Hosting Microservices in Microsoft Azure
- 1. 1CONFIDENTIAL HOSTING MICROSERVICES IN MICROSOFT AZURE Sergii Kryshtop, Software Engineering Manager, EPAM Ukraine APRIL 21, 2018
- 2. 2CONFIDENTIAL ABOUT THE SPEAKER Sergii has 12+ years of experience in software development (6+ years with EPAM). He works with Microsoft Technology Stack and Azure cloud platform being focused on software architecture and design, cloud services, software engineering practices, delivery management and resource management. During his carrier, Sergii has worked with both green field and legacy systems, with both small teams working in one room and big ones involving 30 people from three distant geo locations. Sergii is passionate about technology, since it enables engineers to make people’s life easier and better. He is also fond of management, since it helps development teams transform the intangible ideas and lines of code into tangible products, which altogether brings real value for business and people. SERGII KRYSHTOP EPAM Systems, Software Engineering Manager
- 4. 4CONFIDENTIAL MONOLITHS AND MICROSERVICES source: https://martinfowler.com/articles/microservices.html
- 5. 5CONFIDENTIAL • Single application as a suite of small services (100+) • Services are independently deployable and scalable • Different services to be written in different programming languages • Different services can be managed by different teams • Design for failure, requires monitoring and logging MONOLITHS VS. MICROSERVICES • Change cycles are tied together • Hard to keep a good modular structure • Scaling requires scaling of the entire application MICROSERVICE ARCHITECTURE MONOLITHIC ARCHITECTURE
- 6. 6CONFIDENTIAL • Networking, Service Discovery • Configuration management • Deployment w/o downtime • Scaling • Monitoring COMPONENTS
- 9. 9CONFIDENTIAL • App/OS Customization: ability to select OS and support of different programming languages and runtimes • Deployment: Built-in tools for application deployment, upgrade and integration with CI/CD tools • Scalability: built-in capabilities for scaling-out application on specific condition • Monitoring: built-in features for monitoring • Cost Efficiency: hosting service cost per single application. The higher efficiency, the lower cost • Security: ability to secure access to application on different OSI model levels. Higher mark indicates ability to control access on lower levels (network, transport, application) CRITERIA
- 10. 10CONFIDENTIAL • OS: Linux, Windows, Custom Image • Deployment: No built-in support, Infrastructure only • Scalability: autoscale with Azure Insights Autoscale • Infrastructure monitoring (CPU, Memory, App metrics) • SLA: – 1 x VM: 99.9% – 2 x VM in Availability Zones: 99.99% • Cost for 50 microservices: – 2 x 50 * A1 v2 ($31.39) = $3,139 / month VIRTUAL MACHINE / VM SCALE SET App/OS Customization: Deployment: Scalability: Monitoring: Cost Efficiency: Security:
- 11. 11CONFIDENTIAL • The most popular service for running Web and API applications (including microservices) • DevOps friendly • Scale out within several minutes • Supported languages: .NET, Java, Node.js, PHP, and Python • Monitoring and Integration with Application Insight • Docker image support is in preview APP SERVICE / APP SERVICE PLAN PROS App Service Plan VM1 Web App 1 IIS Site Web Site Web App 2 IIS Site Web Job VM2 Web App 1 IIS Site Web Site Web App 2 IIS Site Web Job AUTOSCALING LOAD BALANCING
- 12. 12CONFIDENTIAL • No real isolation for each application. Applications runs in IIS Application Pool Process • Scaling is focused on scaling VMs, but not the applications • No control on per-application restrictions on consuming resources • Internal load balancer only in Premium service App Service Environment APP SERVICE / APP SERVICE PLAN CONS
- 13. 13CONFIDENTIAL • OS: Windows and Linux • Platforms: .NET Fx, .NET Core , Java, Node.js, PHP, Python • Deployment: – Integrated CD with major Git Repositories, – Blue/Green and Canary support with Slots • Scalability: autoscale (5 min window) • Infrastructure monitoring (CPU, Memory, App metrics) • Automatic OS update management • SLA: 1 x App Service Plan: 99.95% • Cost for 50 microservices: – 1 x 50 * S1 ($73.00) = $3,650 / month APP SERVICE / WEB APPS App/OS Customization: Deployment: Scalability: Monitoring: Cost Efficiency: Security:
- 14. 14CONFIDENTIAL • OS: Linux • Deployment: – CI/CD with Docker Hub, Azure Container Registry, and GitHub – Blue/Green and Canary support with Slots • Scalability: autoscale (min 5 min window) • Infrastructure monitoring (CPU, Memory, App metrics) • SLA: 1 x App Service Plan: 99.95% • Cost for 50 microservices: – 1 x 50 * S1 ($73.00) = $3,650 / month APP SERVICE / WEB APPS FOR CONTAINERS App/OS Customization: Deployment: Scalability: Monitoring: Cost Efficiency: Security:
- 15. 15CONFIDENTIAL • Dedicated App Service environment • Predicted performance on RPS • Deployed into Virtual Network • All features of Azure App Service • Up to 100 App Service Plans • Cost for 50 microservices: – Flat fee for each ASE: $981.19/month – 50 * I1 (255.50) = $12,775 / month APP SERVICE ENVIRONMENT (ASE) App/OS Customization: Deployment: Scalability: Monitoring: Cost Efficiency: Security:
- 16. 16CONFIDENTIAL • Serverless architecture • Reach deployment options from App Service Plan • Predicted cost per transaction or use App Service Plan • Languages: C#, JavaScript, F# • Triggers: schedule, Azure Service events (Blob, Queue, etc.), webhook • Cost for 50 microservices: – Execution Time: $0.000016/GB-s – Total Executions: $0.20 per mln executions 3 mln executions, 1 second, 512MB ~ $20 /month AZURE FUNCTIONS App/OS Customization: Deployment: Scalability: Monitoring: Cost Efficiency: Security:
- 18. 18CONFIDENTIAL • Fully managed platform and programming model for microservices • Proven platform used by Azure and other Microsoft services • Run in Azure, on-premise or other clouds • Introduce additional Actor programming model • Java, Docker and Windows containers are in roadmap. Linux support Is in Preview AZURE SERVICE FABRIC (ASF) PROS CONS • Vendor lock on Platform and model • On-premise support were released in 2016 • Need to develop expertise
- 19. 19CONFIDENTIAL • OS: Windows, Linux • Deployment: Rolling updates, etc. • Scalability: built-in autoscale • Infrastructure monitoring, App monitoring • Programming Models: from containers and guest- executables to microservices and actors. • SLA: 99.99% (from VM) • Cost for 50 microservices: – 5 x D2 V3 ($80.30) = $401.5 / month – 5 x D4 V3 ($160.60) = $803.0 / month AZURE SERVICE FABRIC App/OS Customization: Deployment: Scalability: Monitoring: Cost Efficiency: Security:
- 21. 21CONFIDENTIAL CONTAINERS ARE LIGHTWEIGHT SERVER HOST OS GUEST OS HYPERVISOR (TYPE 2) LIBRARIES APP A GUEST OS LIBRARIES APP B VM SERVER HOST OS LIBRARIES APP A LIBRARIES APP B CONTAINERS • Containers are isolated but share OS • Increase the compute density and memory utilization DOCKER ENGINE
- 22. 22CONFIDENTIAL VIRTUAL MACHINES CONTAINERS AND THE HOST
- 23. 23CONFIDENTIAL • Application Deployment, including rollouts and rollbacks • Horizontal Scaling • Self-healing • Discovery and load balancing • Configuration management • Monitoring CONTAINER ORCHESTRATION CONTAINER ORCHESTRATION = CLOUD PLATFORM
- 24. 24CONFIDENTIAL • Optimized provisioning of open-source container orchestration tools – Kubernetes – DC/OS – Docker Swarm • Full automation of all orchestration platform components • Compatible with standard open-source orchestration API • ACS is a free service that clusters Virtual Machines (VMs) into a container • Hybrid deployment for applications AZURE CONTAINER SERVICE (ACS)
- 25. 25CONFIDENTIAL CONTAINER ORCHESTRATION BASELINE ARCHITECTURE Master Master App 1 App 2 App 3 App 1 App 2 Worker Node Worker Node OrchestrationOrchestration Orchestration Services Admin UI / CLI Load Balancing Application Users
- 27. 27CONFIDENTIAL ACS DOCKER SWARM DEPLOYMENT
- 28. 28CONFIDENTIAL ACS DC/OS (APACHE MESOS)
- 29. 30CONFIDENTIAL • Optimized provisioning of open-source container orchestration tools • Full automation of all orchestration platform components • Compatible with standard open-source orchestration API • ACS is a free service that clusters Virtual Machines (VMs) into a container • Hybrid deployment for applications AZURE KUBERNETES SERVICE (AKS)
- 30. 31CONFIDENTIAL • Azure-Hosted Control Plane • Control version of Kubernetes • Upgrade to new version of Kubernetes • No payment for Master Nodes AZURE CONTAINER SERVICE (MANAGED) (AKS) UNMANAGED KUBERNETES (ACS) MANAGED KUBERNETES (AKS) Control Plane Master VM Master VM Master VM Agent Pool Agent VM Agent VM Agent VM Hosted Control Pane Agent Pool Agent VM Agent VM Agent VM
- 31. 32CONFIDENTIAL • OS: Linux • Deployment: Rolling updates, etc. • Scalability: in-cluster scaling • Infrastructure monitoring, App monitoring (K8s tools) • Programming Models: Any that can run in Linux containers • SLA: 99.99% (from VM when 2+ workers) • Cost for 50 microservices: – Master: 3 x D2 V3 ($80.30) ~ $241.0 / month – Worker: 3 x D4 V3 ($160.60) ~ $483.0 / month AZURE KUBERNETES SERVICE (AKS) App/OS Customization: Deployment: Scalability: Monitoring: Cost Efficiency: Security:
- 32. 33CONFIDENTIAL • Store and manage container images across all types of Azure deployments • Azure Container Registry is a managed Docker registry service based on the open-source Docker Registry 2.0. • Maintain Windows and Linux container images in a single Docker registry • Use familiar, open-source Docker command line interface (CLI) tools • Simplify registry access management with Azure Active Directory • Managed Registry and Classic AZURE CONTAINER REGISTRY SERVICE
- 33. 34CONFIDENTIAL ACR PRICING Managed Registry SLA 99.9% successfully process Registry Transactions. The SLA for Classic Registry is provided through Azure Storage (99.9%).
- 35. 37CONFIDENTIAL • Get more DTU density for given cost • Databases runs in isolation • Scale up/down database manually or using autoscale SQL DATABASE ELASTIC POOL SQL ELASTIC DATABASE POOL eDTU POOL
- 36. 38CONFIDENTIAL DEMO: DEPLOY TO ACS KUBERNETES
- 37. 39CONFIDENTIAL 1. Provision Managed Kubernetes Cluster 2. Provision Azure Container Service 3. Get sample application from GitHub and create docker package 4. Push app image to registry 5. Deploy application to cluster SCENARIO https://github.com/sergekryshtop/gab2018_aksdemo
- 39. 43CONFIDENTIAL CREATE DOCKER REGISTRY SECRET WITH JMESPATH http://jmespath.org/
- 40. 44CONFIDENTIAL • Available Docker Integration Build Task • Docker Registry Credentials Management • Docker Build command • Docker Push command • Docker Run command • Run Docker commands • Docker Compose support • Requires VSTS agent with Docker tools or Agent running in Docker Container VISUAL STUDIO TEAM SERVICES: BUILD
- 41. 45CONFIDENTIAL BUILD STEPS AND RESULT
- 42. 46CONFIDENTIAL BUILD STEPS FOR AGENT IN DOCKER
- 44. 48CONFIDENTIAL • Depending on product engineering maturity and system architecture you have several Azure services to deployment • Containers and Linux are the first-class citizens in Azure • Microsoft wrap all Container orchestration components with easy-to-use services which allows to build full CI and CD solution in short time • It is a trend from Microsoft and industry to move into better resource utilization and containers • If you build microservices on Microsoft stack and not interested in Containers and open-source orchestration, than Azure Service Fabric might be a good option to go WRAP UP