The Stack That Helped Medium Scale To 2.6 Millennia Of Reading Time

26,417
Evfnneus
Medium
Medium is a different kind of place on the internet. A place where the measure of success isn’t views, but viewpoints. Where the quality of the idea matters, not the author’s qualifications. A place where conversation pushes ideas forward and words still matter.

Editor's note: By Daniel Pupius, ‎Head of Engineering at Medium


medium

Background

Medium is a network. It’s a place to share stories and ideas that matter — it’s where you move thinking forward, and people have spent 1.4 billion minutes — or 2.6 millennia — reading on Medium.

We get over 25 million unique readers every month and tens of thousands of posts published each week. But we want Medium to be a place where the measure of success isn’t views, but viewpoints. Where the quality of the idea matters, not the author’s qualifications. A place where conversation pushes ideas forward and words still matter.

I lead the engineering team. I was previously a Staff Software Engineer at Google, where I worked on Google+ and Gmail, and co-founded the Closure project. In past lives I’ve raced snowboards, jumped out of planes, and lived in the jungle.

Engineering Team

I couldn’t be prouder of this team. It’s an awesome bunch of talented, curious, mindful individuals who come together to do great work.

We operate in cross-functional, mission-driven teams, so while some people specialize, everyone should feel able to touch any part of the stack. We believe that exposure to different disciplines makes you a stronger engineer. I wrote about our other values here.

The teams have a lot of flexibility in how they organize around their work, but as a company we set quarterly goals and encourage iterative sprints. We use GitHub for code reviews and bug tracking and Google Apps for email, docs, and spreadsheets. We’re heavy users of Slack — and slack bots — and many teams use Trello.

Initial Stack

We deployed to EC2 from the start. The main app servers were written in Node.js, and we migrated to DynamoDB for the public launch.

There was a node server that we used for image processing, delegating to GraphicsMagick for the actual hard work. And another server acted as a SQS queue processor for background tasks.

We used SES for email, S3 for static assets, CloudFront as CDN, and nginx as a reverse proxy. We used Datadog for monitoring and PagerDuty for alerting.

The site used TinyMCE as a foundation for the editor. Before launch we were already using the Closure Compiler and some portions of the Closure Library, but Handlebars for templates.

Current Stack

For a site as seemingly simple as Medium, it may be surprising how much complexity is behind the scenes. It’s just a blog, right? You could probably knock something out using Rails in a couple of days. :)

Anyway, enough snark. Let’s start at the bottom.

Production Environment

We are on Amazon’s Virtual Private Cloud. We use Ansible for system management, which allows us to keep our configuration under source control and easily roll out changes in a controlled way.

We have a service-oriented architecture, running about a dozen production services (depending on how you count them and some more micro than others). The primary choice as to whether to deploy a separate service is the specificity of the work it performs, how likely dependent changes are to be made across service boundaries, and the resource utilization characteristics.

Our main app servers are still written in Node, which allows us to share code between server and client, something we use quite heavily with the editor and post transformations. Node has worked pretty well for us, but performance problems have emerged where we block the event loop. To alleviate this, we run multiple instances per machine and route expensive endpoints to specific instances, thus isolating them. We’ve also hooked into the V8 runtime to get insights into what ticks are taking a long time; generally it’s due to object reification during JSON deserialization.

We have several auxiliary services written in Go. We’ve found Go very easy to build, package, and deploy. We like the type-safety without the verbosity and JVM tuning of Java. Personally, I’m a fan of using opinionated languages in a team environment; it improves consistency, reduces ambiguity, and ultimately gives you less rope to hang yourself. 

We now serve static assets using CloudFlare, though we send 5% of traffic to Fastly and 5% to CloudFront to keep their caches warm should we need to cut over in an emergency. Recently we turned up CloudFlare for application traffic as well — primarily for DDOS protection but we’ve been happy with the performance gains.

We use a combination of Nginx and HAProxy as reverse proxies and load balancers, to satisfy the Venn Diagram of features we need.

We still use Datadog for monitoring and PagerDuty for alerts, but we now heavily use ELK (Elasticsearch, Logstash, Kibana) for debugging production issues.

Databases

DynamoDB is still our primary datastore, but it hasn’t been completely smooth sailing. One of the perennial issues we’ve hit is the hotkey issue during viral events or fanouts for million-follower users. We have a Redis cache cluster sitting in front of Dynamo, which mitigates these issues with reads. Optimizing for developer convenience and production stability have often seemed at odds, but we’re working to close the gap.

We’re starting to use Amazon Aurora for some newer data, which allows more flexible querying and filtering than Dynamo.

We use Neo4J to store relations between the entities that represent the Medium network, running a master with two replicas. People, posts, tags, and collections are nodes in the graphs. Edges are created on entity creation and when people perform actions such as follow, recommend, and highlight. We walk the graph to filter and recommend posts.

Data Platform

From early on we’ve been very data hungry, investing in our analytics infrastructure to help us make business and product decisions. More recently we’re able to use the same data pipelines to feed back into production systems to power data-driven features such as Explore.

We use Amazon Redshift as our data warehouse, providing the scalable storage and processing system our other tools build on. We continuously import our core data set (e.g. users, posts) from Dynamo into Redshift, and event logs (e.g. post viewed, post scrolled) from S3 to Redshift.

Jobs are scheduled by Conduit, an internal tool that manages scheduling, data dependencies, and monitoring. We use an assertion-based scheduling model, where jobs will only be executed if their dependencies are satisfied (e.g. daily job that depends on an entire day of event logs). In production this has proved indispensable — data producers are decoupled from their consumers, simplifying configuration, and the system is very predictable and debuggable.

While SQL queries running in Redshift work well for us, we need to get data into and out of Redshift. We’ve increasingly turned to Apache Spark for ETL because of its flexibility and ability to scale with our growth. Over time Spark will likely become the tool of choice for our data pipelines.

We use Protocol Buffers for our schemas (and schema evolution rules) to keep all layers of the distributed system in sync, including mobile apps, web service, and data warehouse. Using custom options, we annotate our schemas with configuration details like table name and indexes, and validation constraints like max length for strings, or acceptable ranges for numbers.

People need to remain in sync too so mobile and web app developers can all log the same way, and Product Scientists can interpret fields, in the same way. We help our people work with data by treating the schemas as the spec, rigorously documenting messages and fields and publishing generated documentation from the .proto files.

Images

Our image server is now written in Go and uses a waterfall strategy for serving processed images. The servers use groupcache, which provides a memcache alternative while helping to reduce duplicated work across the fleet. The in-memory cache is backed by a persistent S3 cache; then images are processed on demand. This gives our designers the flexibility to change image presentation and optimize for different platforms without having to do large batch jobs to generate resized images.

While it’s now mainly used for resizing and cropping, earlier versions of the site allowed for color washes, blurring, and other image effects. Processing animated gifs has been a huge pain for reasons that should be another post.

TextShots

The fun TextShots feature is powered by a small Go server that interfaces with PhantomJS as a renderer process.



I always imagined switching the rendering engine to use something like Pango, but in practice, the ability to lay out the image in HTML is way more flexible and convenient. And the frequency at which the feature is used means we can handle the throughput quite easily.

Custom Domains

We allow people to set up custom domains for their Medium publications. We wanted single sign-on and HTTPS everywhere, so it wasn’t super trivial to get working. We have a set of dedicated HAProxy servers that manage certs and route traffic to the main fleet of application servers. There is some manual work required when setting up a domain, but we’ve automated large swathes of it through a custom integration with Namecheap. The cert provisioning and publication linking is handled by a dedicated service.

Web Frontend

On the web, we tend to want to stay close to the metal. We have our own Single Page Application framework that uses Closure as a standard library. We use Closure Templates for rendering on both the client and the server, and we use the Closure Compiler to minify the code and split it into modules. The editor is the most complex part of our web app, which Nick has written about.

iOS

Both our apps are native, making minimal use of web views.

On iOS, we use a mixture of homegrown frameworks and built-in components. In our network layer, we use NSURLSession for making requests and Mantle for parsing JSON into models. We have a caching layer built on top of NSKeyedArchiver. We have a generic way to render items in a list with a common styling, which allows us to quickly build new lists with different types of content. The post view is built with a UICollectionView with a custom layout. We use shared components to render the full post and the post preview.

Every commit is built and pushed to Medium employees, so that we can try out the app as quickly as possible. The cadence of our release to the appstore is beholden to the review cycle, but we try to keep pushing as fast as we can, even if there are only minimal updates.

For tests, we use XCTest and OCMock.

Android

On Android, we stay current with the very latest editions of the SDK and support libraries. We don’t use any comprehensive frameworks, preferring instead to establish consistent patterns for repeated problems. We use guava for all the things missing from Java. But otherwise, we tend to use 3rd party tools that aim to solve more narrow problems. We define our API responses using protocol buffers and then generate the objects we use in the app.

We use mockito and robolectric. We write high-level tests that spin up activities and poke around — we create basic versions of these when we first add a screen or to prepare for refactoring. They grow as we reproduce bugs to shield against regression. We write low-level tests to exercise the particulars of a single class — we create these as we build out new features and they help us reason about how our classes interact.

Every commit is automatically pushed to the play store as an alpha build, which goes out to Medium staff right away. (This includes another flavor of the app, for our internal version of Medium — Hatch). Most Fridays we promote the latest alpha to our beta group and have them play with things over the weekend. Then, on Monday, we promote it from beta to production. Since the latest code is always ready for release, when we find a bad bug, we get the fix out to production immediately. When we’re worried about a new feature, we let the beta group play with things a little longer; when we’re excited, we release even more frequently.

A|B Testing & Feature Flags

All our clients use server-supplied feature flags, called variants, for A|B testing and guarding unfinished features.

Misc

There are a lot of other things on the fringe of the product that I haven’t mentioned above: Algolia has allowed us to iterate quickly on search-related functionality, SendGrid for inbound and outbound email, Urban Airship for notifications, SQS for queue processing, Bloomd for bloom filters, PubSubHubbub and Superfeedr for RSS, etc. etc.

Build, Test, Deploy Workflow

We embrace continuous integration and delivery, pushing on green as fast as possible. Jenkins manages all those processes.

Historically we’ve used Make for our build system, but we’re migrating to Pants for newer projects.

We have a combination of unit tests and HTTP level functional tests. All commits have to pass tests before they can be merged. We worked with the team at Box to use Cluster Runner to distribute the tests and make this fast. There’s nice integration with GitHub.

We deploy to a staging environment as quickly as we can — currently about 15 minutes — and successful builds are then used as candidates for production. The main app servers normally deploy around five times a day, but sometimes as many as 10 times.

We do blue/green deploys. For production we send traffic to a canary instance, and the release process monitors error rates before proceeding with the deploy. Rollbacks are internal DNS flips.

What's Next

We’re starting to work on monetization features for authors and publishers. This is a green field project, and we’re approaching the problem space with open minds. We think the future needs new mechanisms for funding content, and we want to make sure our features incentivize quality content and value to the network.


medium


Joining Medium Engineering

Generally we’re always interested in talking with mission-driven engineers who have experience working in the consumer space. We’re not prescriptive about what languages you know, since we think that good engineers can quickly learn new disciplines, but you should be curious, aware, resolute, and empathetic. That said, experience with iOS, Android, Node, or Go would give you a head start.

We’re growing our Product Science team, so are looking for people with experience building data pipelines and large analytical systems.

And I’m looking for engineering leaders to help scale the team. They should be interested in organizational theory, be willing to get their hands dirty, and be proponents of servant leadership.

You can find out more about Medium Engineering here.

Evfnneus
Medium
Medium is a different kind of place on the internet. A place where the measure of success isn’t views, but viewpoints. Where the quality of the idea matters, not the author’s qualifications. A place where conversation pushes ideas forward and words still matter.
Senior Engineering Manager
New York
At Medium, we’re building a platform that gives people a place to share their stories and ideas. To do that, we create simple and beautiful product experiences that prioritize the user experience. We built the best editor on the web using the cutting-edge technologies available at the time, and we’re now creating a stack for the future using modern tooling like React, Jest, and GraphQL. We prioritize fast iteration without compromising quality, while maintaining an attention to detail that creates the kind of thoughtful interactions we’re known for. We’re looking for mindful, empathetic, and self-aware leader to join our team to help us move thinking forward. With a strong focus on growth and personal development, Medium is the ideal place to develop your skills in an environment that is built on mutual trust and respect.
  • You will ship features that will delight our users.
  • You will plan and support career development for the members of your team.
  • You will inspire and grow engineers through mentorship, empowerment, and by challenging them to do their best work.
  • You will partner with product management and design to help set priorities and plan the roadmap for your team.
  • You will facilitate technical design discussions, remove blockers and align stakeholders
  • You will drive programs in engineering reliability, productivity, process and quality
  • You will recruit and retain best-in-class engineering talent to build a diverse high performing team
  • You will inspire an atmosphere of continuous improvement by serving as a coach, mentor, and technical advisor.
  • You have 5+ years of leading and managing high performance engineering teams
  • You have experience running teams in an Agile or Scrum environment
  • You are flexible, dedicated, and curious
  • You believe in the craft of software engineering, but are pragmatic when necessary.
  • You have a passion for mentoring and coaching engineers
  • You have a track record of delivering features and functionality on time with high quality
  • You have excellent communication skills
  • You have experience in Javascript (Node.js, React or GraphQL)
  • You have experience with AWS(RDS, SQS, DynamoDB)
  • You have experience working in a service-oriented or microservice architecture
  • You have deep knowledge of API design principles and best-practices, caching, resiliency, RPC, REST, and HTTP
  • Comments
    Open jobs at Medium
    Senior Software Engineer
    San Francisco/New York
    Build the future Medium with us. Who are we? We are the engineering team at Medium. Engineering is deeply integrated into the product development process. We organize in cross-functional, mission-driven teams, working side-by-side with product and design. This is a team sport. We are better together; we support each other, ask for help, and collaborate. We aspire to simple, yet powerful, systems that enable unique experiences and interactions. We prioritize quick, iterative progress, without losing sight of the big picture or compromising quality. We build best-in-class editors across web and mobile. We develop data processing systems that provide product insight and enable a new currency of attention. We build the systems that give voice to ideas and stories across the globe. We take that responsibility seriously and we have a lot of fun along the way! Who are you? You’ve built, launched, and maintained mobile and web services, and have learned many lessons along the way. You are flexible, proactive, and curious. You believe in the craft of software engineering, but are pragmatic and are driven to make a positive change in the world. While you draw inspiration from prior art, you don’t accept the status quo and challenge yourself to come up with unique solutions. In the same way, you respectfully challenge your peers, and welcome being challenged. You are willing to learn, and willing to help us learn. We have positions at Medium’s HQ in San Francisco, and in New York. We have a multinational team and will consider people from around the world who are willing to relocate. We’d love to hear from you. Please send us a résumé with cover letter here. If you’re curious what it’s like to work at Medium, check out this story from behind the scenes.
    Senior Platform Engineer
    San Francisco/New York
    Senior Android Engineer
    San Francisco
    Let’s bring Medium to the world’s largest mobile platform. At Medium, we are about stories — creation, experience, and distribution — and we believe mobile is the key to our success. We’re looking for mobile developers interested in delivering unique experiences on Android to join our team. What’s Android at Medium like? We’re on the cusp of releasing an exciting new app where users can read, write and engage with Medium content, but that’s just the beginning. We’re looking to build one of the best Android apps available on the platform by leveraging material design and Android specific technologies. We have big aspirations for our Android experience and we need the brightest minds to get there. We are moving fast and trying new things. A/B testing, feature switches (we call them variants), and continuous deployment allow us to innovate and experiment. Everyone — from engineers to designers to editorial staff — is encouraged to be a part of the mobile conversation, offer suggestions, and put ideas into action. Our engineers work out of our San Francisco office. We value colocation highly and find working together in person is most effective for our team. But, we do have varying work schedules and promote work/life balance. Our goal for the platform is to create the most engaging and interactive Medium possible. We want the writer’s content to shine within an intuitive user experience. Not only should people enjoy using our app, but it should be the go-to app for anyone who likes to read and write. It’s surprising how many challenges can arise when organizing words and images on a screen. One of our mobile engineers, Elizabeth Ford, explains, “Our web user experience has set a very high bar, so it’s not just about displaying text…robust local caching, handling network degradation gracefully and more are crucial to create the best experience.” What about skills? You should be an experienced software engineer and have released apps in the Play Store. We use Git to collaborate and peer code review as our primary mechanism for keeping code bases healthy. Engineers are expected to provide meaningful code review feedback to others regularly. Though not a requirement, willingness to jump into the back-end is definitely a plus. Do you think big? Our long-term goals for the mobile Medium experience are not modest: narration, internationalization, content creation, collaboration, live discussion, and more. We’re only just getting started. We want strong engineers who think systematically and innovate eagerly on the path to building amazing products. Many of the features we plan to launch by the end of this year were conceived by engineers who were able to lead the efforts to launch them. You can read more broadly about the qualities of an engineer that resonate with us. So what’s the catch? Well, there isn’t one. If you love building mobile products, having input in what you are building, and working with smart people, we want to hear from you!
    Verified by
    4007917
    Community Manager
    294886
    Engineer
    1677490
    Software Engineer
    206453
    Head of Engineering
    You may also like
    E-Commerce at Scale: Inside Shopify's Tech Stack
    How SendGrid Scaled to 40 Billion Emails Per Month
    How Stream Built a Modern RSS Reader With JavaScript
    How Heap Built an Analytics Platform that Auto-Tracks Every User Event