Node.js vs Golang: Performance Benchmark in Kubernetes #210

Summary

This video compares Node.js and Golang performance using a web application. Both are deployed on a Kubernetes cluster in AWS. Metrics include CPU, memory usage, availability, CPU throttling, and network pressure. A persistence layer test using Postgres database measures data insertion latency and connection pool scaling. Node.js shows higher CPU and memory usage with larger latency, handling around 9,000 requests compared to Golang's 70,000. The video invites viewer feedback and contributions for improvement.

Takeaways

• 🌐 The video compares Node.js with Golang by creating web applications using Express for Node.js and the standard library for Golang.
• 📊 Both applications are deployed on a Kubernetes cluster in AWS using the latest EC2 instances to measure performance metrics like CPU and memory usage.
• 📈 The first test measures CPU usage, memory usage, application availability, CPU throttling, and network pressure without a database.
• 🔄 Node.js uses more CPU and transmits more data due to sending more headers by default compared to Golang.
• 🚀 Golang can handle significantly more requests (around 70,000) compared to Node.js (around 9,000) under the same conditions.
• 🗂️ The second test introduces a persistence layer using a Postgres database and measures latency for data insertion.
• 📚 Node.js and Golang show similar latency at the beginning of the second test, but Node.js reaches its limit at 4,000 requests per second.
• 🚨 Node.js experiences degradation at 60% CPU usage, which is significantly lower than Golang's performance ceiling.
• 🔌 The connection pool for both applications reaches its maximum size quickly, indicating the need for efficient database connection management.
• 💻 The video invites viewers to contribute to the applications' improvement by submitting pull requests on GitHub.
• 📊 The video provides detailed graphs for each test, including requests per second, latency, CPU and memory usage, availability, and CPU throttling.

Q & A

What are the two web application frameworks being compared in the video?

The video compares Node.js with Golang, using Express for Node.js and the standard library for Golang.

What environment is used to deploy and test the applications?

The applications are deployed to a production-ready Kubernetes cluster in AWS using the latest generation EC2 instances as Kubernetes nodes.

What performance metrics are measured during the tests?

The tests measure CPU usage, memory usage, application availability, CPU throttling by Kubernetes, network pressure, and the number of bytes each application receives and transmits.

What is the impact of using KeepAlive on the performance metrics?

Using KeepAlive allows a single TCP connection to handle multiple HTTP requests, which can affect the number of headers and the amount of data transmitted.

What changes are made for the second test after introducing a persistence layer?

In the second test, a POST request with a JSON payload is sent to each application, which then generates a UUID for the device and saves it to a database. The latency for each application and the connection pool scaling are also measured.

Which database is used for the benchmark in the second test?

A Postgres relational database is used for the benchmark in the second test.

What is the maximum connection pool size set for both applications in the second test?

The maximum connection pool size is set to 20 for both applications in the second test.

What is the observed CPU usage behavior of Node.js when the applications start to fail?

Node.js reaches 60% CPU usage and gets stuck, with only the latency of the requests increasing for the remainder of the test.

How does the video presenter suggest improving Node.js performance?

The presenter suggests that increasing the number of replicas might be better for Node.js performance, and they also welcome pull requests for improvements.

What is the final result of the first test in terms of requests each application can handle?

Go can handle around 70,000 requests, and Node around 9,000.

What is the presenter's approach to running the database migrations?

The presenter uses an open-source Postgres Docker image to run migrations, including creating a user, cleaning up idle connections, and creating a table, using an init container to run these migrations every time the application is deployed.