2023-10-16 –, Grand Ballroom II-III
Production server infrastructure is a complicated beast that requires configuring and coordinating dozens of tools and services. You have a new Django application and you're ready to deploy it; what next? You have an existing Django application and you set up the servers yourself; what can you do better?
I’m the co-founder CTO of Zagaran, Inc., a software consulting company. Over the past 10 years, we’ve built, maintained, and deployed dozens of Django websites, and have an extensive playbook for how to do that well. In this talk, we'll draw from that playbook and go through the main issues for a creating a robust and secure Django deployment. For each issue, we'll look at the technologies and techniques to solve it. We'll focus on AWS as a hosting platform, but the techniques at play will work on any major cloud provider. This talk will cover the following:
* Application hosting
* Resilience to server failures
* Automating deployment
* Secrets management
* File storage
* Error monitoring
* Server maintenance
* (and more!)
Motivation
One mistake in a server setup can lead to major downtime or a security breach of a website. Short of that, it can cause a lot of headaches for developers. This talk will cover the key issues that people encounter in creating and maintaining production deployments, and present one or more battle-tested solutions to each of those problems.
There are a lot of technologies involved in a good server setup, and the list at first may seem overwhelming. This talk is going to be an overview of the technologies in question. For each suggested technology, the goal is to leave the audience with an understanding of where each technology fits in the deployment ecosystem and what they would be using it for, so they know what documentation to look for when they are working with it.
Target Audience
This talk is aimed at anyone that is working with the servers and/or deployment process of a production Django project. For people setting up a new set of servers for a Django project, this talk should be a good overview of the approach they need to take. For people with an existing server setup for a Django project, this talk should help them identify issues that they had missed and ways to make their servers easier to manage.
Technological goals
A good server setup has the following properties:
- Server setup is resilient to individual server failures
- Deployments are robust and fully automated
- Code can be tested on a representative environment (ex. staging) before being deployed to production
- Data and secrets are properly secured and backed-up
- Errors are sent to developers
- Servers can be analyzed for problems
- Resources can be scaled in response to load
These properties are the framing for the talk. All the technologies and techniques in this talk serve at least one of these goals.
Outline
The following is an outline of the presentation. Each bullet point roughly corresponds to one slide (on average, one minute of content).
Intro:
* Who I am
* Motivation
* Goals of a good deployment
* Talk outline
Hosting:
* Elastic beanstalk
* ECS
* EKS (Kubernetes)
* Terraform
* Gunicorn
* Docker
* Load balancers
Deployment:
* Deploying EB
* Deploying ECS
* Deploying EKS
* Full continuous deployment (zero-click deployments)
* Updating system packages
Databases and Migrations:
* RDS and backups
* Encryption in transit and at rest
* Security groups for RDS
* Running migrations on EB (configure a deployment command)
* Running migrations on ECS (run a task)
* Running migrations on EKS (run a job)
* Django Migrations race condition
* Downtime flag for deployments
* Automatic database resizing
Static assets and Media:
* WhiteNoise plus on-server hosting for static assets
* CloudFront for static assets
* django-storages + S3 for media assets
* Encryption in transit and at rest
Secrets management:
* django-environ
* Amazon secrets manager
HTTPS:
* Amazon certificate manager
* Disabling insecure versions of SSL and Amazon SSL policies
* HTTP -> HTTPS redirect
Monitoring/Error Reporting:
* Sentry (error monitoring tool)
* UptimeRobot (uptime monitoring tool)
* DataDog (error monitoring plus profiling tool)
Performance:
* CloudWatch (profiling tool)
* NewRelic (profiling tool)
* Load testing
* Manual scaling
* Auto-scaling
Extras:
* Dependency pinning: use pip-tools or poetry
* SSH-ing on to a server: use Amazon Session Manager; for EKS, install Amazon SSM agent
Zags is the co-founder and CTO of Zagaran, Inc., a software consulting company. He has led dozens of full-stack software development projects between both the private and public sectors. Zags graduated from Harvard in 2012 with bachelor’s and master’s degrees in Computer Science. He previously worked at Google, mentors for TechStars, and is a published game theorist.