June 2022 Summaries

Will Jordan, an SRE worker at Fly.io, explains how serverless platforms like Lambda work by using virtual machines (VMs) to run functions on dedicated hardware around the world. He describes the design of a serverless platform and its implications for concurrency, pricing, APIs, and resilience. The author emphasizes that while "serverless" may seem like magic, there are servers involved behind the scenes.

In recent weeks, Fly.io has made significant updates to its Phoenix/LiveView-based web UI, including the addition of a new YUL region in Montreal, Canada. The company has also optimized API queries in flyctl for faster performance and added support for Laravel launcher projects using Swoole or RoadRunner. Additionally, Fly.io has improved DNS propagation pre-verification with Let's Encrypt, updated logic to prevent multiple reloads of data about an app's current image, and optimized API calls used for first deploys. Other updates include source coverage and test analytics for fly-proxy's all-Rust codebase, additional TLVs in the Rust proxy's PROXYPROTO v2 support, and a dedicated internal reference doc for diagnostics and troubleshooting. Furthermore, Fly.io has made UI improvements to its Account Settings dashboard components, added a page showcasing serious documents like DPA and HIPAA BAA, and simplified the Plans UI with an organization dropdown selector. Several bugs have also been fixed, including issues related to flyctl's env variables, deleting organizations, password-reset links, listing app secrets, moving apps, and a bad if causing a 500 error on the app activity dashboard.

The text discusses how loading a web page works and provides an overview of Fly.io's Rust-based proxy, fly-proxy. It explains that when typing https://fly.io in your browser address bar and hitting enter, the computer processes events in-order through various stages including DNS lookup, TCP connection establishment, and HTTP/2 request handling. The text also discusses memory leaks in fly-proxy and how they were fixed using a Rust profiler called Bytehound. It highlights that even though RAII is working as expected in Rust, there can still be leaks due to improper resource management or cancellation of background tasks before shutting down the application. The issue was resolved by removing custom Honeycomb tracer implementation and optimizing memory usage for queues and latency tracking.

Fly.io, a platform that makes it easy to host applications worldwide like a CDN hosts HTML pages, has released its latest changelog with several updates and features. These include faster syncing of WireGuard peers, improved proxy lock contention for tracing instrumentation, build-time secrets support in fly deploy, faster rolling deploys, Jupyter launcher requiring credit card information, intelligent longpolling fallback, permanent redirects to prevent Google indexing, and more. The platform also fixed issues with automatic redeployment of Turboku apps from Heroku deploys, deletion of users who were admins on deleted orgs, loading app metrics page, and visibility of coupons for all customers.

Fly.io, a platform that runs container images on hardware worldwide, has released its latest changelog highlighting several new features and improvements. These include an example Node.js app demonstrating how to expose multiple internal ports on separate public ports, rolling restarts for the fly restart command, and extended visibility of failed VMs. Additionally, a sample Terraform project for orchestrating Fly API, machines, and DNSimple has been created. Furthermore, a getting-started guide for Laravel apps is now available along with an official Terraform provider. Other updates include the addition of direct GraphQL endpoints for fetching volumes, IPs, and certificates by ID, improved web dashboard features, and bug fixes across various areas of the platform.