Unleash Rust's Potential on AWS

Rust at Scale on AWS

Amazon Aurora DSQL - 100% Rust

• Amazon Aurora DSQL is a serverless, distributed SQL database built 100% in Rust
• Originally started as a 100% JVM/Kotlin project, but faced issues with garbage collection and latency
• The Rust-based adjudicator component was 10x faster than the Kotlin version
• The entire data plane was rewritten in Rust, achieving 10x performance over Kotlin with no optimization
• Rust was chosen for its speed, safety, and ability to handle the latency-sensitive, globally distributed nature of the database

Other Rust Projects at AWS

• Bottlerocket (the OS behind AWS Firecracker microVMs) is built in Rust
• Firecracker (the microVM runtime) is also written in Rust
• The AWS MCP (Identity and Access Management) server and CLI tool are built in Rust
• Rust is the default choice for data plane components across many AWS services

Rust on AWS Lambda

• Cargo Lambda makes it easy to develop, test, and deploy Rust-based Lambda functions
• Rust Lambda functions can achieve cold starts of 1.2 seconds and warm starts of 4ms
• Rust Lambda functions use minimal memory (29MB) and are faster than most other runtimes, including Python and JavaScript

Rust for Front-end and Full-stack

• Rust can be used to build front-end applications using technologies like HTMX
• The presenter demonstrated a link shortener application built entirely in Rust, from the static website to the Lambda functions and database

Rust at DataDog

Serverless Observability with Rust

• DataDog's legacy Go-based Lambda extension was replaced with a Rust version, reducing cold start times from 700-800ms to just 80ms
• The Rust extension was able to achieve significantly lower post-runtime durations (from 60-140ms down to 500μs) by optimizing the flushing strategy
• This allowed DataDog to remove 53,000 lines of Go code from their agent

Rust in the DataDog Agent

• DataDog is rewriting parts of their agent's data plane in Rust, achieving a 25% reduction in CPU usage and 33% reduction in memory usage compared to the Go version
• Rust's performance characteristics and memory management are well-suited for the high-volume, time-series data processing required in the agent

Rust in DataDog's Backend

• DataDog's backend uses a distributed, time-series database powered by a custom Rust storage engine (Monle)
• The Rust-based storage engine was able to handle 60x more write throughput (3.5 million points/second) compared to the previous Go-based implementation
• Rust's low-level control and lack of garbage collection overhead were key to achieving this level of performance

Production Tips for Rust Developers

Linting and Error Handling

• Use Clippy to lint your Rust code and enforce best practices, such as avoiding unwrap() calls
• Treat Clippy warnings as failures to ensure they are addressed early in development

Monitoring Rust Programs

• Monitor for blocking tasks in your Tokio-based Rust programs using the Metrics crate
• Leverage tools like the Tokio runtime's stats dump and the Tokio Console to debug performance issues

Working with AWS SDKs

• Carefully parse AWS SDK errors to handle specific error conditions, such as provisioned throughput exceeded exceptions
• This allows for more intelligent application-level throttling and error handling

Key Takeaways

• Rust has become a critical language for performance-sensitive components across AWS services, from databases to serverless functions
• Rust's speed, safety, and concurrency features make it an ideal choice for building highly scalable, distributed systems
• DataDog has seen significant benefits from adopting Rust, including improved observability, reduced resource usage, and higher throughput in their backend
• Rust developers should leverage the ecosystem's tooling, such as Clippy and Tokio's monitoring capabilities, to ensure production-ready Rust applications