway-api-inference-extension/pkg/epp/datastore"
+	"sigs.k8s.io/gateway-api-inference-extension/api/v1alpha2"
 	schedulingtypes "sigs.k8s.io/gateway-api-inference-extension/pkg/epp/scheduling/types"
 	errutil "sigs.k8s.io/gateway-api-inference-extension/pkg/epp/util/error"
 	logutil "sigs.k8s.io/gateway-api-inference-extension/pkg/epp/util/logging"
@@ -77,7 +77,7 @@ func (s *Server) HandleRequestBody(
 	llmReq := &schedulingtypes.LLMRequest{
 		Model:               model,
 		ResolvedTargetModel: modelName,
-		Critical:            datastore.IsCritical(modelObj),
+		Critical:            modelObj.Spec.Criticality != nil && *modelObj.Spec.Criticality == v1alpha2.Critical,
 	}
 	loggerVerbose.Info("LLM request assembled", "request", llmReq)
 
diff --git a/pkg/epp/handlers/streamingserver.go b/pkg/epp/handlers/streamingserver.go
@@ -348,7 +348,7 @@ func (s *StreamingServer) HandleRequestBody(
 	llmReq := &schedulingtypes.LLMRequest{
 		Model:               model,
 		ResolvedTargetModel: modelName,
-		Critical:            datastore.IsCritical(modelObj),
+		Critical:            modelObj.Spec.Criticality != nil && *modelObj.Spec.Criticality == v1alpha2.Critical,
 	}
 	logger.V(logutil.DEBUG).Info("LLM request assembled", "model", llmReq.Model, "targetModel", llmReq.ResolvedTargetModel, "critical", llmReq.Critical)
 

Original file line number	Diff line number	Diff line change
`@@ -69,7 +69,7 @@ type Datastore interface {`
`69`	`69`	`Clear()`
`70`	`70`	`}`
`71`	`71`
`72`		`-func NewDatastore(parentCtx context.Context, pmf backendmetrics.PodMetricsFactory) datastore {`
	`72`	`+func NewDatastore(parentCtx context.Context, pmf *backendmetrics.PodMetricsFactory) Datastore {`
`73`	`73`	`store := &datastore{`
`74`	`74`	`parentCtx: parentCtx,`
`75`	`75`	`poolAndModelsMu: sync.RWMutex{},`
`@@ -302,10 +302,3 @@ func stripLabelKeyAliasFromLabelMap(labels map[v1alpha2.LabelKey]v1alpha2.LabelV`
`302`	`302`	`}`
`303`	`303`	`return outMap`
`304`	`304`	`}`
`305`		`-`
`306`		`-func IsCritical(model *v1alpha2.InferenceModel) bool {`
`307`		`- if model.Spec.Criticality != nil && *model.Spec.Criticality == v1alpha2.Critical {`
`308`		`- return true`
`309`		`- }`
`310`		`- return false`
`311`		`-}`
Original file line number	Diff line number	Diff line change
`@@ -348,7 +348,7 @@ func (s *StreamingServer) HandleRequestBody(`
`348`	`348`	`llmReq := &schedulingtypes.LLMRequest{`
`349`	`349`	`Model: model,`
`350`	`350`	`ResolvedTargetModel: modelName,`
`351`		`- Critical: datastore.IsCritical(modelObj),`
	`351`	`+ Critical: modelObj.Spec.Criticality != nil && *modelObj.Spec.Criticality == v1alpha2.Critical,`
`352`	`352`	`}`
`353`	`353`	`logger.V(logutil.DEBUG).Info("LLM request assembled", "model", llmReq.Model, "targetModel", llmReq.ResolvedTargetModel, "critical", llmReq.Critical)`
`354`	`354`