250717 sync master

.flake8

@@ -4,6 +4,8 @@ extend-ignore = E203,E501,E701, B017
 exclude =
   .venv,
   venv,
+  prod_venv,
+  python/type=local/*,
   python/kserve/kserve/exceptions.py,
   python/kserve/kserve/configuration.py,
   python/kserve/kserve/rest.py,

Dockerfile

@@ -1,7 +1,6 @@
 # Build the manager binary
 # Upstream already is on go 1.24, however there is no gotoolset for 1.24 yet.
-# TODO move to ubi9/go-toolset:1.24 when available
-FROM registry.access.redhat.com/ubi9/go-toolset:1.23 as builder
+FROM registry.access.redhat.com/ubi9/go-toolset:1.24 as builder
 
 # Copy in the go src
 WORKDIR /go/src/github.com/kserve/kserve

Makefile

@@ -139,7 +139,7 @@ poetry-lock: $(POETRY)
 # Update the kserve package first as other packages depends on it.
 	cd ./python && \
 	cd kserve && $(POETRY) lock --no-update && cd .. && \
-	for file in $$(find . -type f -name "pyproject.toml" -not -path "./pyproject.toml" -not -path "*.venv/*"); do \
+	for file in $$(find . -type f -name "pyproject.toml" -not -path "./pyproject.toml" -not -path "*.venv/*" -not -path "*/prod_venv/" -not -path "*/site-packages/*" -not -path "*/__pypackages__/*" -not -path "*/build/*"); do \
 		folder=$$(dirname "$$file"); \
 		echo "moving into folder $$folder"; \
 		case "$$folder" in \

agent.Dockerfile

@@ -1,7 +1,7 @@
 # Build the inference-agent binary
 # Upstream already is on go 1.24, however there is no gotoolset for 1.24 yet.
 # TODO move to ubi9/go-toolset:1.24 when available
-FROM registry.access.redhat.com/ubi9/go-toolset:1.23 AS builder
+FROM registry.access.redhat.com/ubi9/go-toolset:1.24 AS builder
 
 # Copy in the go src
 WORKDIR /go/src/github.com/kserve/kserve

charts/kserve-crd-minimal/templates/serving.kserve.io_llminferenceserviceconfigs.yaml

@@ -0,0 +1,32 @@
+apiVersion: apiextensions.k8s.io/v1
+kind: CustomResourceDefinition
+metadata:
+  annotations:
+    controller-gen.kubebuilder.io/version: v0.16.2
+  name: llminferenceserviceconfigs.serving.kserve.io
+spec:
+  group: serving.kserve.io
+  names:
+    kind: LLMInferenceServiceConfig
+    listKind: LLMInferenceServiceConfigList
+    plural: llminferenceserviceconfigs
+    singular: llminferenceserviceconfig
+  scope: Namespaced
+  versions:
+  - name: v1alpha1
+    schema:
+      openAPIV3Schema:
+        properties:
+          apiVersion:
+            type: string
+          kind:
+            type: string
+          metadata:
+            type: object
+          spec:
+            type: object
+            x-kubernetes-map-type: atomic
+            x-kubernetes-preserve-unknown-fields: true
+        type: object
+    served: true
+    storage: true

charts/kserve-crd-minimal/templates/serving.kserve.io_llminferenceservices.yaml

@@ -0,0 +1,57 @@
+apiVersion: apiextensions.k8s.io/v1
+kind: CustomResourceDefinition
+metadata:
+  annotations:
+    controller-gen.kubebuilder.io/version: v0.16.2
+  name: llminferenceservices.serving.kserve.io
+spec:
+  group: serving.kserve.io
+  names:
+    kind: LLMInferenceService
+    listKind: LLMInferenceServiceList
+    plural: llminferenceservices
+    shortNames:
+    - llmisvc
+    singular: llminferenceservice
+  scope: Namespaced
+  versions:
+  - additionalPrinterColumns:
+    - jsonPath: .status.url
+      name: URL
+      type: string
+    - jsonPath: .status.conditions[?(@.type=='Ready')].status
+      name: Ready
+      type: string
+    - jsonPath: .status.conditions[?(@.type=='Ready')].reason
+      name: Reason
+      type: string
+    - jsonPath: .metadata.creationTimestamp
+      name: Age
+      type: date
+    - jsonPath: .status.addresses[*].url
+      name: URLs
+      priority: 1
+      type: string
+    name: v1alpha1
+    schema:
+      openAPIV3Schema:
+        properties:
+          apiVersion:
+            type: string
+          kind:
+            type: string
+          metadata:
+            type: object
+          spec:
+            type: object
+            x-kubernetes-map-type: atomic
+            x-kubernetes-preserve-unknown-fields: true
+          status:
+            type: object
+            x-kubernetes-map-type: atomic
+            x-kubernetes-preserve-unknown-fields: true
+        type: object
+    served: true
+    storage: true
+    subresources:
+      status: {}

cmd/agent/main.go

@@ -63,6 +63,8 @@ var (
 	workers             = flag.Int("workers", 5, "Number of workers")
 	sourceUri           = flag.String("source-uri", "", "The source URI to use when publishing cloudevents")
 	logMode             = flag.String("log-mode", string(v1beta1.LogAll), "Whether to log 'request', 'response' or 'all'")
+	logStorePath        = flag.String("log-store-path", "", "The path to the log output")
+	logStoreFormat      = flag.String("log-store-format", "json", "Format for log output, 'json' or 'yaml'")
 	inferenceService    = flag.String("inference-service", "", "The InferenceService name to add as header to log events")
 	namespace           = flag.String("namespace", "", "The namespace to add as header to log events")
 	endpoint            = flag.String("endpoint", "", "The endpoint name to add as header to log events")
@@ -153,7 +155,7 @@ func main() {
 	var loggerArgs *loggerArgs
 	if *logUrl != "" {
 		logger.Info("Starting logger")
-		loggerArgs = startLogger(*workers, logger)
+		loggerArgs = startLogger(*workers, logStorePath, logStoreFormat, logger)
 	}
 
 	var batcherArgs *batcherArgs
@@ -264,18 +266,18 @@ func startBatcher(logger *zap.SugaredLogger) *batcherArgs {
 	}
 }
 
-func startLogger(workers int, logger *zap.SugaredLogger) *loggerArgs {
+func startLogger(workers int, logStorePath *string, logStoreFormat *string, log *zap.SugaredLogger) *loggerArgs {
 	loggingMode := v1beta1.LoggerType(*logMode)
 	switch loggingMode {
 	case v1beta1.LogAll, v1beta1.LogRequest, v1beta1.LogResponse:
 	default:
-		logger.Errorf("Malformed log-mode %s", *logMode)
+		log.Errorf("Malformed log-mode %s", *logMode)
 		os.Exit(-1)
 	}
 
 	logUrlParsed, err := url.Parse(*logUrl)
 	if err != nil {
-		logger.Errorf("Malformed log-url %s", *logUrl)
+		log.Errorf("Malformed log-url %s", *logUrl)
 		os.Exit(-1)
 	}
 
@@ -285,23 +287,35 @@ func startLogger(workers int, logger *zap.SugaredLogger) *loggerArgs {
 
 	sourceUriParsed, err := url.Parse(*sourceUri)
 	if err != nil {
-		logger.Errorf("Malformed source_uri %s", *sourceUri)
+		log.Errorf("Malformed source_uri %s", *sourceUri)
 		os.Exit(-1)
 	}
 
-	var annotationKVPair map[string]string = map[string]string{}
+	annotationKVPair := map[string]string{}
 	for _, annotations := range *metadataAnnotations {
 		k, v, found := strings.Cut(annotations, "=")
 		if found {
 			annotationKVPair[k] = v
 		} else {
-			logger.Errorf("annotation does not adhere to desired format got key: %s value: %s", k, v)
+			log.Errorf("annotation does not adhere to desired format got key: %s value: %s", k, v)
 			os.Exit(-1)
 		}
 	}
 
-	logger.Info("Starting the log dispatcher")
-	kfslogger.StartDispatcher(workers, logger)
+	var store kfslogger.Store
+	if kfslogger.GetStorageStrategy(*logUrl) != kfslogger.HttpStorage {
+		if logStoreFormat != nil && *logStoreFormat != "" && logStorePath != nil && *logStorePath != "" {
+			log.Infow("Logger storage is enabled", "path", logStorePath, "logStoreFormat", logStoreFormat)
+			store, err = kfslogger.NewStoreForScheme(logUrlParsed.Scheme, *logStorePath, *logStoreFormat, log)
+			if err != nil {
+				log.Errorw("Error creating logger store", zap.Error(err))
+				os.Exit(-1)
+			}
+		}
+	}
+
+	log.Info("Starting the log dispatcher")
+	kfslogger.StartDispatcher(workers, store, log)
 	return &loggerArgs{
 		loggerType:       loggingMode,
 		logUrl:           logUrlParsed,

config/crd/full/serving.kserve.io_clusterservingruntimes.yaml

@@ -793,6 +793,8 @@ spec:
                                   - port
                                 type: object
                             type: object
+                          stopSignal:
+                            type: string
                         type: object
                       livenessProbe:
                         properties:
@@ -2772,6 +2774,8 @@ spec:
                                       - port
                                     type: object
                                 type: object
+                              stopSignal:
+                                type: string
                             type: object
                           livenessProbe:
                             properties:

config/crd/full/serving.kserve.io_clusterstoragecontainers.yaml

@@ -261,6 +261,8 @@ spec:
                             - port
                             type: object
                         type: object
+                      stopSignal:
+                        type: string
                     type: object
                   livenessProbe:
                     properties: