End to end Machine Learning using Kubeflow - Build, Train, Deploy and Manage
4 likes2,131 views
This document discusses Kubeflow, an end-to-end machine learning platform for Kubernetes. It covers various Kubeflow components like Jupyter notebooks, distributed training operators, hyperparameter tuning with Katib, model serving with KFServing, and orchestrating the full ML lifecycle with Kubeflow Pipelines. It also talks about IBM's contributions to Kubeflow and shows how Watson AI Pipelines can productize Kubeflow Pipelines using Tekton.
![Define Pipeline with Python SDK
@dsl.pipeline(name='Taxi Cab Classification Pipeline Example’)
def taxi_cab_classification(
output_dir,
project,
Train_data = 'gs://bucket/train.csv',
Evaluation_data = 'gs://bucket/eval.csv',
Target = 'tips',
Learning_rate = 0.1, hidden_layer_size = '100,50’, steps=3000):
tfdv = TfdvOp(train_data, evaluation_data, project, output_dir)
preprocess = PreprocessOp(train_data, evaluation_data, tfdv.output[“schema”], project, output_dir)
training = DnnTrainerOp(preprocess.output, tfdv.schema, learning_rate, hidden_layer_size, steps,
target, output_dir)
tfma = TfmaOp(training.output, evaluation_data, tfdv.schema, project, output_dir)
deploy = TfServingDeployerOp(training.output)
Compile and Submit Pipeline Run
dsl.compile(taxi_cab_classification, 'tfx.tar.gz')
run = client.run_pipeline(
'tfx_run', 'tfx.tar.gz', params={'output': ‘gs://dpa22’, 'project': ‘my-project-33’})](https://image.slidesharecdn.com/kubeflow-e2e-dojo-external-200707165703/85/End-to-end-Machine-Learning-using-Kubeflow-Build-Train-Deploy-and-Manage-32-320.jpg)
End to end Machine Learning using Kubeflow - Build, Train, Deploy and Manage
- 1. Jupyter Notebooks Workflow Building Pipelines Tools Serving Metadata Kale Fairing TFX KF Pipelines HP Tuning Tensorboard KFServing Seldon Core TFServing, + Training Operators Pytorch XGBoost, + Tensorflow Prometheus Kubeflow: End to End ML Platform Animesh Singh MPI MXNet
- 2. © 2019 IBM Corporation Animesh Singh STSM and Chief Architect - Data and AI Open Source Platform o CTO, IBM RedHat Data and AI Open Source Alignment o IBM Kubeflow Engagement Lead, Kubeflow Committer o Chair, Linux Foundation AI - Trusted AI o Chair, CD Foundation MLOps Sig o Ambassador, CNCF o Member of IBM Academy of Technology (IBM AoT) Kubeflow github.com/kubeflow Your Speaker Today: CODAIT 2
- 3. Prepared and Analyzed Data Trained Model Deployed Model Prepared Data Untrained Model Kubeflow: Current IBM Contributors Christian Kadner Weiqiang Zhuang Tommy Li Andrew Butler Jin Chi He Feng Li Ke Zhu Kevin Yu
- 4. IBM is the 2nd Largest Contributor
- 5. IBM is the 2nd Largest Contributor
- 6. IBMers contributing across projects in Kubeflow
- 7. Kubeflow Services High Level Services Low Level APIs / Services Katib Pipelines Notebooks TFJob PyTorchJob Jupyter CR Seldon CR Kubebench Pipelines CR Argo Study Job MPIJob Spark Job KFServing TFX Developed By Kubeflow Developed Outside Kubeflow Adapted from Kubeflow Contributor Summit 2019 talk: Kubeflow and ML Landscape (Not all components are shown) Kubernetes API Server Istio Mesh and Gateway kubectl apply -f tfjob
- 10. Prepared and Analyzed Data Trained Model Deployed Model Prepared Data Untrained Model ML Lifecycle: Build: Development, Training and HPO
- 11. Develop (Kubeflow Jupyter Notebooks) – Data Scientist – Self-service Jupyter Notebooks provide faster model experimentation – Simplified configuration of CPU/GPU, RAM, Persistent Volumes – Faster model creation with training operators, TFX, magics, workflow automation (Kale, Fairing) – Simplify access to external data sources (using stored secrets) – Easier protection, faster restoration & sharing of “complete” notebooks – IT Operator – Profile Controller, Istio, Dex enable secure RBAC to notebooks, data & resources – Smaller base container images for notebooks, fewer crashes, faster to recover
- 12. Develop (Kubeflow Jupyter Notebooks) 12
- 15. Distributed Tensorflow Operator • A distributed Tensorflow Job is collection of the following processes o Chief – The chief is responsible for orchestrating training and performing tasks like checkpointing the model o Ps – The ps are parameters servers; the servers provide a distributed data store for the model parameters to access o Worker – The workers do the actual work of training the model. In some cases, worker 0 might also act as the chief o Evaluator - The evaluators can be used to compute evaluation metrics as the model is trained
- 16. Distributed MPI Operator - AllReduce • AllReduce is an operation that reduces many arrays spread across multiple processes into a single array which can be returned to all the processes • This ensures consistency between distributed processes while allowing all of them to take on different workloads • The operation used to reduce the multiple arrays back into a single array can vary and that is what makes the different options for AllReduce
- 17. Hyper Parameter Optimization and Neural Architecture Search - Katib • Katib: Kubernetes Native System for Automated tuning of machine learning model’s Hyperparameter Turning and Neural Architecture Search. • Github Repository: https://github.com/kubeflow/katib • Hyperparameter Tuning q Random Search q Tree of Parzen Estimators (TPE) q Grid Search q Hyperband q Bayesian Optimization q CMA Evolution Strategy • Neural Architecture Search q Efficient Neural Architecture Search (ENAS) q Differentiable Architecture Search (DARTS)
- 19. ❑ Rollouts: Is this rollout safe? How do I roll back? Can I test a change without swapping traffic? ❑ Protocol Standards: How do I make a prediction? GRPC? HTTP? Kafka? ❑ Cost: Is the model over or under scaled? Are resources being used efficiently? ❑ Monitoring: Are the endpoints healthy? What is the performance profile and request trace? Prepared and Analyzed Data Trained Model Deployed Model Prepared Data Untrained Model ❑ Frameworks: How do I serve on Tensorflow? XGBoost? Scikit Learn? Pytorch? Custom Code? ❑ Features: How do I explain the predictions? What about detecting outliers and skew? Bias detection? Adversarial Detection? ❑ How do I wire up custom pre and post processing ML Lifecycle: Production Model Serving ❑ How do I handle batch predictions? ❑ How do I leverage standardized Data Plane protocol so that I can move my model across MLServing platforms?
- 20. ● Seldon Core was pioneering Graph Inferencing. ● IBM and Bloomberg were exploring serverless ML lambdas. IBM gave a talk on the ML Serving with Knative at last KubeCon in Seattle ● Google had built a common Tensorflow HTTP API for models. ● Microsoft Kubernetizing their Azure ML Stack Experts fragmented across industry
- 21. ● Kubeflow created the conditions for collaboration. ● A promise of open code and open community. ● Shared responsibilities and expertise across multiple companies. ● Diverse requirements from different customer segments Putting the pieces together
- 22. ● Founded by Google, Seldon, IBM, Bloomberg and Microsoft ● Part of the Kubeflow project ● Focus on 80% use cases - single model rollout and update ● Kfserving 1.0 goals: ○ Serverless ML Inference ○ Canary rollouts ○ Model Explanations ○ Optional Pre/Post processing Model Serving - KFServing
- 23. Manages the hosting aspects of your models • InferenceService - manages the lifecycle of models • Configuration - manages history of model deployments. Two configurations for default and canary. • Revision - A snapshot of your model version • Route - Endpoint and network traffic management Route Default Configuration Revision 1 Revision M 90 % KFService Canary Configuration Revision 1 Revision N 10 % KFServing: Default and Canary Configurations
- 24. Model Servers - TensorFlow - Nvidia TRTIS - PyTorch - XGBoost - SKLearn - ONNX Components: • - Predictor, Explainer, Transformer (pre-processor, post-processor) Storage - AWS/S3 - GCS - Azure Blob - PVC Supported Frameworks, Components and Storage Subsystems
- 25. GPU Autoscaling - KNative solution Ingress Activator (buffers requests) Autoscaler Queue Proxy Model server when scale == 0 or handling burst capacity when scale > 0 metrics ● Scale based on # in-flight requests against expected concurrency ● Simple solution for heterogeneous ML inference autoscaling scale metrics 0...N Replicas API Requests
- 26. But the Data Scientist Sees... ● A pointer to a Serialized Model File ● 9 lines of YAML ● A live model at an HTTP endpoint = http ● Scale to Zero ● GPU Autoscaling ● Safe Rollouts ● Optimized Serving Containers ● Network Policy and Auth ● HTTP APIs (gRPC soon) ● Tracing ● Metrics apiVersion: "serving.kubeflow.org/v1alpha2" kind: "InferenceService" metadata: name: "flowers-sample" spec: default: predictor: tensorflow: storageUri: "gs://kfserving-samples/models/tensorflow/flowers" Production users include: Bloomberg
- 27. ` 27 KFServing: Default, Canary and Autoscaler
- 28. KFServing – Existing Features q Crowd sourced capabilities – Contributions by AWS, Bloomberg, Google, Seldon, IBM, NVidia and others. q Support for multiple runtimes pre-integrated (TFServing, Nvdia Triton (GPU optimization), ONNX Runtime, SKLearn, PyTorch, XGBoost, Custom models. q Serverless ML Inference and Autoscaling: Scale to zero (with no incoming traffic) and Request queue based autoscaling q Canary and Pinned rollouts: Control traffic percentage and direction, pinned rollouts q Pluggable pre-processor/post-processor via Transformer: Gives capabilities to plug in pre-processing/post-processing implementation, control routing and placement (e.g. pre-processor on CPU, predictor on GPU) q Pluggable analysis algorithms: Explainability, Drift Detection, Anomaly Detection, Adversarial Detection (contributed by Seldon) enabled by Payload Logging (built using CloudEvents standardized eventing protocol) q Batch Predictions: Batch prediction support for ML frameworks (TensorFlow, PyTorch, ...) q Integration with existing monitoring stack around Knative/Istio ecosystem: Kiali (Service placements, traffic and graphs), Jaeger (request tracing), Grafana/Prometheus plug-ins for Knative) q Multiple clients: kubectl, Python SDK, Kubeflow Pipelines SDK q Standardized Data Plane V2 protocol for prediction/explainability et all: Already implemented by Nvidia Triton
- 29. q MMS: Multi-Model-Serving for serving multiple models per custom KFService instance q More Data Plane v2 API Compliant Servers: SKLearn, XGBoost, PyTorch… q Multi-Model-Graphs and Pipelines: Support chaining multiple models together in a Pipelines q PyTorch support via AWS TorchServe q gRPC Support for all Model Servers q Support for multi-armed-bandits q Integration with IBM AIX360 for Explainability, AIF360 for Bias detection and ART for Adversarial detection KFServing – Upcoming Features
- 30. Prepared and Analyzed Data Trained Model Deployed Model Prepared Data Untrained Model ML Lifecycle: Orchestrate Build, Train, Validate and Deploy
- 31. Kubeflow Pipelines § Containerized implementations of ML Tasks § Pre-built components: Just provide params or code snippets (e.g. training code) § Create your own components from code or libraries § Use any runtime, framework, data types § Attach k8s objects - volumes, secrets § Specification of the sequence of steps § Specified via Python DSL § Inferred from data dependencies on input/output § Input Parameters § A “Run” = Pipeline invoked w/ specific parameters § Can be cloned with different parameters § Schedules § Invoke a single run or create a recurring scheduled pipeline
- 32. Define Pipeline with Python SDK @dsl.pipeline(name='Taxi Cab Classification Pipeline Example’) def taxi_cab_classification( output_dir, project, Train_data = 'gs://bucket/train.csv', Evaluation_data = 'gs://bucket/eval.csv', Target = 'tips', Learning_rate = 0.1, hidden_layer_size = '100,50’, steps=3000): tfdv = TfdvOp(train_data, evaluation_data, project, output_dir) preprocess = PreprocessOp(train_data, evaluation_data, tfdv.output[“schema”], project, output_dir) training = DnnTrainerOp(preprocess.output, tfdv.schema, learning_rate, hidden_layer_size, steps, target, output_dir) tfma = TfmaOp(training.output, evaluation_data, tfdv.schema, project, output_dir) deploy = TfServingDeployerOp(training.output) Compile and Submit Pipeline Run dsl.compile(taxi_cab_classification, 'tfx.tar.gz') run = client.run_pipeline( 'tfx_run', 'tfx.tar.gz', params={'output': ‘gs://dpa22’, 'project': ‘my-project-33’})
- 33. Visualize the state of various components
- 38. Kubeflow Pipelines can train, deploy and serve Open Source Dojo 38
- 39. Kubernetes Ready ML and AI Platform Operator Hub - operatorhub.io
- 40. Prepared and Analyzed Data Trained Model Deployed Model Prepared Data Untrained Model Watson Productization of Kubeflow Pipelines
- 41. Watson AI Pipelines • Demonstrate that Watson can be used for end-end AI lifecycledata prep/model training/model risk validation/model deployment/monitoring/updating models • Demonstrate that the full lifecycle can be operated programmatically, and have Tekton as a backend instead of Argo
- 42. Pipeline: Train the model and monitor with OpenScale
- 43. Tekton q A PipelineResource defines an object that is an input (such as a git repository) or an output (such as a docker image) of the pipeline. q A PipelineRun defines an execution of a pipeline. It references the Pipeline to run and the PipelineResources to use as inputs and outputs. q A Pipeline defines the set of Tasks that compose a pipeline. q A Task defines a set of build Steps such as compiling code, running tests, and building and deploying images. TASK STEP POD STEP TASK STEP STEP POD Container Container Container Container TEKTON q The Tekton Pipelines project provides Kubernetes-style resources for declaring CI/CD- style pipelines. q Tekton introduces several new CRDs including Task, Pipeline, TaskRun, and PipelineRun. q A PipelineRun represents a single running instance of a Pipeline and is responsible for creating a Pod for each of its Tasks and as many containers within each Pod as it has Steps.
- 44. KFP API Server Components Pipelines Object Store KFP UI Relational DB Argo Pipeline Yaml Tekton Pipeline Yaml KFP – Tekton Phase One Pluggable Components Watson Studio WML Open Scale Spark Kubeflow Training Seldon AIF360 ART KATIB KFSERVING ! ! ! ! ! ! ! … …! COMPILE KFP SDK TASK STEP POD STEP STEP POD POD POD STEP TASK STEP STEP STEP POD Container Container Container Container ARGO TEKTON
- 45. KFP – Tekton Phase Two Pluggable Components Watson Studio WML Open Scale Spark Kubeflow Training Seldon AIF360 ART KATIB KFSERVING ! ! ! ! ! ! ! … …! TASK STEP POD STEP STEP POD POD POD STEP TASK STEP STEP STEP POD Container Container Container Container ARGO TEKTON KFP API Server Components Pipelines Object Store KFP UI Relational DB Argo Pipeline Yaml Tekton Pipeline Yaml COMPILE KFP SDK
- 46. KFP – Tekton Challenges 46 Multiple Moving parts, with different stakeholders Tekton Community: Argo with version 2.6 much more mature than Tekton v0.11 (alpha) when the work started around 5 months ago • Multiple features and capabilities lacking in Tekton when we kick started • The team had to default to a spreadsheet to start tracking and mapping KFP DSL features, and areas where Tekton needed to bring features and functions. Overall 50 DSL capabilities identified and corresponding Tekton features started getting mapped. • Multiple features like Kubernetes resources support to create/patch/update/delete them, image pull secrets, loops, conditionals, support for system params didn’t exist. Or existed partially • Tekton started moving from alpha to beta as the work progressed, and few features left behind in alpha mode • Multiple issues opened on Tekton. Required ramping up the team of Tekton contributors to help drive these issues . Formed a virtual team of IBM Open tech developers (Andrea Frittoli, Priti Desai), IBM Systems team (Vincent Pli) DevOps team (Simon Kaegi), RedHat (Vincent Demeester etc.) to drive Tekton requirements Kubeflow Pipeline and TFX Community: Open source team needed to be formed for the specific mission. And trained. Additionally Google needed to be brought up on the same page, and convinced the validity of integration. • Multiple design reviews established with Google, and jointly agreed on a direction after they were convinced why we were doing it, and why it makes sense. • Convincing to accelerate the IR (Intermediate Representation) strategy with TFX, so as to be able to drive this the right way • Huge dependency in Kubeflow Pipeline code on Argo, including the API backend and UI all written with Argo dependency • Internal IBM team divided to attack different areas: Compiler (Christian Kadner), API (Tommy Li), UI (Andrew), Feng Li (IBM Systems, China) • Inability of Kubeflow Pipeline backend to take multiple CRDs, which is the default model Tekton follows. So everything needed to be bundled in one Pipeline Spec • Type check, workflow utils, and parameter replacement are heavily tied with Argo API. In addition, the persistent agent is watching the resources using the Argo API type. • MLOps Sig in CD Foundation leveraged to bring Kubeflow Pipelines and Tekton team together
- 47. KFP – Tekton: Delivered Pluggable Components Watson Studio WML Open Scale Spark Kubeflow Training Seldon AIF360 ART KATIB KFSERVING ! ! ! ! ! ! ! … …! TASK STEP POD STEP TASK STEP STEP POD Container Container Container Container TEKTON KFP API Server Components Pipelines Object Store KFP UI Relational DB Tekton Pipeline Yaml COMPILE KFP SDK
- 48. Same KFP Experience: DAG, backed by Tekton YAML 48
- 49. Same KFP Exp: Logs, Lineage Tracking and Artifact Tracking 49
- 50. 50 End to end Kubeflow Components : With KFP-Tekton
- 51. Prepared and Analyzed Data Trained Model Deployed Model Prepared Data Untrained Model Kubeflow Adoption: External and Internal
- 52. Telstra AI Lab - (TAIL) - Configuration • Kubernetes – 1.15 • Spectrum Scale CSI Driver • MetalLB for Load Balancing • Istio 1.3.1 for ingress • Kubeflow – 1.0.1 • Jupyter Notebook images are IBM’s multiarchitecture powerai images ( https://hub.docker.com/r/ibmcom/powerai/tags) Telstra: Collaborating with IBM to build an Open Source based OneAnalytics Platform leveraging Kubeflow THINK 2020 Session: End-to-End Data Science and Machine Learning for Telcos: Telstra's Use Case https://www.ibm.com/events/think/watch/replay/126561688
- 53. Telstra AI Lab - (TAIL) – Future state • RedHat Openshift – 4.3 • GPU Operator • Kubeflow Operator • Extending the compute • Integrate feature stores and streaming technologies • Integrate with CI/CD tools (Tekton Pipelines)
- 54. Yara – Working with IBM to build a Data Science Platform for Digital Farming ML use cases based on Kubeflow 54 THINK 2020 Session: Enable Smart Farming using Kubeflow https://www.ibm.com/events/think/watch/replay/126494864
- 55. Watson STT: Kubeflow Pipelines running Operations
- 56. Watson SpeechToText training Kubeflow pipeline
- 57. Prepared and Analyzed Data Trained Model Deployed Model Prepared Data Untrained Model OpenDataHub
- 58. 'Upstream' is about extracting oil and natural gas from the ground; 'midstream' is about safely moving them thousands of miles; and 'downstream' is converting these resources into the fuels and finished products we all depend on. Upstream, Midstream and Downstream
- 59. Upstream, Midstream and Downstream 'Upstream' is about extracting oil and natural gas from the ground; 'midstream' is about safely moving them thousands of miles; and 'downstream' is converting these resources into the fuels and finished products we all depend on.
- 60. Data Platform Operator Hub - operatorhub.io OpenShift Ready
- 61. OPEN DATA HUB - Ecosystem 61
- 62. Red Hat OpenShift Container Platform OPEN DATA HUB REFERENCE ARCHITECTURE Storage Metadata Management Data Analysis AI and ML Security and Governance Monitoring and Orchestratio n Data in Motion Data Lake In Memory Relational Databases Streaming Data Object Storage Data Log Data Big Data Processing Streaming Data Exploration Interactive Notebooks Model Lifecycle ML Applications Business Applications Metastore
- 63. Red Hat OpenShift Container Platform OPEN DATA HUB REFERENCE IMPLEMENTATION Storage Metadata Management Data Analysis AI and ML Security and Governance OpenShift Oauth OpenShift Single SignOn (Keycloak) RedHat Ceph Object Gateway RedHat 3scale Monitoring and Orchestratio n Prometheus Grafana Kubeflow Pipelines Jenkins CI/CD Data in Motion Data Lake RedHat Ceph Storage In Memory RedHat Data Grid (Infinispan) Relational Databases PostgreSQL MySQL Streaming Data RedHat AMQ Streams Kafka Connect Object Storage Data RedHat Ceph S3 API Log Data FluentD Logstash Big Data Processing Spark SparkSQL Thrift Streaming Kafka Streams Elastic Search Data Exploration Hue Kibana Interactive Notebooks JupyterHub Hue Model Lifecycle Kubeflow Seldon MLFlow ML Applications OpenDataHub AI Library Business Applications Superset Metastore Hive
- 65. Initial Goals: OpenDataHub and Kubeflow Initial Goals: • Kubeflow has a great traction, Make it available for OpenShift users Done in https://github.com/opendatahub-io/manifests • Offer ODH users components installed by KF • And offer components from ODH (Kafka, Apache SuperSet, Hive…) to KF community • Decide if we can leverage KF project and community as upstream for ODH • Think Kubernetes -> OpenShift • Frees up ODH maintainers time to make sure KF keeps running well on OpenShift
- 66. Kubeflow Operator – Contributed by IBM to Kubeflow community to help enable OpenDataHub • https://operatorhub.io/operator/kubeflow • Deploy, manage and monitor Kubeflow • On various environments q IBM Cloud q GCP q AWS q Azure q OpenShift q Other K8S
- 67. Outcome: Kubeflow an Upstream for OpenDataHub ● A version of the Operator based on Kubeflow Architecture released: https://developers.redhat.com/blog/2020/05/07/open- data-hub-0-6-brings-component-updates-and-kubeflow- architecture/?sc_cid=7013a000002DTqEAAW ● Most of the components converted: https://github.com/opendatahub-io/odh-manifests ● Still a separate deployment – needs to do both ODH and Kubeflow in one go. Future • KF 1.0 on OpenShift • Disconnected deployment • Open Data Hub CI/CD • Kubeflow on OpenShift CI • UBI based ODH & KF • Multitenancy model • Mixing KF & ODH
- 68. OPEN DATA HUB 0.6.x
- 69. Open Data Hub in OpenShift 69
- 70. Apache Superset 70 Think 2020 / DOC ID / Month XX, 2020 / © 2020 IBM Corporation
- 71. Spark with Open Data Hub 71 • Open Data Hub will also deploy the Spark Operator to manage Spark as an application. • Two versions of Spark – Spark in dedicated mode and Spark on K8s • Currently moving towards Spark on K8s Operator from Google for serverless Spark. IBM Hummingbird team investigating this
- 72. Airflow integration with Open Data Hub 72 • Open Data Hub will also deploy the Airflow Operator to manage Airflow as an application. • Using the Airflow Operator originally developed in the GoogleCloudPlatform repository and later donated to Apache. • The Operator creates a controller-manager pod which will be created as a part of the Open Data Hub deployment. • Users can then install the Airflow components they need from the available options (eg: CeleryExecutor or KubernetesExecutor, Postgres deployment or MySQL deployment etc. )
- 73. Apache Hive with OpenDataHub • Hive was one of the first abstraction engines to be built on top of MapReduce. • Started at Facebook to enable data analysts to analyse data in Hadoop by using familiar SQL syntax without having to learn how to write MapReduce. • Hive an essential tool in the Hadoop ecosystem that provides an SQL dialect for querying data stored in HDFS, other file systems that integrate with Hadoop such as MapR-FS and Amazon’s S3 and databases like HBase(the Hadoop database) and Cassandra. • Hive is a Hadoop based system for querying and analysing large volumes of structured data which is stored on HDFS. • Hive is a query engine built to work on top of Hadoop that can compile queries into MapReduce jobs and run them on the cluster.
- 74. Data Platform Operator Hub - operatorhub.io OpenShift Ready
- 75. Kubernetes Ready ML and AI Platform Operator Hub - operatorhub.io
- 76. Kubernetes Ready Upstream Kubeflow Midstream OpenDataHub OpenShift Ready Operator Hub - operatorhub.io Kubeflow OpenDataHub Open Source End To End Data and AI Platform RedHat MarketPlace https://marketplace.redhat.com/en-us
- 77. Coming Next: Kubeflow Dojo https://github.com/kubeflow https://github.com/opendatahub-io https://github.com/IBM/ KubeflowDojo
- 78. Kubeflow Dojo: Prerequisites • Knowledge of Kubernetes, watch the dojo for Kubernetes project with the IBM internal link or external link • Access to a Kubernetes cluster, either minikube or remote hosted • Source code control and development with git and github, watch the presentation with the IBM internal link or external link for git and external link for pull requests • Get familiar with golang language, watch the introduction dojo with the IBM internal link or external link • (optional) Knowledge of Istio and knative • If you have more time, o Read Kubeflow document to learn more about Kubeflow project o Browse through Kubeflow community github
- 79. Kubeflow Dojo: Tips for success • Access to a Kubernetes cluster • minimal spec: 8vcpu, 16gb ram and at least 50gb disk for docker registry • On IBM Kubernetes Service, provision the cluster with machine type b2c.4x16 and 2 worker nodes • Follow Kubeflow document to have your cluster prepared • On IKS cluster, follow this link to install the IBM Cloud CLI and helm followed by setting up IBM Cloud Block Storage as the default storage class