Delivering Agile Data Science on Openshift - Red Hat Summit 2019
3 likes693 views
The document discusses delivering agile data science solutions on OpenShift, highlighting the challenges faced by data science teams and how OpenShift can facilitate rapid deployment and feedback loops. It details the use of Jupyter notebooks for creating reproducible data science environments and emphasizes the importance of a modern CI/CD pipeline in data science operations. The document also covers the implementation timeline of data science capabilities at ExxonMobil and outlines essential features for data science developers.
Delivering Agile Data Science on Openshift - Red Hat Summit 2019
- 1. Delivering Agile Data Science on Openshift Audrey Reznik Data Scientist May 9th, 2019 John Archer Principal Energy Solution Architect How to create Instant Business Value
- 2. Red Hat Summit May 2019 – Delivering agile data science solutions with OpenShift MEET THE SPEAKERS John Archer Principal Solution Energy Architect Red Hat since 2015 BEA Systems, BSI Consulting, DocuQuest, Andrews & Kurth, SilverStream, Petris and Oracle Upstream Data Management, DoD, APIs, eCommerce, IoT, data science and blockchain SPE, SEG, PPDM, HJUG, HDUG, HAL- PC, Energistics Audrey Reznik Data Scientist Upstream Research Center ExxonMobil since 2007 Chevron, Akamai, Entriq, Digital Medical Registrar, Spider Technologies, Ziff Davis
- 3. Red Hat Summit May 2019 – Delivering agile data science solutions with OpenShift DATA SCIENCE TEAM PRESSURES EXPLOSIVE GROWTH in data analytics teams and analytic tools MULTIPLE TEAMS COMPETING for use of the same storage and computing resources CONGESTION in busy analytic clusters causing frustration and missed SLAs EMERGING DATAOPS Data Scientist Developers vs Full Stack Developer agility and enablement gaps
- 4. What can you envision and share?
- 5. Red Hat Summit May 2019 – Delivering agile data science solutions with OpenShift NEED: SHARE CODE (PRODUCT) WITH USERS Jupyter Notebooks as a technology we could use to combine python code, a GUI, documentation for sharing with customers. Start of a Interactive Data Science environment. Red Hat OpenShift PoC at ExxonMobil. Could this new technology benefit us in creating a Reproducible & Interactive Data Science environment? Prize: This would enable the team to not only quickly obtain customer feedback, but also easily utilize Agile Methodology; therefore, quickly delivering MVPs. Drawback: how does one avoid the setup/configuration issues and reliably deploy the notebook? Pip install required Anaconda libraries Jupyter Notebook Python 3.x (load onto PC – or setup server) Local admin access Access to latest source code OS?SQL Server PC Setup
- 6. Red Hat Summit May 2019 – Delivering agile data science solutions with OpenShift LOCAL PC VS OPENSHIFT PROJECT CONTAINERS Jupyter Notebook Python 3.x (image) Libraries • Numpy • Pandas • Matplotlib • IPyWidgets • SciPy • Lmfit • Seaborne • Plotly SQLite Container v2.0 GIT Image project Code project OpenShift URL to PoCCode Local PC Setup pip install required Anaconda libraries Jupyter Notebook Python 3.x (load onto PC – or setup server) Local admin access Access to latest source code OS?SQL Server Reproducible Data Science environment that users interact with via Chrome. Hardware Freedom & easier Reproduction!
- 7. Red Hat Summit May 2019 – Delivering agile data science solutions with OpenShift For a Data Scientist, the ability to rapidly deploy code and quickly obtain feedback from a user is extremely valuable and Agile! Openshift facilitates these capabilities! REPRODUCIBLE & INTERACTIVE SCIENTIFIC ENVIRONMENT 1. Understand the Problem 2. Suggest Solutions Deliver POC 3. Refine the Problem Agile How to Deploy? URL to PoC Code GIT Image project Code project OpenShift “Interactive” feedback! Nexus Image As a user I want to provide frequent feedback! Python (Pypi) Security
- 8. Red Hat Summit May 2019 – Delivering agile data science solutions with OpenShift DEPLOY SOURCE CODE WITH SOURCE TO IMAGE (S2I) • Re-useable Data Science Applications: data location • To re-useable Data Science Images: can they be re-consumed or modified for particular use cases? • E.g. we have a base python image that has been modified to provide TensorFlow, SciKit Learn for Data Science projects. • Reusable data access containers: SQL Server, Oracle, PI, SAP HANA. Git RepositoryBUILD APP (OpenShift) Developer code Source-to-Image (S2I) Builder Image Image Registry BUILD IMAGE (OpenShift) DEPLOY (OpenShift) deployApplication Container
- 9. Red Hat Summit May 2019 – Delivering agile data science solutions with OpenShift MATURING THE CI/CD PIPELINE Seeing an emerging notion of Data ScienceOps workflows. Current OS production CI/CD in progress. Challenges we are experiencing include: 1. OnPrem databases in different countries 2. Development/Deployment in Jupyter notebooks GIT Jenkins build Package Jenkins Archive Artifacts in Nexus Nexus OS build image deploy to TEST OS build image deploy to PROD Test build Package
- 10. Red Hat Summit May 2019 – Delivering agile data science solutions with OpenShift MACHINE LEARNING ON OPENSHIFT Figure 1. liquid estimates. Marco De Mattia Unique performance computing requirements for Artificial Intelligence, Machine Learning, Neural Networks and GPUs Multiple Data Science images: • TensorFlow • PyTorch • Scikit-learn Testing GPU (NVidia v100) cluster (OCP). Additional service to internal HPC. Next Steps: examine RAPIDS.AI – execute end-to- end data science pipelines in the GPU…
- 11. Red Hat Summit May 2019 – Delivering agile data science solutions with OpenShift OPENSHIFT GPU PROOF OF CONCEPT (POC) GPU POC: read & analyze petro-physical data. Use ML Algorithms to generate analysis/models on GPU cluster. Vetted models can be pushed to Azure for deployment. GPUDB Data Scientist URL to ML App User ML Algorithms (GIT Repo) L4 Network onPrem Database(s) Containers Figure 2. GPU POC workflow, Audrey Reznik
- 12. Red Hat Summit May 2019 – Delivering agile data science solutions with OpenShift READY FOR ANY CLOUD – PRIVATE AND PUBLIC DATA GRAVITY DRIVES THE LOCATION • OpenShift for on premise and Public Cloud (Azure) for Container as a Service (CaaS) 1. CaaS Security enabled through AD groups created onPremise and DevOps practices 2. Self-service for accessing Data Science packages with network, routing and DNS services 3. Storage can be self-service with PVC or extended with Ceph and OCP Storage options Where does your application live? How do you access it? Is my application secure? Enabled Data Science Teams • Perform More Experiments • Spend less time on plumbing • Focus on Delivering Value to ExxonMobil Resulting In
- 13. Red Hat Summit May 2019 – Delivering agile data science solutions with OpenShift EXXONMOBIL DATA SCIENCE OS TIMELINE Started with Data Virtualization for Calgary Optimization Dec 2017 Containerized JBoss Data Virtualization on Openshift on premise - Feb 2018 Spoke with Data Science teams - Python, MATLAB, Julia and R users – Mar 2018 Introduced Graham Dumpleton’s JupyterHub container image – April - 2018 Delivered Data Science Workshop on Openshift to eight different data science teams – Dec 2018 Built “Base” Data Science image. Python 3.x, AI libraries July - 2018 Data Science developers deliver faster and collaborate globally within 2 months – Feb 2019 Successfully deploy ODH supporting multiple notebook kernels and GPU – Mar 2019 Built test OCP 3.10 cluster for NVidia v100 testing for Tensorflow and Keras - Nov 2018
- 14. Red Hat Summit May 2019 – Delivering agile data science solutions with OpenShift MOVING FORWARD: EXXONMOBIL DATA SCIENCE CAPABILITY TODAY As a Data Scientist (all I care about) is that using Openshift, I can now deploy a common Jupyter Notebook / Anaconda image (with all required libraries) in a matter of seconds. Freeing myself (and other Data Scientists) to perform data science and not worry about architecture and delivery mechanisms. Now that is Democratizing Data Science! Selected Openshift on premises and public cloud for Container as a Service (CaaS) • Openshift supports: • One Click Notebooks and JupyterHub/Lab templates • Self-service for accessing data & data science packages • Nexus Repository to allow for Python, Java, R, PHP, .Net package managers • Docker public repository security built-in process – protects against rooted containers and new CVE attacks • NVidia GPU support allows for sharing these resources across multiple teams Jupyter Notebook & select conda libraries image being used for Kearl Mining Optimization Studies
- 15. Red Hat Summit May 2019 – Delivering agile data science solutions with OpenShift DATA SCIENTIST DEVELOPERS NEEDS All Developers need ● Choice of architectures ● Choice of programming languages ● Choice of databases and persistence ● Choice of application services ● Choice of development tools ● Choice of build and deploy workflows Data Science Additional Needs ● Access to GPUs and varied storage ● Access to Curated Data ● Automated ScienceOps pipelines ● Collaboration with the Business ● Access to specific data science languages and toolsets They don’t want to have to worry about the infrastructure.
- 16. Red Hat Summit May 2019 – Delivering agile data science solutions with OpenShift YOUR DIFFERENTIATION DEPENDS ON YOUR ABILITY TO DELIVER INTELLIGENT APPS FASTER CONTAINERS, KUBERNETES, DEVOPS & DATAOPS ARE KEY INGREDIENTS Innovation Culture Cloud-native Applications AI & Machine Learning Internet of Things Virtual GPU
- 17. Red Hat Summit May 2019 – Delivering agile data science solutions with OpenShift
- 18. Red Hat Summit May 2019 – Delivering agile data science solutions with OpenShift OPENDATAHUB.IO ARCHITECTURE CONTAINER STORAGE (CEPH) CONTAINER HOST (RHEL/RHCOS) Microsoft Azure AWSOpenStackDatacenterLaptop Google Cloud CONTAINER ORCHESTRATION AND MANAGEMENT (OPENSHIFT) S3 API Object Store BLOCK FILE GPU FPGA APPLICATION LIFE CYCLE MANAGEMENT (OPENSHIFT) DEVOPS WORKFLOW (CODE & DATA) API GATEWAY (3SCALE) SERVICE MESH (ISTIO) SERVERLESS PRIVATE MICRO SERVICES (CONTAINERIZED CUSTOM APPS) CONTAINER APPS PRE-DEFINED AI LIBRARY (BOTS | ANOMALY | CLASSIFICATION | SENTIMENT | …) AI TOOLCHAIN & WORKFLOW (JUPYTER, SUPERSET, …) COMMON SERVICES SERVICECATALOG&SELFSERVICEUI/CLI IDENTITY/POLICY(ACCESS,PLACEMENT)/LINEAGE(CODE ANDDATA) MANAGEMENTCONSOLE/INSIGHTS/AIOPS (PROMETHEUS|ELASTIC|…) FEDERATION RH Core Platform OpenShift ALM Red Hat Middleware Community & ISV Ecosystem Technology Roadmap Customer Content LEGEND PYTHON / FLASK JAVA JAVASCRIPT ... STREAMING (KAFKA - streamzi) MSG BUS (AMQ) ANALYTICS (SPARK) ML (TENSORFLOW | …) MEMORY CACHE (JDG) || DECISION (BxMS) HDFS | REDIS | SQL | NoSQL | GRAPHDB | TIMESERIES | ELASTIC | ...
- 19. Red Hat Summit May 2019 – Delivering agile data science solutions with OpenShift MODERN DATA ANALYTICS PIPELINE DATA GENERATION INGEST DATA SCIENCE MACHINE LEARNING STREAM PROCESSING TRANSFORM, MERGE, JOIN DATA ANALYTICS • IoT Telemetry • G&G - Well Logs • Transactions • Production • NiFi • Kafka • MQTT • Presto • Impala • SparkSQL • Notebooks • TensorFlow • PyTorch • Keras • scikit-learn • AutoML* • Kafka • MQTT • WebSockets • Hadoop • Spark • Pandas • Apache Arrow • Spark • Hadoop
- 20. CONNECTING THE EDGE TO DATA SCIENTISTS Highly Scalable, flexible, elastic, microservice based architecture Fully Portable – On Premise to any public cloud vendor Leverages the power and agility of open source software without lock-in Architecture Tenets Data Scientist Data Manager s Citizen Data Scientist Cognitive AI Vision Speech Face Audio Video Text Data Models Curation Prep Quality Publishing SecurityPython, R, Jupyter.org, Tensorflow, Keras, Pandas, Bokeh, Dash, Prometheus, Grafana, SciPy, NumPy, SumPy, Julia , Spark, PySpark, Theano, Scikit, FaceDetect Packages: AI/ML/Data Science Pods MongoDB, MariaDB, mySQL, Postgres, Couchbase, Redis, MS-SQL, OraclePersistence : SSO and Authentication OIDC SAML OAuth JWT Kerberos DevOps Node.js, .Net Core, Java, Python, PHP, Ruby, Rails, Javascript, PerlApp Dev: AppDev & App Services and Persistence Pods REST ODBC JDBC WS Predictive Maintenance Autonomous Operations Supply Chain Improvements Downstream Reliability Use Cases Multitenant – CPU and GPU powered workloads REST IoT “Things” MQTT Integration, BPM, Rules, Messaging, API, IoT, Microservices, IstioApp Services: OnPremise Public Cloud WSS Kafka
- 21. Red Hat Summit May 2019 – Delivering agile data science solutions with OpenShift ● JupyterHub on Openshift ○ Jupyter notebook, JupyterHub, JupyterLab, Openshift Templates ● Kubeflow ○ Kube project for Tensorflow, Seldon, JupyterHub/Lab, PyTorch, MPI Operator ● Opendatahub.io ○ Ceph, Spark, JupyterHub/Lab, Tensorflow ○ Simplified Multiple Kernels support ○ GPU Support ○ Resource management and instance culling ● radanalytics.io ○ Openshift Spark ○ Oshinko - Apache Spark Cluster ○ Spark Operator OSS DATA SCIENCE PROJECTS
- 22. Red Hat Summit May 2019 – Delivering agile data science solutions with OpenShift ● Join Openshift Commons - ML SIG - https://commons.openshift.org/ ● Openshift Self Service Education - https://learn.openshift.com ● Install Minishift - https://docs.okd.io/latest/minishift/getting- started/installing.html ○ MacOS - brew cask install minishift ○ Manual - https://github.com/minishift/minishift/releases ● Install Jupyter and JupyterHub Openshift templates ○ https://github.com/jupyter-on-openshift/jupyterhub-quickstart ● Review the OpenDataHub.io project HOW CAN I GET STARTED?
- 23. Delivering Agile Data Science solutions with OpenShift … and providing Business Value!