Under the Hood: Enhancing Karafka’s CPU and Memory Efficiency

Table of Contents

• 1 Introduction
• 2 Shaving Memory Allocations
• 3 If a Tree Falls: Fast Paths for Unmonitored Events
• 4 Timing Time
• 5 Conclusion

Introduction

Now and then, I like to go on a performance improvement hunt because life isn't just about adding new features. Recently, I have been focusing on enhancing efficiency, particularly regarding CPU and memory usage in Karafka. Three of my recent pull requests (PR 117, PR 118, PR 123), have made some minor improvements, and this article is about them. These changes help reduce memory allocation and improve time tracking management in Karafka and WaterDrop.

Most of the time, such improvements are not significant, but when applied in crucial places, they can make things a bit faster.

When doing OSS, I think of myself as a middleman. Karafka runs in tens of thousands of processes, and improvements affecting the consumption or production of messages (especially when applicable per message) can make a difference when multiplied.

Shaving Memory Allocations

PR 117 targets memory savings by optimizing instrumentation data handling. The primary change involves reducing unnecessary array allocations during instrumentation.