[KEP 4205] Blog post for PSI metrics GA by mariafromano-25 · Pull Request #54709 · kubernetes/website

mariafromano-25 · 2026-02-27T04:04:59Z

Description

This commit adds a new blog post to announce that Pressure Stall Information (PSI) Metrics has graduated to Stable (GA) in Kubernetes v1.36.

Issue

Ref: kubernetes/enhancements#4205

/sig node
Closes: #

netlify · 2026-02-27T04:15:16Z

✅ Pull request preview available for checking

Built without sensitive environment variables

Name	Link
🔨 Latest commit	`1a5659f`
🔍 Latest deploy log	https://app.netlify.com/projects/kubernetes-io-main-staging/deploys/69e7f705fd47920008b63f26
😎 Deploy Preview	https://deploy-preview-54709--kubernetes-io-main-staging.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

To edit notification comments on pull requests, go to your Netlify project configuration.

chadmcrowell · 2026-03-09T22:06:58Z

Hi @mariafromano-25 👋 v1.36 Communications team here,

@ttsuuubasa as author of #54541, I'd like you to be a writing buddy for @mariafromano-25 on this PR.

Please:

Review this PR, paying attention to the guidelines and review hints
Update your own PR based on any best practices you identify that should be applied
Remember to be compassionate with your fellow article author

UtkarshUmre · 2026-04-04T15:07:05Z

Hi @mariafromano-25 👋 -- this is Utkarsh (@UtkarshUmre ) from the v1.36 Communications Team!

Just a friendly reminder that we are approaching the feature blog "ready for review" deadline: Monday, 6 April. We ask you to have the blog in non-draft state, and all write-up to be complete, so that we can start the blog review from SIG Docs Blog team.

If you have any questions or need help, please don't hesitate to reach out to me or any of the Communications Team members. We are here to help you!

mariafromano-25 · 2026-04-07T06:45:50Z

cc @roycaihw @ndixita

lmktfy

This is basically OK but I do have a lot of small recommended fixes.

lmktfy · 2026-04-07T18:16:59Z

+
+Since its original implementation in the Linux kernel in 2018, Pressure Stall Information (PSI) has provided users with the high-fidelity signals needed to identify resource saturation before it becomes an outage. Unlike traditional utilization metrics, PSI tells the story of tasks stalled and time lost, all in nicely-packaged percentages of time across the CPU, memory, and I/O.
+
+Today, we are excited to announce that Kubelet-integrated PSI metrics have graduated to **General Availability (GA)** in Kubernetes v1.36. This graduation ensures that users across the ecosystem have a stable, reliable interface to observe resource contention at the node, pod, and container levels.


This is a post release blog article. It is not the announcement; it is a detailed follow up. Please reword accordingly.

Thank you for the callout, I reworded to reference the recent release instead of announcing.

lmktfy · 2026-04-07T18:26:30Z

+
+You can read more about our performance tests [here](https://docs.google.com/document/d/1ffv7pleid3uk0dT9euK5vkso41i9tscNoEVgeshfiCc/edit?usp=sharing&resourcekey=0-_xKPCU4zqiGU0e7Q8jDJRw).
+
+### Getting Started


Do you need to use a particular kernel or does it Just Work?

Yes, in the "Getting Started" section I specified

roycaihw · 2026-04-07T22:25:50Z

+1. **Kernel PSI OFF / Kubelet Feature ON** (Baseline)
+2. **Kernel PSI ON / Kubelet Feature ON** (Kernel Scheduler overhead)
+3. **Kernel PSI ON / Kubelet Feature OFF** (Default Baseline)
+4. **Kernel PSI ON / Kubelet Feature ON** (Feature fallback behavior)


nit: Why is this called "fallback"? Isn't kernal ON + kubelet ON = the feature?

I was trying to list out the total conditions but I can see how it was confusing from the overlap. I reworded to clearly outline both cases, one to isolate kernel overhead and the other for kubelet overhead.

roycaihw · 2026-04-07T22:30:59Z

+{{< figure src="/images/node_sys_cpu_usage_rate_comparison.png" alt="A line graph comparing the Node System (Kernel) CPU usage rate over elapsed time with the PSI feature turned OFF versus ON." title="Node System CPU Usage Rate Comparison" >}}
+*Figure 1: Node System CPU comparison under load (80 pods).*
+
+As seen in Figure 1, the "Kernel Tax" for enabling PSI is remarkably low. Even under heavy I/O and CPU load, the **System CPU** delta between the PSI-enabled (red) and PSI-disabled (blue) clusters remained consistently under **0.2 cores** and over **0.037 cores** for the most part. This confirms that simply enabling the feature does not raise the pre-existing resource use and that the internal kernel bookkeeping for stall tracking is safe for production-scale deployments.


Reading this, it not super clear to me what two scenarios that we are comparing.

the "Kernel Tax" for enabling PSI is remarkably low

Does "enabling PSI" mean the Kubernetes feature? And is the kernel PSI enabled or disabled?

You're right, it was confusing reading back. I added explanations for both cases with their corresponding graphs

roycaihw · 2026-04-07T22:32:10Z

+
+As seen in Figure 1, the "Kernel Tax" for enabling PSI is remarkably low. Even under heavy I/O and CPU load, the **System CPU** delta between the PSI-enabled (red) and PSI-disabled (blue) clusters remained consistently under **0.2 cores** and over **0.037 cores** for the most part. This confirms that simply enabling the feature does not raise the pre-existing resource use and that the internal kernel bookkeeping for stall tracking is safe for production-scale deployments.
+
+{{< figure src="/images/kubelet_cpu_usage_rate_comprison.png" alt="A line graph comparing the Kubelet CPU usage rate over elapsed time with the PSI feature turned OFF versus ON." title="Kubelet CPU Usage Rate Comparison" >}}


I think this should be the main measurement that we are demonstrating-- kubelet overhead when the k8s feature is enabled v.s. disabled (while PSI is enabled at kernel level).

roycaihw · 2026-04-07T22:33:50Z

+You can read more about our performance tests [here](https://docs.google.com/document/d/1ffv7pleid3uk0dT9euK5vkso41i9tscNoEVgeshfiCc/edit?usp=sharing&resourcekey=0-_xKPCU4zqiGU0e7Q8jDJRw).
+
+### Getting Started
+As of v1.36, the `KubeletPSI` feature gate is enabled by default. You can query the Kubelet Summary API to see real-time pressure data:


We should document the requirements:

Kernel version

cgroup v2

See https://kubernetes.io/blog/2025/09/04/kubernetes-v1-34-introducing-psi-metrics-beta/#how-to-enable-psi-metrics as an example

roycaihw · 2026-04-07T22:35:00Z

+
+You can read more about our performance tests [here](https://docs.google.com/document/d/1ffv7pleid3uk0dT9euK5vkso41i9tscNoEVgeshfiCc/edit?usp=sharing&resourcekey=0-_xKPCU4zqiGU0e7Q8jDJRw).
+
+### Getting Started


Can we also mention this improvement that was done in 1.36? kubernetes/kubernetes#137326

Great idea! Just added that as well before the "Getting Started" section.

ttsuuubasa

Everything looks fine to me on my end. Congrats on GA!

Co-authored-by: Tim Bannister <[email protected]>

roycaihw · 2026-04-20T23:11:50Z

+To use PSI metrics in your Kubernetes cluster, your nodes must meet the following requirements:
+
+1. **Ensure your nodes are running a Linux kernel version 4.20 or later and are using cgroup v2.**
+2. **Ensure PSI is enabled at the OS level** (your kernel must be compiled with `CONFIG_PSI=y` and must not be booted with the `psi=0` parameter).


must not be booted with the psi=0 parameter

What's the default? If the default is 1, this looks good. If the default it 0, shall we just say must boot with psi=1 instead?

yup default is 1!

roycaihw · 2026-04-21T19:19:27Z

+Our testing focused on two primary scenarios to isolate the impact of the Kubelet and kernel-level collection respectively:
+1. **Kernel PSI ON / Kubelet Feature OFF** vs **Kernel PSI ON / Kubelet Feature ON** (Kubelet overhead)
+2. **Kernel PSI OFF / Kubelet Feature ON** vs **Kernel PSI ON / Kubelet Feature ON** (Kernel overhead)
+
+#### Scenario 1: The Kubelet Overhead
+First, we evaluated the Kubelet overhead (Case 1) on 4 core machines. For these tests, the Linux kernel was already tracking pressure on both clusters by default(`psi=1`), but we toggled the `KubeletPSI` feature gate to see if the Kubelet actively querying and exposing these metrics impacted the resource usage. As seen in the following graph, the **System CPU** usage lines for the Kubelet PSI-enabled (red) follows the same pattern as the Kubelet PSI-disabled (blue) clusters, with a slight expected increase and delay from the baseline. This visualizes that once the OS is tracking PSI, at around **2.5 cores**, the act of Kubernetes reading those cgroup metrics is negligible to performance. 
+
+{{< figure src="/images/kubeletPSI_sys_cpu_usage_rate_graph.png" alt="A line graph comparing the system CPU usage rate over elapsed time with the PSI feature turned off versus on and kernel PSI off." title="(Case 1) System CPU Usage Rate Comparison" caption="Figure 1: Node System CPU Usage Rate Comparison." >}}


You mentioned case 1 is "Kernel PSI ON / Kubelet Feature OFF vs Kernel PSI ON / Kubelet Feature ON", so I expected kernel PSI to always be ON in the comparison. I don't get why the graph is showing (kernel) PSI ON v.s. OFF. Did I miss something?

You're correct, Kernal PSI stays ON for case 1. The "PSI On/Off" is referring to the Kubelet feature gate. OH i think I put "off" instead of "on" on the alt field

Just updated the alt field and the graph line labels to be clearer

roycaihw

/lgtm
/approve
Thanks!

k8s-ci-robot · 2026-04-21T22:29:05Z

LGTM label has been added.

Details

Git tree hash: 0d73854c29d35c9aabec61610dbc564535c22d03

chadmcrowell · 2026-04-22T20:46:12Z

/assign @nate-double-u

lmktfy · 2026-04-23T18:21:28Z

@ttsuuubasa you were suggested as the writing buddy for this PR.

Would you be willing to provide a review?

lmktfy

/lgtm
/approve

/hold
for release comms to confirm we can include this one

k8s-ci-robot · 2026-04-26T12:57:00Z

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: lmktfy, roycaihw

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Details

Needs approval from an approver in each of these files:

~~OWNERS~~ [lmktfy]
~~content/en/blog/OWNERS~~ [lmktfy]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

SwathiR03 · 2026-04-27T16:54:46Z

/unhold

WIP: Add placeholder blog post for PSI metrics GA

af24850

k8s-ci-robot added do-not-merge/work-in-progress Indicates that a PR should not merge because it is a work in progress. sig/node Categorizes an issue or PR as relevant to SIG Node. labels Feb 27, 2026

github-project-automation Bot added this to SIG Node: code and documentation PRs Feb 27, 2026

github-project-automation Bot moved this to Triage in SIG Node: code and documentation PRs Feb 27, 2026

k8s-ci-robot added the cncf-cla: yes Indicates the PR's author has signed the CNCF CLA. label Feb 27, 2026

k8s-ci-robot requested review from Gauravpadam and lmktfy February 27, 2026 04:05

k8s-ci-robot added area/blog Issues or PRs related to the Kubernetes Blog subproject language/en Issues or PRs related to English language size/S Denotes a PR that changes 10-29 lines, ignoring generated files. labels Feb 27, 2026

haircommander moved this from Triage to Waiting on Author in SIG Node: code and documentation PRs Mar 4, 2026

chadmcrowell mentioned this pull request Mar 9, 2026

Document DRA Device Binding Conditions in v1.36 #54541

Merged

k8s-ci-robot added sig/docs Categorizes an issue or PR as relevant to SIG Docs. size/M Denotes a PR that changes 30-99 lines, ignoring generated files. and removed size/S Denotes a PR that changes 10-29 lines, ignoring generated files. labels Apr 7, 2026

mariafromano-25 force-pushed the psi-ga-blog branch from 7bf0428 to ffa5893 Compare April 7, 2026 06:26

mariafromano-25 marked this pull request as ready for review April 7, 2026 06:55

k8s-ci-robot requested a review from graz-dev April 7, 2026 06:56

mariafromano-25 changed the title ~~WIP: Add placeholder blog post for PSI metrics GA~~ [KEP 4205] Blog post for PSI metrics GA Apr 7, 2026

k8s-ci-robot removed the do-not-merge/work-in-progress Indicates that a PR should not merge because it is a work in progress. label Apr 7, 2026

psi blog with graphs

1fae618

mariafromano-25 force-pushed the psi-ga-blog branch from ffa5893 to 1fae618 Compare April 7, 2026 07:04

lmktfy reviewed Apr 7, 2026

View reviewed changes

roycaihw reviewed Apr 7, 2026

View reviewed changes

ttsuuubasa reviewed Apr 8, 2026

View reviewed changes

ndixita mentioned this pull request Apr 8, 2026

Support PSI based on cgroupv2 kubernetes/enhancements#4205

Closed

34 tasks

mariafromano-25 and others added 4 commits April 10, 2026 12:58

Apply suggestions from code review

76a7986

Co-authored-by: Tim Bannister <[email protected]>

added case 2 graphs and explained changes and more requirements

344e453

fixed typo in figure 2 img path

ecab8fa

more typo clean up, myb

52ed9e3

roycaihw reviewed Apr 20, 2026

View reviewed changes

swaped scenario order, data as percentages

a09dcb1

roycaihw reviewed Apr 21, 2026

View reviewed changes

clearer graph labels, reordered for kubeletpsi first

1a5659f

roycaihw approved these changes Apr 21, 2026

View reviewed changes

k8s-ci-robot assigned roycaihw Apr 21, 2026

k8s-ci-robot added the lgtm "Looks good to me", indicates that a PR is ready to be merged. label Apr 21, 2026

k8s-ci-robot assigned nate-double-u Apr 22, 2026

lmktfy reviewed Apr 26, 2026

View reviewed changes

k8s-ci-robot added the do-not-merge/hold Indicates that a PR should not merge because someone has issued a /hold command. label Apr 26, 2026

k8s-ci-robot assigned lmktfy Apr 26, 2026

k8s-ci-robot added the approved Indicates a PR has been approved by an approver from all required OWNERS files. label Apr 26, 2026

k8s-ci-robot removed the do-not-merge/hold Indicates that a PR should not merge because someone has issued a /hold command. label Apr 27, 2026

k8s-ci-robot merged commit 8f896ea into kubernetes:main Apr 27, 2026
6 checks passed

github-project-automation Bot moved this from Waiting on Author to Done in SIG Node: code and documentation PRs Apr 27, 2026


		Since its original implementation in the Linux kernel in 2018, Pressure Stall Information (PSI) has provided users with the high-fidelity signals needed to identify resource saturation before it becomes an outage. Unlike traditional utilization metrics, PSI tells the story of tasks stalled and time lost, all in nicely-packaged percentages of time across the CPU, memory, and I/O.

		Today, we are excited to announce that Kubelet-integrated PSI metrics have graduated to General Availability (GA) in Kubernetes v1.36. This graduation ensures that users across the ecosystem have a stable, reliable interface to observe resource contention at the node, pod, and container levels.


		You can read more about our performance tests [here](https://docs.google.com/document/d/1ffv7pleid3uk0dT9euK5vkso41i9tscNoEVgeshfiCc/edit?usp=sharing&resourcekey=0-_xKPCU4zqiGU0e7Q8jDJRw).

		### Getting Started


		As seen in Figure 1, the "Kernel Tax" for enabling PSI is remarkably low. Even under heavy I/O and CPU load, the System CPU delta between the PSI-enabled (red) and PSI-disabled (blue) clusters remained consistently under 0.2 cores and over 0.037 cores for the most part. This confirms that simply enabling the feature does not raise the pre-existing resource use and that the internal kernel bookkeeping for stall tracking is safe for production-scale deployments.

		{{< figure src="/images/kubelet_cpu_usage_rate_comprison.png" alt="A line graph comparing the Kubelet CPU usage rate over elapsed time with the PSI feature turned OFF versus ON." title="Kubelet CPU Usage Rate Comparison" >}}

Conversation

mariafromano-25 commented Feb 27, 2026

Description

Issue

Uh oh!

netlify Bot commented Feb 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

✅ Pull request preview available for checking

Uh oh!

chadmcrowell commented Mar 9, 2026

Uh oh!

UtkarshUmre commented Apr 4, 2026

Uh oh!

mariafromano-25 commented Apr 7, 2026

Uh oh!

lmktfy left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ttsuuubasa left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

roycaihw left a comment

Choose a reason for hiding this comment

Uh oh!

k8s-ci-robot commented Apr 21, 2026

Uh oh!

chadmcrowell commented Apr 22, 2026

Uh oh!

lmktfy commented Apr 23, 2026

Uh oh!

lmktfy left a comment

Choose a reason for hiding this comment

Uh oh!

k8s-ci-robot commented Apr 26, 2026

Uh oh!

netlify Bot commented Feb 27, 2026 •

edited

Loading