FASTEST Go Web Framework: gnet vs fiber vs fasthttp vs net/http?

Summary

In this video, the presenter tests and compares the fastest Go web frameworks—Go's standard library, fasthttp, Fiber, and gNet—based on key performance metrics like latency, throughput, CPU usage, memory usage, and database integration. Through two rounds of testing, including PostgreSQL database interaction, the video demonstrates that while the standard library offers the best latency, it struggles with throughput and high CPU usage. fasthttp and Fiber are more scalable, with fasthttp performing slightly better in terms of throughput. gNet is efficient in memory and CPU but faces challenges with multi-core support and connection pooling. The conclusion offers insights for selecting the right framework based on specific needs.

Takeaways

• 😀 The Go standard library initially offers the lowest latency but struggles with high CPU usage under heavy loads.
• 😀 fasthttp and Fiber are closely matched in performance, with Fiber built on top of fasthttp for a better user experience.
• 😀 gNet is the most efficient in terms of CPU and memory usage but has limitations with its low-level API and lack of documentation.
• 😀 At high throughput (55,000 requests/second), the Go standard library becomes the slowest framework compared to others.
• 😀 gNet has one of the lowest latencies under heavy load but falls behind in terms of throughput, handling up to 98,000 requests/second.
• 😀 When testing in Kubernetes, the standard library and Fiber apps started to fail health checks around 90,000 requests/second.
• 😀 fasthttp outperforms all frameworks in the high-throughput test, handling over 100,000 requests/second with stability.
• 😀 PostgreSQL database integration reveals that gNet has a critical limitation in connection pooling, affecting performance.
• 😀 gNet's event loop-based approach limits its connection pool size, which can hinder database performance in production environments.
• 😀 The standard library is the best in terms of latency but requires scaling horizontally due to high CPU consumption, making it less scalable than other frameworks.
• 😀 Fiber provides a good balance of performance and user-friendly API, making it a solid choice for most use cases.

Q & A

Which Go web frameworks were compared in the video?

The video compares the Go standard library, fasthttp, Fiber, and gNet web frameworks.

What performance metrics were measured during the tests?

The performance metrics measured include latency, throughput (requests per second), CPU usage, memory usage, error rates, and CPU throttling.

What is the main advantage of the Go standard library in the tests?

The Go standard library has the lowest latency at the start of the test, which makes it suitable for light loads.

Why does the Go standard library become the slowest framework after a certain threshold?

The Go standard library suffers from high CPU usage after handling around 15,000 requests per second, which causes it to lose its initial advantage in performance.

What role does Fiber play in relation to fasthttp?

Fiber is a web framework built on top of fasthttp. It provides a user-friendly API and additional middleware, making it easier to use while benefiting from the speed of fasthttp.

What is gNet’s unique feature compared to the other frameworks?

gNet stands out for its small memory footprint, low CPU usage, and potential to achieve high throughput, though it has a low-level API and poor documentation.

How does gNet perform under heavy loads, and what limitation was identified?

Under heavy loads, gNet shows low latency but lags behind in throughput. A major limitation was found in its connection pool, which is limited to the number of event loops created, affecting database performance.

What is the significance of Kubernetes in the tests?

Kubernetes is used to deploy the applications, with health checks running in the Kubernetes environment. If a pod fails too quickly, it transitions to a crashloop state, requiring adjustments to scaling capacity.

What did the second test involve, and how did PostgreSQL impact performance?

The second test involved adding a PostgreSQL database to measure latency for insert operations. The database connection pool size became a limiting factor for gNet, especially due to its small pool size.

Why did gNet perform poorly in the second test with PostgreSQL?

gNet performed poorly because it limited the number of database connections to the number of event loops it created. This caused a bottleneck, affecting its ability to handle database interactions efficiently.

What conclusion did the video reach regarding the choice of framework?

The video concluded that the standard library has the best latency performance but high CPU usage. Fiber and fasthttp provide a good balance between performance and resource usage, while gNet is efficient but hindered by limitations such as its small connection pool and low-level API.