A kubectl plugin that allows you to profile production applications with low-overhead by generating FlameGraphs
Running kubectlf-flame
does not require any modification to existing pods.
- Supported languages: Go, Java (any JVM based language), Python, Ruby, and NodeJS
- Kubernetes cluster that use Docker as the container runtime (tested on GKE, EKS and AKS)
This plugin is forked from the upstream kubectl-flame project so it is not available on Krew. To install the plugin checkout this repo and run:
kubectl krew install --manifest=.krew.yaml
This will clash with any existing installations of kubectl-flame.
In order to profile a Java application in pod mypod
for 1 minute and save the flamegraph as /tmp/flamegraph.svg
run:
kubectl flame mypod -t 1m --lang java -f /tmp/flamegraph.svg
Profiling Java application in alpine based containers require using --alpine
flag:
kubectl flame mypod -t 1m -f /tmp/flamegraph.svg --lang java --alpine
NOTICE: this is only required for Java apps, the --alpine
flag is unnecessary for Go profiling.
Pods that contains more than one container require specifying the target container as an argument:
kubectl flame mypod -t 1m --lang go -f /tmp/flamegraph.svg mycontainer
Profiling Go application in pods that contains more than one process require specifying the target process name via --pgrep
flag:
kubectl flame mypod -t 1m --lang go -f /tmp/flamegraph.svg --pgrep go-app
Java profiling assumes that the process name is java
. Use --pgrep
flag if your process name is different.
To run this tool on Kubernetes clusters that use containerd as the runtime engine, you must specify the path to the containerd runtime files:
kubectl flame mypod -t 1m --docker-path /run/containerd
You can install kubectl flame
using the Krew, the package manager for kubectl plugins.
Once you have Krew installed just run:
kubectl krew install flame
See the release page for the full list of pre-built assets.
kubectl-flame
launch a Kubernetes Job on the same node as the target pod.
Under the hood kubectl-flame
use async-profiler in order to generate flame graphs for Java applications.
Interaction with the target JVM is done via a shared /tmp
folder.
Golang support is based on ebpf profiling.
Python support is based on py-spy.
Ruby support is based on rbspy.
NodeJS support is based on perf. In order for Javascript Symbols to be resolved, node process needs to be run with --perf-basic-prof
flag.
Please refer to the contributing.md file for information about how to get involved. We welcome issues, questions, and pull requests.
This project is licensed under the terms of the Apache 2.0 open source license. Please refer to LICENSE for the full terms.