Why Render Sticks with Go and Never Plans to Rewrite Its Load Balancer in Rust

Remote Firefight: Render’s 58 k QPS vs Cloudflare’s 58 M QPS

Render’s CEO Anurag Goel posted a tongue‑in‑cheek tweet stating that their Go‑written load balancer handles more than 1500 billion HTTP requests per month, roughly 58 k queries per second (QPS). In contrast, Cloudflare rewrote its proxy (Pingora) in Rust to break through a 58 M QPS bottleneck—about a thousand‑fold higher.

“We process over 1500 billion requests per month with Go.” “The number of times we want to rewrite it in Rust is: zero.” “Go is the most underrated language in infrastructure. Its ultimate feature is being ‘boring.’”

Commenters quickly highlighted the core mismatch: the scale at which Rust becomes necessary (tens of millions of QPS) versus Render’s current load (tens of thousands of QPS). At 58 k QPS, Go, Java, or even Node.js can comfortably serve the traffic, and Go’s concise syntax, fast compilation, and mature concurrency model make it the optimal choice for this stage.

“If a system is stable at this scale, there’s no reason to switch stacks. Bringing Rust in just to showcase Go’s strengths feels unnecessary.”

The discussion then shifted from pure performance to philosophy. One developer posed the classic trolley‑problem: “If you could only use one language for the rest of your life, which would you choose?” The responses revealed two opposing mindsets.

Go supporters champion an “80‑point” approach: 80 % performance for 95 % development efficiency, avoiding the extra complexity of chasing the last 20 % of speed. Rust advocates argue for “100‑point” performance and safety, accepting roughly a 50 % drop in development speed and longer compile times.

In the AI era, another angle emerged. A developer claimed Go is currently the best language for writing large language model (LLM) code because its simple syntax, enforced gofmt, and clear control flow lower the cognitive load of human review of AI‑generated code. Conversely, the Rust‑based streaming database RisingWave argues that Rust’s zero‑cost abstractions and hardware‑level performance make it ideal for building the next generation of AI infrastructure.

The article concludes with three takeaways for architects:

1. Beware the “language hammer”

Choosing a tool should be driven by the problem, not by a bias toward a single language.

2. Embrace “boring” reliability

Stable systems that silently handle massive traffic deliver more business value than flashy, experimental stacks that require constant firefighting.

3. Value the language you can maintain

The best language is the one your team can sustain for years, backed by deep domain knowledge and a culture of rapid issue resolution.

Ultimately, the debate has no winner; it underscores that Go’s pragmatic simplicity and Rust’s rigorous performance are both valid paths toward building reliable software at scale.