Skip to content

Refactor scheduler to make it more readable #645

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Apr 7, 2025
+592 −450
Merged

Refactor scheduler to make it more readable #645

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 2 additions & 1 deletion pkg/epp/backend/metrics/metrics.go
View file
Original file line number Diff line number Diff line change
@@ -109,6 +109,7 @@ func (p *PodMetricsClientImpl) promToPodMetrics(

if loraMetrics != nil {
updated.ActiveModels = make(map[string]int)
updated.WaitingModels = make(map[string]int)
for _, label := range loraMetrics.GetLabel() {
if label.GetName() == LoraInfoRunningAdaptersMetricName {
if label.GetValue() != "" {
@@ -122,7 +123,7 @@ func (p *PodMetricsClientImpl) promToPodMetrics(
if label.GetValue() != "" {
adapterList := strings.Split(label.GetValue(), ",")
for _, adapter := range adapterList {
updated.ActiveModels[adapter] = 0
updated.WaitingModels[adapter] = 0
}
}
}
11 changes: 7 additions & 4 deletions pkg/epp/backend/metrics/metrics_test.go
View file
Original file line number Diff line number Diff line change
@@ -404,7 +404,8 @@ func TestPromToPodMetrics(t *testing.T) {
expectedMetrics: &Metrics{
WaitingQueueSize: 7,
KVCacheUsagePercent: 0.8,
ActiveModels: map[string]int{"lora1": 0, "lora2": 0, "lora3": 0},
ActiveModels: map[string]int{"lora1": 0, "lora2": 0},
WaitingModels: map[string]int{"lora3": 0},
MaxActiveModels: 3,
},
},
@@ -416,8 +417,8 @@ func TestPromToPodMetrics(t *testing.T) {
KVCacheUtilization: &MetricSpec{MetricName: "vllm_usage"},
LoraRequestInfo: &MetricSpec{MetricName: "vllm:lora_requests_info"},
},
existingMetrics: &Metrics{ActiveModels: map[string]int{}},
expectedMetrics: &Metrics{ActiveModels: map[string]int{}},
existingMetrics: &Metrics{ActiveModels: map[string]int{}, WaitingModels: map[string]int{}},
expectedMetrics: &Metrics{ActiveModels: map[string]int{}, WaitingModels: map[string]int{}},
expectedErr: multierr.Combine(errors.New("metric family \"vllm_waiting\" not found"), errors.New("metric family \"vllm_usage\" not found"), errors.New("metric family \"vllm:lora_requests_info\" not found")),
},
{
@@ -439,7 +440,8 @@ func TestPromToPodMetrics(t *testing.T) {
expectedMetrics: &Metrics{
WaitingQueueSize: 0,
KVCacheUsagePercent: 0.8,
ActiveModels: map[string]int{"lora1": 0, "lora2": 0, "lora3": 0},
ActiveModels: map[string]int{"lora1": 0, "lora2": 0},
WaitingModels: map[string]int{"lora3": 0},
MaxActiveModels: 3,
},
expectedErr: errors.New("metric family \"vllm_waiting\" not found"),
@@ -457,6 +459,7 @@ func TestPromToPodMetrics(t *testing.T) {
existingMetrics: &Metrics{},
expectedMetrics: &Metrics{
ActiveModels: map[string]int{"lora1": 0},
WaitingModels: map[string]int{},
MaxActiveModels: 0, // Should still default to 0.

},
2 changes: 2 additions & 0 deletions pkg/epp/backend/metrics/pod_metrics_test.go
View file
Original file line number Diff line number Diff line change
@@ -44,6 +44,7 @@ var (
"foo": 1,
"bar": 1,
},
WaitingModels: map[string]int{},
}
updated = &Metrics{
WaitingQueueSize: 9999,
@@ -53,6 +54,7 @@ var (
"foo": 1,
"bar": 1,
},
WaitingModels: map[string]int{},
}
)

26 changes: 22 additions & 4 deletions pkg/epp/backend/metrics/types.go
View file
Original file line number Diff line number Diff line change
@@ -41,16 +41,17 @@ type PodMetricsFactory struct {
}

func (f *PodMetricsFactory) NewPodMetrics(parentCtx context.Context, in *corev1.Pod, ds Datastore) PodMetrics {
pod := toInternalPod(in)
pm := &podMetrics{
pmc: f.pmc,
ds: ds,
interval: f.refreshMetricsInterval,
parentCtx: parentCtx,
once: sync.Once{},
done: make(chan struct{}),
logger: log.FromContext(parentCtx),
logger: log.FromContext(parentCtx).WithValues("pod", pod.NamespacedName),
}
pm.pod.Store(toInternalPod(in))
pm.pod.Store(pod)
pm.metrics.Store(newMetrics())

pm.startRefreshLoop()
@@ -77,9 +78,20 @@ func (p *Pod) String() string {
return fmt.Sprintf("%+v", *p)
}

func (p *Pod) Clone() *Pod {
return &Pod{
NamespacedName: types.NamespacedName{
Name: p.NamespacedName.Name,
Namespace: p.NamespacedName.Namespace,
},
Address: p.Address,
}
}

type Metrics struct {
// ActiveModels is a set of models(including LoRA adapters) that are currently cached to GPU.
ActiveModels map[string]int
ActiveModels map[string]int
WaitingModels map[string]int
// MaxActiveModels is the maximum number of models that can be loaded to GPU.
MaxActiveModels int
RunningQueueSize int
@@ -93,7 +105,8 @@ type Metrics struct {

func newMetrics() *Metrics {
return &Metrics{
ActiveModels: make(map[string]int),
ActiveModels: make(map[string]int),
WaitingModels: make(map[string]int),
}
}

@@ -109,8 +122,13 @@ func (m *Metrics) Clone() *Metrics {
for k, v := range m.ActiveModels {
cm[k] = v
}
wm := make(map[string]int, len(m.WaitingModels))
for k, v := range m.WaitingModels {
wm[k] = v
}
clone := &Metrics{
ActiveModels: cm,
WaitingModels: wm,
MaxActiveModels: m.MaxActiveModels,
RunningQueueSize: m.RunningQueueSize,
WaitingQueueSize: m.WaitingQueueSize,
3 changes: 3 additions & 0 deletions pkg/epp/datastore/datastore_test.go
View file
Original file line number Diff line number Diff line change
@@ -236,6 +236,7 @@ var (
"foo": 1,
"bar": 1,
},
WaitingModels: map[string]int{},
}
pod2 = &corev1.Pod{
ObjectMeta: metav1.ObjectMeta{
@@ -250,6 +251,7 @@ var (
"foo1": 1,
"bar1": 1,
},
WaitingModels: map[string]int{},
}
pod1NamespacedName = types.NamespacedName{Name: pod1.Name, Namespace: pod1.Namespace}
pod2NamespacedName = types.NamespacedName{Name: pod2.Name, Namespace: pod2.Namespace}
@@ -305,6 +307,7 @@ func TestMetrics(t *testing.T) {
// Failed to fetch pod2 metrics so it remains the default values.
{
ActiveModels: map[string]int{},
WaitingModels: map[string]int{},
WaitingQueueSize: 0,
KVCacheUsagePercent: 0,
MaxActiveModels: 0,
4 changes: 2 additions & 2 deletions pkg/epp/handlers/request.go
View file
Original file line number Diff line number Diff line change
@@ -27,7 +27,7 @@ import (
"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/pkg/epp/scheduling"
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"
)
@@ -74,7 +74,7 @@ func (s *Server) HandleRequestBody(
return nil, errutil.Error{Code: errutil.BadConfiguration, Msg: fmt.Sprintf("error getting target model name for model %v", modelObj.Name)}
}
}
llmReq := &scheduling.LLMRequest{
llmReq := &schedulingtypes.LLMRequest{
Model: model,
ResolvedTargetModel: modelName,
Critical: datastore.IsCritical(modelObj),
5 changes: 2 additions & 3 deletions pkg/epp/handlers/server.go
View file
Original file line number Diff line number Diff line change
@@ -26,10 +26,9 @@ import (
"google.golang.org/grpc/codes"
"google.golang.org/grpc/status"
"sigs.k8s.io/controller-runtime/pkg/log"
backendmetrics "sigs.k8s.io/gateway-api-inference-extension/pkg/epp/backend/metrics"
"sigs.k8s.io/gateway-api-inference-extension/pkg/epp/datastore"
"sigs.k8s.io/gateway-api-inference-extension/pkg/epp/metrics"
"sigs.k8s.io/gateway-api-inference-extension/pkg/epp/scheduling"
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"
)
@@ -57,7 +56,7 @@ type Server struct {
}

type Scheduler interface {
Schedule(ctx context.Context, b *scheduling.LLMRequest) (targetPod backendmetrics.PodMetrics, err error)
Schedule(ctx context.Context, b *schedulingtypes.LLMRequest) (targetPod schedulingtypes.Pod, err error)
}

func (s *Server) Process(srv extProcPb.ExternalProcessor_ProcessServer) error {
4 changes: 2 additions & 2 deletions pkg/epp/handlers/streamingserver.go
View file
Original file line number Diff line number Diff line change
@@ -37,7 +37,7 @@ import (
backendmetrics "sigs.k8s.io/gateway-api-inference-extension/pkg/epp/backend/metrics"
"sigs.k8s.io/gateway-api-inference-extension/pkg/epp/datastore"
"sigs.k8s.io/gateway-api-inference-extension/pkg/epp/metrics"
"sigs.k8s.io/gateway-api-inference-extension/pkg/epp/scheduling"
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"
)
@@ -343,7 +343,7 @@ func (s *StreamingServer) HandleRequestBody(
return reqCtx, errutil.Error{Code: errutil.BadConfiguration, Msg: fmt.Sprintf("error getting target model name for model %v", modelObj.Name)}
}
}
llmReq := &scheduling.LLMRequest{
llmReq := &schedulingtypes.LLMRequest{
Model: model,
ResolvedTargetModel: modelName,
Critical: datastore.IsCritical(modelObj),
Loading