changes for multilora

Author: kaushikmitr

pkg/epp/scheduling/filter.go

@@ -20,14 +20,10 @@ import (
 	"errors"
 	"math"
 
-<<<<<<< HEAD:pkg/epp/scheduling/filter.go
 	"github.com/go-logr/logr"
-	"sigs.k8s.io/gateway-api-inference-extension/pkg/epp/datastore"
-	logutil "sigs.k8s.io/gateway-api-inference-extension/pkg/epp/util/logging"
-=======
 	klog "k8s.io/klog/v2"
+	"sigs.k8s.io/gateway-api-inference-extension/pkg/epp/datastore"
 	"sigs.k8s.io/gateway-api-inference-extension/pkg/ext-proc/backend"
->>>>>>> 78c0740 (changes for multilora):pkg/ext-proc/scheduling/filter.go
 )
 
 type Filter interface {
@@ -66,14 +62,8 @@ func (f *filter) Name() string {
 	return f.name
 }
 
-<<<<<<< HEAD:pkg/epp/scheduling/filter.go
-func (f *filter) Filter(logger logr.Logger, req *LLMRequest, pods []*datastore.PodMetrics) ([]*datastore.PodMetrics, error) {
-	loggerTrace := logger.V(logutil.TRACE)
-	loggerTrace.Info("Running a filter", "name", f.Name(), "podCount", len(pods))
-=======
 func (f *filter) Filter(req *LLMRequest, pods []*backend.PodMetrics) ([]*backend.PodMetrics, error) {
 	klog.InfoS("Running a filter", "name", f.Name(), "request", req, "podCount", len(pods))
->>>>>>> 78c0740 (changes for multilora):pkg/ext-proc/scheduling/filter.go
 
 	filtered, err := f.filter(logger, req, pods)
 
@@ -86,12 +76,8 @@ func (f *filter) Filter(req *LLMRequest, pods []*backend.PodMetrics) ([]*backend
 		if f.nextOnSuccess != nil {
 			next = f.nextOnSuccess
 		}
-<<<<<<< HEAD:pkg/epp/scheduling/filter.go
-		loggerTrace.Info("Filter succeeded", "filter", f.Name(), "next", next.Name(), "filteredPodCount", len(filtered))
-=======
 		klog.InfoS("Filter succeeded", "filter", f.Name(), "next", next.Name(), "filteredPodCount", len(filtered))
 
->>>>>>> 78c0740 (changes for multilora):pkg/ext-proc/scheduling/filter.go
 		// On success, pass the filtered result to the next filter.
 		return next.Filter(logger, req, filtered)
 	} else {
@@ -102,11 +88,7 @@ func (f *filter) Filter(req *LLMRequest, pods []*backend.PodMetrics) ([]*backend
 		if f.nextOnFailure != nil {
 			next = f.nextOnFailure
 		}
-<<<<<<< HEAD:pkg/epp/scheduling/filter.go
-		loggerTrace.Info("Filter failed", "filter", f.Name(), "next", next.Name())
-=======
 		klog.InfoS("Filter failed", "filter", f.Name(), "next", next.Name())
->>>>>>> 78c0740 (changes for multilora):pkg/ext-proc/scheduling/filter.go
 		// On failure, pass the initial set of pods to the next filter.
 		return next.Filter(logger, req, pods)
 	}
@@ -229,13 +211,8 @@ func loRAAffinityPredicate(req *LLMRequest, pod *datastore.PodMetrics) bool {
 }
 
 // canAcceptNewLoraPredicate is a filter function to check whether a pod has room to load the adapter.
-<<<<<<< HEAD:pkg/epp/scheduling/filter.go
-func canAcceptNewLoraPredicate(req *LLMRequest, pod *datastore.PodMetrics) bool {
-	return len(pod.ActiveModels) < pod.MaxActiveModels
-=======
 func canAcceptNewLoraPredicate(req *LLMRequest, pod *backend.PodMetrics) bool {
 	return len(pod.ActiveModels) < maxLoRACost
->>>>>>> 78c0740 (changes for multilora):pkg/ext-proc/scheduling/filter.go
 }
 
 func criticalRequestPredicate(req *LLMRequest, pod *datastore.PodMetrics) bool {

pkg/epp/scheduling/scheduler.go

@@ -18,12 +18,11 @@ limitations under the License.
 package scheduling
 
 import (
-	"context"
 	"fmt"
 	"math/rand"
 
 	"github.com/go-logr/logr"
-	"sigs.k8s.io/controller-runtime/pkg/log"
+	klog "k8s.io/klog/v2"
 	"sigs.k8s.io/gateway-api-inference-extension/pkg/epp/datastore"
 	errutil "sigs.k8s.io/gateway-api-inference-extension/pkg/epp/util/error"
 	logutil "sigs.k8s.io/gateway-api-inference-extension/pkg/epp/util/logging"
@@ -122,26 +121,14 @@ type Scheduler struct {
 }
 
 // Schedule finds the target pod based on metrics and the requested lora adapter.
-<<<<<<< HEAD:pkg/epp/scheduling/scheduler.go
-func (s *Scheduler) Schedule(ctx context.Context, req *LLMRequest) (targetPod datastore.PodMetrics, err error) {
-	logger := log.FromContext(ctx).WithValues("request", req)
-	podMetrics := s.datastore.PodGetAll()
-	logger.V(logutil.VERBOSE).Info("Scheduling a request", "metrics", podMetrics)
-	pods, err := s.filter.Filter(logger, req, podMetrics)
-=======
 func (s *Scheduler) Schedule(req *LLMRequest) (targetPod backend.Pod, err error) {
 	klog.InfoS("Scheduling a request", "request", req, "metrics", s.podMetricsProvider.AllPodMetrics())
 	pods, err := s.filter.Filter(req, s.podMetricsProvider.AllPodMetrics())
->>>>>>> 78c0740 (changes for multilora):pkg/ext-proc/scheduling/scheduler.go
 	if err != nil || len(pods) == 0 {
 		return datastore.PodMetrics{}, fmt.Errorf(
 			"failed to apply filter, resulted %v pods, this should never happen: %w", len(pods), err)
 	}
-<<<<<<< HEAD:pkg/epp/scheduling/scheduler.go
-	logger.V(logutil.VERBOSE).Info("Selecting a random pod from the candidates", "candidatePods", pods)
-=======
 	klog.InfoS("Selecting a random pod from the candidates", "candidatePods", pods)
->>>>>>> 78c0740 (changes for multilora):pkg/ext-proc/scheduling/scheduler.go
 	i := rand.Intn(len(pods))
 	return *pods[i], nil
 }