feat(k8s): support multi-container (sidecar) SkyPilot pods by php-workx · Pull Request #8444 · skypilot-org/skypilot

php-workx · 2025-12-27T06:27:47Z

SkyPilot’s Kubernetes provisioning and control-plane interactions currently assume SkyPilot pods contain a single container and implicitly target containers[0]. In real clusters, mutating webhooks often inject sidecars (log shippers, mesh proxies, security agents), which breaks this assumption and can cause provisioning failures or commands executing in the wrong container.

This PR adds multi-container awareness by consistently targeting the primary workload container (default: ray-node) when executing commands or inspecting resources.

Key changes

Select the primary container explicitly instead of assuming the first container is the workload container.
Wire container selection through Kubernetes command execution, so kubectl exec targets the intended container when sidecars are present.
Centralize the default workload container name (ray-node) as a constant: sky/provision/kubernetes/constants.py:RAY_NODE_CONTAINER_NAME.

Behavior

Existing single-container pods continue to work unchanged.
When sidecars are injected, SkyPilot continues to exec into the workload container, preventing provisioning/runtime regressions.

Tested (run the relevant ones):

Code formatting: install pre-commit (auto-check on commit) or bash format.sh
Any manual or new tests for this PR (please specify below)
All smoke tests: /smoke-test (CI) or pytest tests/test_smoke.py (local)
Relevant individual tests: /smoke-test -k test_name (CI) or pytest tests/test_smoke.py::test_name (local)
Backward compatibility: /quicktest-core (CI) or pytest tests/smoke_tests/test_backward_compat.py (local)

gemini-code-assist · 2025-12-27T06:28:07Z

Summary of Changes

Hello @php-workx, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request significantly enhances SkyPilot's compatibility with Kubernetes environments that utilize multi-container pods, particularly those where sidecars are automatically injected. By introducing explicit targeting of the primary workload container, SkyPilot can now reliably execute commands and gather resource information, preventing operational disruptions and ensuring correct behavior in complex Kubernetes deployments. This change improves the robustness and flexibility of SkyPilot's Kubernetes provisioning capabilities without affecting existing single-container setups.

Highlights

Multi-Container Pod Support: SkyPilot now explicitly targets the primary workload container (defaulting to 'ray-node') within Kubernetes pods, addressing issues caused by sidecar injection from mutating webhooks.
Consistent Container Targeting: Command execution (kubectl exec) and resource inspection now consistently use the specified primary container, preventing commands from running in unintended sidecar containers.
Centralized Configuration: The default primary container name ('ray-node') has been centralized into a constant, RAY_NODE_CONTAINER_NAME, for easier management and consistency.
Improved Resource Parsing: Resource requests (CPU, memory, GPU) are now correctly parsed from the identified primary container, ensuring accurate resource allocation and monitoring.
Removed Single-Container Assumption: The previous assertion limiting deployments to a single container has been removed, allowing for more flexible pod configurations.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	`/gemini review`	Performs a code review for the current pull request in its current state.
Pull Request Summary	`/gemini summary`	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	`/gemini help`	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

gemini-code-assist

Code Review

This pull request introduces explicit targeting of the primary workload container, named 'ray-node', within Kubernetes pods to better support multi-container deployments. Changes include adding a RAY_NODE_CONTAINER_NAME constant, modifying KubernetesCommandRunner to accept a container argument for kubectl exec and rsync operations, and implementing a get_pod_primary_container utility function. This utility is used across the codebase to ensure pre-initialization, AppArmor annotation removal, resource parsing (CPU, memory, GPU), and command execution correctly interact with the intended container. The previous restriction of single-container deployments has been removed. A review comment suggests a more concise Pythonic approach for retrieving CPU requests using dictionary get with or operators.

aylei · 2026-01-16T06:50:05Z

/smoke-test

aylei

Thanks for adding this @php-workx ! Could you include a smoke test with sidecar injected to make this PR self-contained?

Signed-off-by: Aylei <rayingecho@gmail.com>

aylei · 2026-01-23T01:06:15Z

/smoke-test -k test_kubernetes_pod_config_sidecar

aylei · 2026-01-23T11:17:40Z

/smoke-test -k test_kubernetes_pod_config_sidecar

aylei · 2026-01-23T12:52:50Z

/smoke-test -k test_kubernetes_pod_config_sidecar --kubernetes

aylei · 2026-01-23T15:35:19Z

/smoke-test -k test_kubernetes_pod_config_sidecar --kubernetes

Signed-off-by: Aylei <rayingecho@gmail.com>

aylei · 2026-01-23T16:43:28Z

/smoke-test -k test_kubernetes_pod_config_sidecar --kubernetes

php-workx · 2026-01-23T18:08:24Z

Oh, seems like there was quite a bit of work left?! Thanks so much for pushing this forward, @aylei ! 🙏🏻

aylei · 2026-01-24T00:49:47Z

/smoke-test --kubernetes

aylei · 2026-01-24T00:49:54Z

/quicktest-core

aylei

Thanks @php-workx !

php-workx added 5 commits December 26, 2025 08:58

test(k8s): cover container-scoped kubectl exec and rsync

f862038

feat(k8s): support multi-container (sidecar) SkyPilot pods

96e7557

fix(k8s): force rsync remote path resolution

7e84ea7

revert(k8s): drop rsync --rsync-path workaround

dbfa5b2

refactor(k8s): centralize ray-node container name

655c73f

gemini-code-assist Bot reviewed Dec 27, 2025

View reviewed changes

Comment thread sky/provision/kubernetes/instance.py Outdated

php-workx and others added 3 commits December 27, 2025 07:35

refactor(k8s): simplify CPU request lookup

6a06c1c

Merge branch 'master' into upstream/k8s-multi-container-pods

1efa683

fix: formatting issue after merge

bc26bb0

Michaelvll requested a review from aylei January 15, 2026 22:25

aylei reviewed Jan 16, 2026

View reviewed changes

Comment thread sky/provision/kubernetes/instance.py Outdated

php-workx and others added 3 commits January 18, 2026 14:02

chore: remove optional CPU default

0a7109e

chore: smoke test with sidecar container

e915cdc

Merge master

dc97013

Signed-off-by: Aylei <rayingecho@gmail.com>

aylei force-pushed the upstream/k8s-multi-container-pods branch 2 times, most recently from 33844cb to 441f613 Compare January 23, 2026 15:38

Fix sidecar overlay and fix smoke test

781ef64

Signed-off-by: Aylei <rayingecho@gmail.com>

aylei force-pushed the upstream/k8s-multi-container-pods branch from 441f613 to 781ef64 Compare January 23, 2026 16:12

Fix backward compatibility

267e3f6

Signed-off-by: Aylei <rayingecho@gmail.com>

aylei approved these changes Jan 24, 2026

View reviewed changes

aylei merged commit f0c38c4 into skypilot-org:master Jan 24, 2026
23 checks passed

kyuds mentioned this pull request Jan 24, 2026

[Kubernetes] Rename k8s container name to skypilot-node #8673

Closed

7 tasks

Uh oh!

Conversation

php-workx commented Dec 27, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

gemini-code-assist Bot commented Dec 27, 2025

Summary of Changes

Highlights

Footnotes

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

Uh oh!

aylei commented Jan 16, 2026

Uh oh!

aylei left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

aylei commented Jan 23, 2026

Uh oh!

aylei commented Jan 23, 2026

Uh oh!

aylei commented Jan 23, 2026

Uh oh!

aylei commented Jan 23, 2026

Uh oh!

aylei commented Jan 23, 2026

Uh oh!

php-workx commented Jan 23, 2026

Uh oh!

aylei commented Jan 24, 2026

Uh oh!

aylei commented Jan 24, 2026

Uh oh!

aylei left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

php-workx commented Dec 27, 2025 •

edited

Loading