OpenShift AI
Red Hat® OpenShift® AI とは、柔軟でスケーラブルな人工知能 (AI) および機械学習 (ML) プラットフォームです。このプラットフォームにより、企業はハイブリッドクラウド環境全体で AI 対応アプリケーションを大規模に作成および提供できるようになります。
OpenShift AI はオープンソース・テクノロジーを使用して構築されており、実験、モデル提供、革新的なアプリケーションの提供のための、信頼性と一貫性に優れた運用機能を提供します。
Data Science Project
ここでは、OpenShift AI 2.9.1 / Red Hat OpenShift on IBM Cloud (ROKS) 4.14 の環境で、以下の Tutorial を用いて Data Science Project の実装を確認します。
OpenShift AI UI
Data Science Project は、OpenShift AI UI の Data Science Projects から作成します。Tutorial に従って一連のリソースを作成すると、以下のようになります。ここから Edit project や Delete project なども可能です。
Data Science Project
作成した Data Science Project を開くと幾つかの Tab があり、それれぞれの内容は以下のとおりです。
Overview
Workbenches
Pipelines
Models
Cluster storage
Data connections
Permissions
fraud-detection Project
Tutorial の手順に従うと、Data Science Project のリソースは fraud-detection Project に配置されます。
$ oc get all,pvc -n fraud-detection -o name
Warning: apps.openshift.io/v1 DeploymentConfig is deprecated in v4.14+, unavailable in v4.10000+
pod/create-ds-connections-j62fq
pod/create-minio-buckets-sqstp
pod/create-minio-root-user-gxcpj
pod/create-s3-storage-hjqr2
pod/ds-pipeline-dspa-7ffd8ff549-dhtnx
pod/ds-pipeline-metadata-envoy-dspa-9c745bb5d-bphwb
pod/ds-pipeline-metadata-grpc-dspa-57d47dd549-r5zds
pod/ds-pipeline-persistenceagent-dspa-6b96b9c8d8-sgwkj
pod/ds-pipeline-scheduledworkflow-dspa-64c767c896-lv8d2
pod/ds-pipeline-workflow-controller-dspa-5b4cdb559-bzg6g
pod/fraud-detection-0
pod/mariadb-dspa-8685d54888-8qhhj
pod/minio-6bff775986-hh2kz
pod/modelmesh-serving-model-server-778b47cc76-487x2
service/ds-pipeline-dspa
service/ds-pipeline-metadata-envoy-dspa
service/ds-pipeline-metadata-grpc-dspa
service/ds-pipeline-workflow-controller-metrics-dspa
service/fraud-detection
service/fraud-detection-tls
service/mariadb-dspa
service/metadata-grpc-service
service/minio
service/ml-pipeline
service/modelmesh-serving
deployment.apps/ds-pipeline-dspa
deployment.apps/ds-pipeline-metadata-envoy-dspa
deployment.apps/ds-pipeline-metadata-grpc-dspa
deployment.apps/ds-pipeline-persistenceagent-dspa
deployment.apps/ds-pipeline-scheduledworkflow-dspa
deployment.apps/ds-pipeline-workflow-controller-dspa
deployment.apps/mariadb-dspa
deployment.apps/minio
deployment.apps/modelmesh-serving-model-server
replicaset.apps/ds-pipeline-dspa-7ffd8ff549
replicaset.apps/ds-pipeline-metadata-envoy-dspa-9c745bb5d
replicaset.apps/ds-pipeline-metadata-grpc-dspa-57d47dd549
replicaset.apps/ds-pipeline-persistenceagent-dspa-6b96b9c8d8
replicaset.apps/ds-pipeline-scheduledworkflow-dspa-64c767c896
replicaset.apps/ds-pipeline-workflow-controller-dspa-5b4cdb559
replicaset.apps/mariadb-dspa-8685d54888
replicaset.apps/minio-6bff775986
replicaset.apps/modelmesh-serving-model-server-778b47cc76
statefulset.apps/fraud-detection
job.batch/create-ds-connections
job.batch/create-minio-buckets
job.batch/create-minio-root-user
job.batch/create-s3-storage
route.route.openshift.io/ds-pipeline-dspa
route.route.openshift.io/ds-pipeline-metadata-envoy-dspa
route.route.openshift.io/fraud-detection
route.route.openshift.io/minio-console
route.route.openshift.io/minio-s3
persistentvolumeclaim/fraud-detection
persistentvolumeclaim/mariadb-dspa
persistentvolumeclaim/minio
詳細を確認すると、以下の CR から実装されたことが分かります。
$ oc get svc,deploy,sts,job,route,pvc -n fraud-detection -o custom-columns-file=RES_OWNER.txt
NAMESPACE KIND NAME NODE OWNER_KIND OWNER_NAME OWNER_APIV
fraud-detection Service ds-pipeline-dspa <none> DataSciencePipelinesApplication dspa datasciencepipelinesapplications.opendatahub.io/v1alpha1
fraud-detection Service ds-pipeline-metadata-envoy-dspa <none> DataSciencePipelinesApplication dspa datasciencepipelinesapplications.opendatahub.io/v1alpha1
fraud-detection Service ds-pipeline-metadata-grpc-dspa <none> DataSciencePipelinesApplication dspa datasciencepipelinesapplications.opendatahub.io/v1alpha1
fraud-detection Service ds-pipeline-workflow-controller-metrics-dspa <none> DataSciencePipelinesApplication dspa datasciencepipelinesapplications.opendatahub.io/v1alpha1
fraud-detection Service fraud-detection <none> Notebook fraud-detection kubeflow.org/v1beta1
fraud-detection Service fraud-detection-tls <none> Notebook fraud-detection kubeflow.org/v1
fraud-detection Service mariadb-dspa <none> DataSciencePipelinesApplication dspa datasciencepipelinesapplications.opendatahub.io/v1alpha1
fraud-detection Service metadata-grpc-service <none> DataSciencePipelinesApplication dspa datasciencepipelinesapplications.opendatahub.io/v1alpha1
fraud-detection Service minio <none> <none> <none> <none>
fraud-detection Service ml-pipeline <none> DataSciencePipelinesApplication dspa datasciencepipelinesapplications.opendatahub.io/v1alpha1
fraud-detection Service modelmesh-serving <none> Namespace fraud-detection v1
fraud-detection Deployment ds-pipeline-dspa <none> DataSciencePipelinesApplication dspa datasciencepipelinesapplications.opendatahub.io/v1alpha1
fraud-detection Deployment ds-pipeline-metadata-envoy-dspa <none> DataSciencePipelinesApplication dspa datasciencepipelinesapplications.opendatahub.io/v1alpha1
fraud-detection Deployment ds-pipeline-metadata-grpc-dspa <none> DataSciencePipelinesApplication dspa datasciencepipelinesapplications.opendatahub.io/v1alpha1
fraud-detection Deployment ds-pipeline-persistenceagent-dspa <none> DataSciencePipelinesApplication dspa datasciencepipelinesapplications.opendatahub.io/v1alpha1
fraud-detection Deployment ds-pipeline-scheduledworkflow-dspa <none> DataSciencePipelinesApplication dspa datasciencepipelinesapplications.opendatahub.io/v1alpha1
fraud-detection Deployment ds-pipeline-workflow-controller-dspa <none> DataSciencePipelinesApplication dspa datasciencepipelinesapplications.opendatahub.io/v1alpha1
fraud-detection Deployment mariadb-dspa <none> DataSciencePipelinesApplication dspa datasciencepipelinesapplications.opendatahub.io/v1alpha1
fraud-detection Deployment minio <none> <none> <none> <none>
fraud-detection Deployment modelmesh-serving-model-server <none> ServingRuntime model-server serving.kserve.io/v1alpha1
fraud-detection StatefulSet fraud-detection <none> Notebook fraud-detection kubeflow.org/v1beta1
fraud-detection Job create-ds-connections <none> <none> <none> <none>
fraud-detection Job create-minio-buckets <none> <none> <none> <none>
fraud-detection Job create-minio-root-user <none> <none> <none> <none>
fraud-detection Job create-s3-storage <none> <none> <none> <none>
fraud-detection Route ds-pipeline-dspa <none> DataSciencePipelinesApplication dspa datasciencepipelinesapplications.opendatahub.io/v1alpha1
fraud-detection Route ds-pipeline-metadata-envoy-dspa <none> DataSciencePipelinesApplication dspa datasciencepipelinesapplications.opendatahub.io/v1alpha1
fraud-detection Route fraud-detection <none> Notebook fraud-detection kubeflow.org/v1
fraud-detection Route minio-console <none> <none> <none> <none>
fraud-detection Route minio-s3 <none> <none> <none> <none>
fraud-detection PersistentVolumeClaim fraud-detection <none> <none> <none> <none>
fraud-detection PersistentVolumeClaim mariadb-dspa <none> DataSciencePipelinesApplication dspa datasciencepipelinesapplications.opendatahub.io/v1alpha1
fraud-detection PersistentVolumeClaim minio <none> <none> <none> <none>
$ oc get DataSciencePipelinesApplication,Notebook,ServingRuntime -n fraud-detection
NAME AGE
datasciencepipelinesapplication.datasciencepipelinesapplications.opendatahub.io/dspa 15d
NAME AGE
notebook.kubeflow.org/fraud-detection 15d
NAME DISABLED MODELTYPE CONTAINERS AGE
servingruntime.serving.kserve.io/model-server openvino_ir ovms 15d
DataSciencePipelinesApplication
$ oc explain DataSciencePipelinesApplication.spec
GROUP: datasciencepipelinesapplications.opendatahub.io
KIND: DataSciencePipelinesApplication
VERSION: v1alpha1
FIELD: spec <Object>
DESCRIPTION:
<empty>
FIELDS:
apiServer <Object>
DS Pipelines API Server configuration.
database <Object>
Database specifies database configurations, used for DS Pipelines metadata
tracking. Specify either the default MariaDB deployment, or configure your
own External SQL DB.
dspVersion <string>
<no description>
mlmd <Object>
<no description>
mlpipelineUI <Object>
Deploy the KFP UI with DS Pipelines UI. This feature is unsupported, and
primarily used for exploration, testing, and development purposes.
objectStorage <Object> -required-
ObjectStorage specifies Object Store configurations, used for DS Pipelines
artifact passing and storage. Specify either the your own External Storage
(e.g. AWS S3), or use the default Minio deployment (unsupported, primarily
for development, and testing) .
persistenceAgent <Object>
DS Pipelines PersistenceAgent configuration.
scheduledWorkflow <Object>
DS Pipelines Scheduled Workflow configuration.
workflowController <Object>
WorkflowController is an argo-specific component that manages a DSPA's
Workflow objects and handles the orchestration of them with the central Argo
server
Notebook
$ oc explain notebook.spec.template.spec
GROUP: kubeflow.org
KIND: Notebook
VERSION: v1
FIELD: spec <Object>
DESCRIPTION:
<empty>
FIELDS:
activeDeadlineSeconds <integer>
<no description>
affinity <Object>
<no description>
automountServiceAccountToken <boolean>
<no description>
containers <[]Object> -required-
<no description>
dnsConfig <Object>
<no description>
dnsPolicy <string>
<no description>
enableServiceLinks <boolean>
<no description>
ephemeralContainers <[]Object>
<no description>
hostAliases <[]Object>
<no description>
hostIPC <boolean>
<no description>
hostNetwork <boolean>
<no description>
hostPID <boolean>
<no description>
hostname <string>
<no description>
imagePullSecrets <[]Object>
<no description>
initContainers <[]Object>
<no description>
nodeName <string>
<no description>
nodeSelector <map[string]string>
<no description>
os <Object>
<no description>
overhead <map[string]Object>
<no description>
preemptionPolicy <string>
<no description>
priority <integer>
<no description>
priorityClassName <string>
<no description>
readinessGates <[]Object>
<no description>
restartPolicy <string>
<no description>
runtimeClassName <string>
<no description>
schedulerName <string>
<no description>
securityContext <Object>
<no description>
serviceAccount <string>
<no description>
serviceAccountName <string>
<no description>
setHostnameAsFQDN <boolean>
<no description>
shareProcessNamespace <boolean>
<no description>
subdomain <string>
<no description>
terminationGracePeriodSeconds <integer>
<no description>
tolerations <[]Object>
<no description>
topologySpreadConstraints <[]Object>
<no description>
volumes <[]Object>
<no description>
ServingRuntime
$ oc explain servingruntime.spec
GROUP: serving.kserve.io
KIND: ServingRuntime
VERSION: v1alpha1
FIELD: spec <Object>
DESCRIPTION:
<empty>
FIELDS:
affinity <Object>
<no description>
annotations <map[string]string>
<no description>
builtInAdapter <Object>
<no description>
containers <[]Object> -required-
<no description>
disabled <boolean>
<no description>
grpcDataEndpoint <string>
<no description>
grpcEndpoint <string>
<no description>
httpDataEndpoint <string>
<no description>
imagePullSecrets <[]Object>
<no description>
labels <map[string]string>
<no description>
multiModel <boolean>
<no description>
nodeSelector <map[string]string>
<no description>
protocolVersions <[]string>
<no description>
replicas <integer>
<no description>
storageHelper <Object>
<no description>
supportedModelFormats <[]Object>
<no description>
tolerations <[]Object>
<no description>
volumes <[]Object>
<no description>
Tutorial に従うと PVC は3つ作成され、それぞれ以下の Pod に割り当てられます。
$ oc get pvc -n fraud-detection
NAME STATUS VOLUME CAPACITY ACCESS MODES STORAGECLASS AGE
fraud-detection Bound pvc-32db862a-b2bb-45c7-8218-6462b18a2472 20Gi RWO ibmc-vpc-block-10iops-tier 15d
mariadb-dspa Bound pvc-a1be0870-cf1e-49fe-a564-b0c6d6e23fd0 10Gi RWO ibmc-vpc-block-10iops-tier 15d
minio Bound pvc-1ec0b7a6-9d38-4d0c-b76b-2fdd5599ba9b 10Gi RWO ibmc-vpc-block-10iops-tier 15d
$ for P in $(oc get pvc -n fraud-detection -o custom-columns=:.metadata.name --no-headers); do echo "PVC: $P -> POD(s) [ $(oc get pod -o json | jq --arg P $P -rc '.items[] | select(.spec.volumes[].persistentVolumeClaim.claimName == $P) | .metadata.name') ]"; done
PVC: fraud-detection -> POD(s) [ fraud-detection-0 ]
PVC: mariadb-dspa -> POD(s) [ mariadb-dspa-8685d54888-8qhhj ]
PVC: minio -> POD(s) [ minio-6bff775986-hh2kz ]