How the Ruby Global VM Lock impacts app performance

Summary of the Talk "How the Ruby Global VM Lock impacts app performance"

In this talk, Ivo Anjo, a senior software engineer at Datadog, discusses the implications of Ruby's Global VM Lock (GVL) on application performance. He begins by explaining that the GVL, often compared to Python's Global Interpreter Lock (GIL), regulates access to the Ruby VM. This prevents multiple Ruby threads from executing Ruby code simultaneously, impacting latency and overall application performance. The key points covered include:

What is the GVL?: The GVL is a mechanism ensuring the correctness of the multi-threaded C program that is Ruby. It allows for concurrency (the illusion of simultaneous execution) but not true parallelism.
Why Concurrency Primitives are Necessary: While the GVL protects Ruby VM’s internal state, it does not safeguard application-level code, where race conditions can lead to unexpected results.
Impact on Performance: The GVL can significantly affect performance as adding more threads increases contention for the GVL, leading to higher latency. Through various examples, he demonstrates how threads can impact each other, especially in I/O operations or when waiting for the GVL to become available.
JVL Tracing Gem: The introduction of the gvl-tracing gem allows developers to visualize the state of threads and assess their waiting times and interactions with the GVL. This tool helps identify performance bottlenecks.
Ractors and M:N Scheduling: Anjo discusses how Ruby 3.0 introduced Ractors, which provide a way to achieve true parallelism by creating separate execution contexts, enabling multiple threads to run concurrently without being gated by the GVL. Additionally, M:N scheduling allows more Ruby threads to share fewer operating system (OS) threads, thus improving throughput.

In conclusion, Anjo emphasizes the importance of profiling Ruby applications in real-world environments to identify performance issues related to the GVL. He encourages developers to experiment with different concurrency models, such as Ractors, to optimize application performance and manage latency more effectively. He leaves the audience with practical recommendations for tuning Ruby applications and highlights the continuous evolution of the Ruby ecosystem.

How the Ruby Global VM Lock impacts app performance
Ivo Anjo • November 13, 2024 • Chicago, IL • Talk

Have you ever heard of Ruby's Global VM Lock? Maybe you've heard of the GIL instead?

Spoiler: It's an implementation detail of Ruby that can have a big impact on latency and performance of applications.
At a high-level, it prevents Ruby code across multiple threads from running in parallel, while still safely allowing concurrency.

Join me as we explore what the GVL is; what it is not; and how we can use the `gvl-tracing` gem to show it in action. We’ll also look into how Ractors, M:N scheduling and fibers fit into this story. By the end of the talk you’ll have the tools to evaluate if your own applications are being impacted, as well as learn a number of do’s and don’ts of Ruby performance.

RubyConf 2024