Kong Plugin Memory and CPU Usage: Find the Top Consumer Online (OpenResty XRay)

To find which Kong plugin uses the most memory or CPU, OpenResty XRay samples a running Kong gateway process and breaks resource usage down per plugin — no restarts, no code changes, and no special build of Kong. In the case below, the bot-detection plugin held the most memory while prometheus burned the most CPU.

Kong is a popular API gateway built on top of our open-source OpenResty framework, with a flexible plugin mechanism covering both community plugins and ones you write yourself. But those plugins can cause CPU bottlenecks, high memory usage, or memory leaks, and it is often hard to tell which one is to blame. OpenResty XRay is a dynamic-tracing product that answers that question on live processes — efficiently and safely, like taking an X-ray image of your software. You just point it at your Kong process; you don’t install any special plugins or rebuild Kong.

In this article, we will walk through how to use OpenResty XRay to analyze the CPU and memory usage of Kong plugins in a server process, with real example results and what they mean.

CPU usage among all Kong plugins in a server process

OpenResty XRay can sample a Kong process for some time, from a few seconds to a couple of minutes, depending on how busy the process is. It will then show you how the CPU time is distributed across all the plugins currently loaded.

For example, here is a pie chart generated by OpenResty XRay for a Kong process:

As you can see, the prometheus plugin takes most of the CPU time in this case, followed by the ip-restriction plugin. The other plugins have negligible CPU usage. This means that if you want to optimize the CPU performance of your Kong server, you should focus on these two plugins and see if they can be improved or replaced.

Of course, this is just one example. In more complex Kong servers, you may see more plugins show up here, and the distribution may vary depending on the traffic and configuration.

CPU usage inside a Kong plugin

OpenResty XRay can also help you drill down into the details of any Kong plugins and see which Lua or C code paths are consuming more CPU time. One easy way to do this is to look at the Lua-land CPU flame graphs like this:

This graph shows you the call stacks of the Lua code that are running on the CPU during the sampling period. The wider the bar, the more CPU time it takes. You can hover over any bar to see more information, such as the function name, file name, line number, and percentage of CPU time.

By looking at this graph, you can quickly spot any hot spots or performance issues in your Lua code and optimize them accordingly. You can also compare different plugins or different versions of the same plugin and see how they differ in terms of CPU usage.

When a plugin is not just heavy but abnormally hot, the same flame graphs reveal exactly why. For a worked example, see how we traced a Kong CPU bottleneck to hidden Lua exceptions in a custom plugin, all the way down to the offending string.lower call.

Memory usage among all Kong plugins in a server process

Similarly, OpenResty XRay can sample the memory usage inside any Kong server processes with ease.

Below is another pie chart generated by OpenResty XRay for the same Kong process:

Here we can see that the bot-detection plugin takes most of the memory in this case, while the prometheus plugin comes next. The other plugins have much lower memory usage. This means that if you want to reduce the memory footprint of your Kong server, you should look into these two plugins and see if they can be optimized or replaced.

Again, this is just one example. In different Kong servers, you may see different results depending on the plugins and settings.

Memory usage inside a Kong plugin

OpenResty XRay can also help you analyze memory usage across all the Lua GC objects inside any of the Kong plugins. One easy way to do this is to look at the GC object reference flame graphs showing how memory is quantitatively distributed across all the GC object reference paths.

This graph shows you the reference paths of the Lua GC objects that are occupying memory during the sampling period. The wider the bar, the more memory it takes. You can hover over any bar to see more information, such as the object type, size, and percentage of memory.

By looking at this graph, you can quickly spot any memory leaks or inefficiencies in your Lua code and optimize them accordingly. You can also compare different plugins or different versions of the same plugin and see how they differ in terms of memory usage.

If a Kong plugin’s memory keeps climbing over time, the same GC object reference flame graph traces the leaking reference path down to the objects holding it — with no restart and no code changes. See our production memory-leak detection workflow for how to take that all the way to the source line, and a real case where we traced a Lua GC leak to cached SSL certificates.

Extra overhead for the servers

You may wonder if using OpenResty XRay will affect the performance of your Kong servers. The answer is no. The extra overhead added to the Kong server processes when sampling is usually so small that it can be ignored. And when not sampling, the processes are running at absolute full speed.

OpenResty XRay is designed to be noninvasive and lightweight. It does not interfere with your normal operations or require any changes to your code or configuration.

Frequently asked questions

Which Kong plugin uses the most memory?

It depends on your plugins and traffic, so the only reliable answer is to measure your own gateway. In the process sampled here, the bot-detection plugin held the most memory, with prometheus next. OpenResty XRay breaks memory down per loaded plugin on a live Kong process, so you can see the real top consumer in your own setup — no special build required.

How can I find a memory leak in a Kong plugin?

OpenResty XRay’s GC object reference flame graph shows the reference paths holding memory, down to the object type and size. When a plugin’s memory keeps climbing, that graph pinpoints the leaking reference path, with no restart and no code changes. See our production memory-leak detection workflow for the full method, from RSS curve to source line.

Which Kong plugin uses the most CPU?

Again, measure rather than guess. In the sampled process, the prometheus plugin took most of the CPU time, followed by ip-restriction. OpenResty XRay distributes sampled CPU time across all loaded plugins, and its Lua-land CPU flame graphs drill into the exact hot code paths inside any single plugin.

Does profiling Kong plugins with OpenResty XRay add overhead?

The extra overhead added to Kong server processes during sampling is usually so small it can be ignored, and when it is not sampling, the processes run at absolute full speed. OpenResty XRay is noninvasive: it needs no changes to your Kong code or configuration and no special plugins or build options.

What’s next?

We are not stopping here. We have more plans to make OpenResty XRay even more powerful and useful for you. Some of the features we are working on include:

• Showing real-time distribution of disk I/O, network I/O, and other resource metrics among Kong plugins. This will help you identify any bottlenecks or hotspots in your system and optimize them accordingly.
• Supporting other technical stacks and open-source software. We want to make OpenResty XRay a universal tool that can analyze any online applications, regardless of the underlying technologies. Some of the targets we have in mind are Nginx modules, Envoy extensions, PostgreSQL extensions, Perl/Python/Ruby modules and libraries, and many more.

If you have any suggestions or requests for more metrics or features, please let us know. We are always listening to your feedback and improving our product to meet your needs.

Conclusion

In this article, we showed you how to use OpenResty XRay to analyze the CPU and memory usage of Kong plugins in a server process. We also showed you some examples of the results and explained what they mean.

With OpenResty XRay, you can easily find out which plugins are consuming more resources than others and how much impact they have on your overall performance. You can then use this information to optimize your Kong servers and make them run faster and smoother.

If you want to learn more about OpenResty XRay and how it can help you with other aspects of your online applications, please visit our website or contact us for more details.

About The Author

Yichun Zhang (Github handle: agentzh), is the original creator of the OpenResty^® open-source project and the CEO of OpenResty Inc..

Yichun is one of the earliest advocates and leaders of “open-source technology”. He worked at many internationally renowned tech companies, such as Cloudflare, Yahoo!. He is a pioneer of “edge computing”, “dynamic tracing” and “machine coding”, with over 22 years of programming and 16 years of open source experience. Yichun is well-known in the open-source space as the project leader of OpenResty^®, adopted by more than 40 million global website domains.

OpenResty Inc., the enterprise software start-up founded by Yichun in 2017, has customers from some of the biggest companies in the world. Its flagship product, OpenResty XRay, is a non-invasive profiling and troubleshooting tool that significantly enhances and utilizes dynamic tracing technology. And its OpenResty Edge product is a powerful distributed traffic management and private CDN software product.

As an avid open-source contributor, Yichun has contributed more than a million lines of code to numerous open-source projects, including Linux kernel, Nginx, LuaJIT, GDB, SystemTap, LLVM, Perl, etc. He has also authored more than 60 open-source software libraries.