The rise of microservices - containers and orchestration

Editor's Notes

• #4: Microservices were pioneered in the web and then mobile App worlds; at one time called micro-web-services. Now other enterprises are looking for the same benefits. Microservice architectures implement applications as a series of small, self-contained, loosely coupled software components. Each has a specific and well understood role. Benefits of microservices: - Development Speed - Rapid Iteration Evolve quickly, continuous deployment Isolate impact of changes to existing functions or just add a new one Reactive development Maintainable Independent, empowered work teams
• #5: Monolith is like Spaghetti Changing anything impacts everything else. <= 1990s Pre-SOA (monolithic) Tight coupling For a monolith to change, all must agree on each change. Each change has unanticipated effects requiring careful testing beforehand
• #6: Service Oriented Architecture is like a Pie Easier to remove or change a piece but still need to make sure that it fits back together properly 2000s Traditional SOA Looser coupling Elements in SOA are developed more autonomously but must be coordinated with others to fit into the overall design SOA was billed as a way of delivering code re-use but people ended up designing for efficiency instead.
• #7: Microservices are like Cupcakes Can add new ones with different flavors, remove ones that you no longer need, add more pink ones if there’s greater demand Developers can create and activate new microservices without prior coordination with others. Their adherence to MSA principles makes continuous delivery of new or modified services possible Greater modularity, looser coupling. Started in the web and mobile app world, moving to Enterprise. Big in media and startups Plan for flexibility rather than reuse
• #8: Each of the ovals represents a microservice. Each source of social media feeds has its own microservice which is specialised in interfacing with the relevant API. Each of those microservices passes messages to the ‘feed merge’ microservice which can then make them available for further microservices to work with. Communication between the microservices is over the network – they can be local to the same machine or distributed. Best practice is for each microservice to be stateless and to have its own database or schema
• #9: Individual microservices can be updated in isolation or even removed if their role is no longer needed
• #10: When a new role (or even a change to an existing one) appears, best practice is to implement a new microservice rather than extending an existing one.
• #11: When a new role (or even a change to an existing one) appears, best practice is to implement a new microservice rather than extending an existing one.
• #12: Microservices allow scale-out. Each type of microservice can be scaled independently – add extra instances just for the functions that are being overworked. Multiple instances of each service can provide High Availability
• #13: An alternate view is that each microservice is much larger – in this case User Account, Product Catalog, Inventory & Order. The key is to do what makes sense for your enterprise: which pieces need scaling independently, which would you like to upgrade independently, how does it fit into your organisation? Functions are decoupled into four separate services (separate WAR files) and organized by business boundaries Web pages, Classes, Config files to a service are grouped together Services can evolve and iterate independently as each team can develop, test, and deploy code separately Best practice is for each microservice to have its own database.
• #14: Melvin Edward Conway is a computer scientist, computer programmer, and hacker who coined what's now known as Conway's Law in 1967. -Organizations design systems which copy the organization -If the parts of the organization do not closely reflect the essential parts of the product, or if the relationship between organizations do not reflect relationship between product parts, than the project will be in trouble -Make sure the organization is compatible with the product architecture
• #15: Best practice is for each microservice to be small enough that a single developer should be able to understand its entire codebase (think in hundreds rather than 10,000s of lines of code. The code for a microservice should be owned by the organization responsible for that function; for example the marketing development team should own the microservice responsible for sending nurture track emails. Netflix was one of the pioneers of Micrservices with their ”Cloud Native” approach – it was really all about being able to scale their development organisation.
• #24: To do useful work, microservices need a way of communicating – Apache Kafka Kafka provides a flexible, scalable, and reliable method to distribute streams of event data from one or more **producers** to one or more **consumers**. Examples of **events** (or **messages**) include: A periodic sensor reading such as the current temperature A user adding an item to the shopping cart in an online store A Tweet being sent with a specific hashtag A log entry generated for each click in a web application Streams of Kafka events are organized into **topics**. A producer chooses a topic to send a given event to and consumers select which topics they pull events from. For example, a financial application could pull NYSE stock trades from one topic, and company financial announcements from another in order to look for trading opportunities. Kafka actually stores all of the messages that it passes around – this makes it ideal for production microservice deployments A microservice can be upgraded and then catch up on everything it missed Or even apply its updated business logic to the full history of events A new microservice can be added and it can be brought up to speed with everything that’s gone before If one service is generating more work than another can keep up with then Kafka operates as a buffer
• #27: [Apache Aurora](http://aurora.apache.org/) – a highly scalable service scheduler for long-running services and `cron` jobs; it's used by Twitter. Aurora extends Mesos by adding rolling updates, service registration, and resource quotas. [Chronos](https://github.com/mesos/chronos) – a fault tolerant service scheduler, to be used as a replacement for `cron`, to orchestrate scheduled jobs within Mesos. [Marathon](https://mesosphere.github.io/marathon/) – a simple to use service scheduler; it builds upon Mesos and Chronos by ensuring that two Chronos instances are running.
• #30: Key Message here: Containers were originally assumed to be stateless but there’s now a demand for running DBs in a container. Kubernetes has been adding some features to help (e.g. persistent volumes) Still some gotchas for MongoDB but we have a white paper explaining exactly how to build and deploy a replica set with Docker and Kubernetes
• #31: Because we’re using external IP addresses, we can instead create the replica set with nodes in different regions
• #32: DevOps & Continuous Delivery Low impact & risk; update one container at a time Replicating environments Instantiate clones for development, QA, production, support… Scalability Add and remove containers based on demand Accurate Testing Confident your stack exactly matches what’s in production Isolation Safely run multiple environments on the same hosts Performance Minimal impact from container overhead High Availability Redundancy from multiple containers fulfilling a role Faster Time to Market Organisational alignment Cost reduction More efficient use of infrastructure
• #33: Fast is more important than elegant. Change in the application’s functionality and usage is frequent. Change occurs at different rates within the application, so functional isolation and simple integration are more important than module cohesiveness. Functionality is easily separated into simple, isolatable components. When you have the developer/DevOps skillsets. Where development org boundaries match service boundaries. Don’t forget that you’re building a distributed system -> complexity but there are precedents to read up on. One argument is that you shouldn’t bother with microservices unless you need either: - Scale your team - Design for change Sagrada Familia – designed by Gaudi; construction started on March 19, 1882. Expected to be finished in 2026.
• #34: Flexible data model Fits in with the need to be agile Monitoring & automation You have lots of moving parts in a microservice architecture – need to monitor and automate as much as possible Redundancy MongoDB replica sets -> can perform online upgrades and easily cope with rescheduling Scalability MongoDB sharding lets your database scale just as easily as your stateless microservices Remember web-scale? Best practice for each service to have its own schema or database; MongoDB’s simplicity helps
• #35: GAP moved their purchase order system from a monolith architecture to microservices. Due to MongoDB’s flexible schema, it took just 75 days to build the new system. When requirements changed and they had to add new types of purchase orders, it took days instead of months. FuboTV is a North American soccer streaming service. Using Microservices with Kubernetes, Docker & MongoDB. Isolation means that they can use a single cluster of machines (in Google Cloud) for dev, QA & production. Very birsty application – scalability lets them handle 100x increases in traffic. Otto – the key was to have an architecture that fits with their organization. Microservices empower loosely couple development teams (business, project management, IT). This is all enabling Fast test & deployment + Iterative, Continuous Delivery Backcountry.com is an online specialty retailer that sells outdoor clothing and gear. The driver to Microservices for them was a growing, distributed development team. As more and more developers joined and made contributions to the code, the schemas became convoluted and harder to maintain; contributing to 20% of the Scrum backlog. Taking advantage of MongoDB’s flexible data model, Backcountry was able to iterate faster, reduce development time, and mitigate technical debt.