Files
2026-04-03 16:31:08 +05:30

163 lines
7.6 KiB
YAML

---
apiVersion: v1
kind: ConfigMap
metadata:
name: inferenceservice-config
data:
agent: |-
{
"image" : "{{ .Values.kserve.agent.image }}:{{ .Values.kserve.agent.tag | default .Values.kserve.version }}",
"memoryRequest": "100Mi",
"memoryLimit": "1Gi",
"cpuRequest": "100m",
"cpuLimit": "1"
}
batcher: |-
{
"image" : "{{ .Values.kserve.agent.image }}:{{ .Values.kserve.agent.tag | default .Values.kserve.version }}",
"memoryRequest": "1Gi",
"memoryLimit": "1Gi",
"cpuRequest": "1",
"cpuLimit": "1",
"maxBatchSize": "32",
"maxLatency": "5000"
}
router: |-
{
"image" : "{{ .Values.kserve.router.image }}:{{ .Values.kserve.router.tag | default .Values.kserve.version }}",
"memoryRequest": "100Mi",
"memoryLimit": "1Gi",
"cpuRequest": "100m",
"cpuLimit": "1",
"imagePullPolicy": "{{ .Values.kserve.router.imagePullPolicy }}",
"imagePullSecrets": {{ .Values.kserve.router.imagePullSecrets }}
}
credentials: |-
{
"storageSpecSecretName": "{{ .Values.kserve.storage.storageSpecSecretName }}",
"storageSecretNameAnnotation": "{{ .Values.kserve.storage.storageSecretNameAnnotation }}",
"gcs": {
"gcsCredentialFileName": "gcloud-application-credentials.json"
},
"s3": {
"s3AccessKeyIDName": "{{ .Values.kserve.storage.s3.accessKeyIdName }}",
"s3SecretAccessKeyName": "{{ .Values.kserve.storage.s3.secretAccessKeyName }}",
"s3Endpoint": "{{ .Values.kserve.storage.s3.endpoint }}",
"s3UseHttps": "{{ .Values.kserve.storage.s3.useHttps }}",
"s3Region": "{{ .Values.kserve.storage.s3.region }}",
"s3VerifySSL": "{{ .Values.kserve.storage.s3.verifySSL }}",
"s3UseVirtualBucket": "{{ .Values.kserve.storage.s3.useVirtualBucket }}",
"s3UseAnonymousCredential": "{{ .Values.kserve.storage.s3.useAnonymousCredential }}",
"s3CABundleConfigMap": "{{ .Values.kserve.storage.s3.CABundleConfigMap }}",
"s3CABundle": "{{ .Values.kserve.storage.s3.CABundle }}"
}
}
deploy: |-
{
"defaultDeploymentMode": "{{ .Values.kserve.controller.deploymentMode }}"
}
service: |-
{
"serviceClusterIPNone": {{ .Values.kserve.service.serviceClusterIPNone }}
}
explainers: |-
{
"art": {
"image" : "{{ .Values.kserve.servingruntime.art.image }}",
"defaultImageVersion": "{{ .Values.kserve.servingruntime.art.defaultVersion | default .Values.kserve.version }}"
}
}
ingress: |-
{
"enableGatewayApi": {{ .Values.kserve.controller.gateway.ingressGateway.enableGatewayApi }},
"kserveIngressGateway" : "{{ .Values.kserve.controller.gateway.ingressGateway.kserveGateway }}",
"ingressGateway" : "{{ .Values.kserve.controller.gateway.ingressGateway.gateway }}",
"knativeLocalGatewayService" : "{{ .Values.kserve.controller.gateway.localGateway.knativeGatewayService }}",
"localGateway" : "{{ .Values.kserve.controller.gateway.localGateway.gateway }}",
"localGatewayService" : "{{ .Values.kserve.controller.gateway.localGateway.gatewayService }}",
"ingressClassName" : "{{ .Values.kserve.controller.gateway.ingressGateway.className }}",
"ingressDomain" : "{{ .Values.kserve.controller.gateway.domain }}",
"additionalIngressDomains": [
{{- range $index, $flag := .Values.kserve.controller.gateway.additionalIngressDomains }}
{{ $flag | quote }} {{ if ne $index (sub (len $.Values.kserve.controller.gateway.additionalIngressDomains) 1)}}, {{ end }}
{{- end}}
],
"domainTemplate": "{{ .Values.kserve.controller.gateway.domainTemplate }}",
"pathTemplate": "{{ .Values.kserve.controller.gateway.pathTemplate }}",
"urlScheme": "{{ .Values.kserve.controller.gateway.urlScheme }}",
"disableIstioVirtualHost": {{ .Values.kserve.controller.gateway.disableIstioVirtualHost }},
"disableIngressCreation": {{ .Values.kserve.controller.gateway.disableIngressCreation }},
"disableHTTPRouteTimeout": {{ .Values.kserve.controller.gateway.disableHTTPRouteTimeout }}
}
logger: |-
{
"image" : "{{ .Values.kserve.agent.image }}:{{ .Values.kserve.agent.tag | default .Values.kserve.version }}",
"memoryRequest": "100Mi",
"memoryLimit": "1Gi",
"cpuRequest": "100m",
"cpuLimit": "1",
"defaultUrl": "http://default-broker"
}
storageInitializer: |-
{
"image" : "{{ .Values.kserve.storage.image }}:{{ .Values.kserve.storage.tag | default .Values.kserve.version }}",
"memoryRequest": "100Mi",
"memoryLimit": "1Gi",
"cpuRequest": "100m",
"cpuLimit": "1",
"caBundleConfigMapName": "{{ .Values.kserve.storage.caBundleConfigMapName }}",
"caBundleVolumeMountPath": "{{ .Values.kserve.storage.caBundleVolumeMountPath }}",
"enableModelcar": {{ .Values.kserve.storage.enableModelcar }},
"uidModelcar": {{ .Values.kserve.storage.uidModelcar }},
"cpuModelcar": "{{ .Values.kserve.storage.cpuModelcar }}",
"memoryModelcar": "{{ .Values.kserve.storage.memoryModelcar }}"
}
metricsAggregator: |-
{
"enableMetricAggregation": "{{ .Values.kserve.metricsaggregator.enableMetricAggregation }}",
"enablePrometheusScraping" : "{{ .Values.kserve.metricsaggregator.enablePrometheusScraping }}"
}
localModel: |-
{
"enabled": {{ .Values.kserve.localmodel.enabled }},
"jobNamespace": "{{ .Values.kserve.localmodel.jobNamespace }}",
"jobTTLSecondsAfterFinished": {{ .Values.kserve.localmodel.jobTTLSecondsAfterFinished }},
"defaultJobImage": "{{ .Values.kserve.localmodel.defaultJobImage }}:{{ .Values.kserve.localmodel.defaultJobTag | default .Values.kserve.version }}",
"fsGroup": {{ .Values.kserve.localmodel.securityContext.fsGroup }},
"reconcilationFrequencyInSecs": {{ .Values.kserve.localmodel.agent.reconcilationFrequencyInSecs }},
"disableVolumeManagement": {{ .Values.kserve.localmodel.disableVolumeManagement }}
}
security: |-
{
"autoMountServiceAccountToken": {{ .Values.kserve.security.autoMountServiceAccountToken }}
}
inferenceService: |-
{
"resource": {
"cpuLimit": "{{ .Values.kserve.inferenceservice.resources.limits.cpu }}",
"cpuRequest": "{{ .Values.kserve.inferenceservice.resources.requests.cpu }}",
"memoryLimit": "{{ .Values.kserve.inferenceservice.resources.limits.memory }}",
"memoryRequest": "{{ .Values.kserve.inferenceservice.resources.requests.memory }}"
}
}
opentelemetryCollector: |-
{
"scrapeInterval": "{{ .Values.kserve.opentelemetryCollector.scrapeInterval }}",
"metricReceiverEndpoint": "{{ .Values.kserve.opentelemetryCollector.metricReceiverEndpoint }}",
"metricScalerEndpoint": "{{ .Values.kserve.opentelemetryCollector.metricScalerEndpoint }}",
"resource": {
"cpuLimit": "{{ .Values.kserve.opentelemetryCollector.resource.cpuLimit }}",
"memoryLimit": "{{ .Values.kserve.opentelemetryCollector.resource.memoryLimit }}",
"cpuRequest": "{{ .Values.kserve.opentelemetryCollector.resource.cpuRequest }}",
"memoryRequest": "{{ .Values.kserve.opentelemetryCollector.resource.memoryRequest }}"
}
}
autoscaler: |-
{
"scaleUpStabilizationWindowSeconds": "{{ .Values.kserve.autoscaler.scaleUpStabilizationWindowSeconds }}",
"scaleDownStabilizationWindowSeconds": "{{ .Values.kserve.autoscaler.scaleDownStabilizationWindowSeconds }}"
}