feat(helm)!: Kata Containers as the DinD runtime — drop Sysbox (closes #116)#129
Merged
Conversation
Sysbox no longer works on containerd 2.x (RKE2/k3s/kubeadm >= 1.34) — its public releases are frozen on containerd 1.x and there is no upstream fix. This adds Kata Containers (kata-qemu) as a supported DinD runtime. Chart changes (all gated — Sysbox installs render unchanged): - dind.privileged: explicit override of the runtime/privileged coupling. null default = legacy auto-derivation. Kata needs true (dockerd needs CAP_NET_ADMIN/RAW for iptables NAT; safe — caps confined to the microVM). - dind.kataInit: chart-templated entrypoint wrapper (ConfigMap, not a custom image) — installs fuse-overlayfs, mknod /dev/fuse, formats+mounts the Block PVC, runs the cgroup-v2 PID-1 evacuation shim. - persistence.varLibDocker.persistentVolume: optional Block-mode PVC for /var/lib/docker — required under Kata for xattr-dependent workloads (virtio-fs drops security.capability xattrs, CVE-2021-20263). - values.schema.json, NOTES.txt warnings, examples/helm/kata/values.yaml. - CI: helm.yml gains a Kata render matrix + a backward-compat regression guard asserting the Sysbox path picks up no Kata machinery. Closes #116 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- docs/kata-runtime.md: install kata-deploy, configure the chart, what the kataInit wrapper does, verification checklist, troubleshooting table, and a Sysbox vs Kata vs runc+privileged tradeoffs comparison. - ADR-0011: records making Kata first-class now (deviates from ADR-0004's Phase-9 timeline, forced by the Sysbox EOL). - kubernetes.md / chart README / DOCKER.md / README.md: present Sysbox and Kata as the two supported runtimes instead of Sysbox-only. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Plus Run ID: 📒 Files selected for processing (2)
📝 WalkthroughWalkthroughThis PR migrates the Helm chart from Sysbox to Kata Containers ( ChangesKata Containers as First-Class DinD Runtime
Estimated code review effort🎯 4 (Complex) | ⏱️ ~60 minutes Possibly related PRs
Poem
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 2
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
.github/workflows/helm.yml (1)
32-35:⚠️ Potential issue | 🟠 Major | ⚡ Quick winSet explicit least-privilege permissions for the
lintjob.
lintcurrently inherits default token permissions. Please pin this job to read-only permissions to avoid unnecessary write scope.Suggested patch
jobs: lint: name: helm lint + template runs-on: ubuntu-latest + permissions: + contents: read steps: - uses: actions/checkout@v6🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In @.github/workflows/helm.yml around lines 32 - 35, The lint job currently inherits default token permissions; to enforce least-privilege add an explicit permissions block under the lint job in the Helm workflow (job name "lint") that restricts the GITHUB_TOKEN to read-only, e.g. add permissions: contents: read (or the minimal read scopes you require) directly beneath the "lint:" stanza so the job cannot perform write actions.
🧹 Nitpick comments (3)
examples/helm/kata/values.yaml (1)
21-23: ⚡ Quick winConsider emphasizing placeholder replacement.
The
mcpApiKeyplaceholder "change-me-32-hex-chars" is clearly marked, but users might overlook it. While line 19 mentions replacing placeholders, consider adding a more prominent comment near the value itself, such as:# ⚠️ REPLACE THIS - Generate a real key: openssl rand -hex 32 mcpApiKey: "change-me-32-hex-chars"This reduces the risk of deploying with an obvious/weak key.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@examples/helm/kata/values.yaml` around lines 21 - 23, Add a prominent inline warning comment directly above the secrets.mcpApiKey value to make replacement explicit (e.g., "⚠️ REPLACE THIS - Generate a real key: openssl rand -hex 32") so users won't miss the placeholder; update the block around the secrets.create and mcpApiKey entries to include that comment and consider leaving the example value unchanged but clearly marked as a placeholder using the mcpApiKey symbol for easy location.helm/computer-use-server/README.md (1)
167-175: ⚡ Quick winConsider clarifying the auto-derivation logic.
The explanation "empty ⇒
true, set ⇒false" describes the legacy Sysbox behavior, but the flow could be clearer for users configuring Kata. Consider restructuring to explain the logic more sequentially: "Whendind.privileged: null, the chart setsprivileged: falsefor any non-emptyruntimeClassName(Sysbox path) andprivileged: truewhenruntimeClassNameis empty (stock runc fallback). For Kata, you must override withdind.privileged: trueexplicitly."🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@helm/computer-use-server/README.md` around lines 167 - 175, Rewrite the paragraph describing how dind.privileged is derived from runtimeClassName to be sequential and unambiguous: state that when dind.privileged is null the chart auto-derives privileged as false if runtimeClassName is non-empty (Sysbox path) and true if runtimeClassName is empty (stock-runc fallback), call out that this is legacy Sysbox behavior, and explicitly instruct users that Kata requires setting dind.privileged: true; also mention that the chart prints a warning in NOTES.txt when runtimeClassName is empty and keep the cautionary sentence about not shipping production clusters with the stock-runc+privileged escape hatch.helm/computer-use-server/templates/pvc-var-lib-docker.yaml (1)
8-9: ⚡ Quick winUse the shared PVC gate here too.
deployment.yamlandconfigmap-dind-init.yamlboth rely oncomputer-use-server.varLibDockerIsPVC, but this template re-derives the condition from raw values. If that helper carries any extra Kata-only guard, this file will still provision a Block PVC that the pod never attaches.♻️ Proposed fix
{{- $pv := .Values.persistence.varLibDocker.persistentVolume | default dict }} -{{- if and $pv.enabled (not $pv.existingClaim) }} +{{- if and (eq (include "computer-use-server.varLibDockerIsPVC" .) "true") (not $pv.existingClaim) }}🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@helm/computer-use-server/templates/pvc-var-lib-docker.yaml` around lines 8 - 9, This template re-derives the PVC condition from .Values.persistence.varLibDocker.persistentVolume ($pv.enabled / $pv.existingClaim) instead of using the shared helper, causing a Block PVC to be created when other templates expect computer-use-server.varLibDockerIsPVC; change the conditional to call and rely on the existing helper (computer-use-server.varLibDockerIsPVC) wherever the template currently checks $pv.enabled and not $pv.existingClaim so the PVC creation logic is consistent with deployment.yaml and configmap-dind-init.yaml.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@docs/kata-runtime.md`:
- Line 39: Update the broken markdown link fragment: replace the incorrect
fragment "`#configure-the-helm-chart`" with the actual generated heading ID
"`#step-2--configure-the-helm-chart`" so the inline link correctly points to the
"Step 2 — Configure the Helm chart" heading in docs/kata-runtime.md.
In `@helm/computer-use-server/values.schema.json`:
- Around line 52-63: Add JSON Schema conditionals to enforce runtime-specific
combos: add an if/then that detects Kata runtime (e.g., runtimeClassName equals
the Kata identifier) and in the then require dind.privileged to be true and
ensure dind.kataInit.enabled may only be present/true under that same Kata
runtime; likewise add an if/then for sysbox-runc (runtimeClassName ==
"sysbox-runc") that rejects
persistence.varLibDocker.persistentVolume.enabled=true and rejects a non-empty
persistence.varLibDocker.existingClaim. Update the top-level schema near the
storageDriver/privileged/kataInit blocks to include these conditional rules so
invalid combinations (dind.privileged=false with Kata, kataInit used without
Kata, or persistentVolume/existingClaim with sysbox) fail schema validation.
---
Outside diff comments:
In @.github/workflows/helm.yml:
- Around line 32-35: The lint job currently inherits default token permissions;
to enforce least-privilege add an explicit permissions block under the lint job
in the Helm workflow (job name "lint") that restricts the GITHUB_TOKEN to
read-only, e.g. add permissions: contents: read (or the minimal read scopes you
require) directly beneath the "lint:" stanza so the job cannot perform write
actions.
---
Nitpick comments:
In `@examples/helm/kata/values.yaml`:
- Around line 21-23: Add a prominent inline warning comment directly above the
secrets.mcpApiKey value to make replacement explicit (e.g., "⚠️ REPLACE THIS -
Generate a real key: openssl rand -hex 32") so users won't miss the placeholder;
update the block around the secrets.create and mcpApiKey entries to include that
comment and consider leaving the example value unchanged but clearly marked as a
placeholder using the mcpApiKey symbol for easy location.
In `@helm/computer-use-server/README.md`:
- Around line 167-175: Rewrite the paragraph describing how dind.privileged is
derived from runtimeClassName to be sequential and unambiguous: state that when
dind.privileged is null the chart auto-derives privileged as false if
runtimeClassName is non-empty (Sysbox path) and true if runtimeClassName is
empty (stock-runc fallback), call out that this is legacy Sysbox behavior, and
explicitly instruct users that Kata requires setting dind.privileged: true; also
mention that the chart prints a warning in NOTES.txt when runtimeClassName is
empty and keep the cautionary sentence about not shipping production clusters
with the stock-runc+privileged escape hatch.
In `@helm/computer-use-server/templates/pvc-var-lib-docker.yaml`:
- Around line 8-9: This template re-derives the PVC condition from
.Values.persistence.varLibDocker.persistentVolume ($pv.enabled /
$pv.existingClaim) instead of using the shared helper, causing a Block PVC to be
created when other templates expect computer-use-server.varLibDockerIsPVC;
change the conditional to call and rely on the existing helper
(computer-use-server.varLibDockerIsPVC) wherever the template currently checks
$pv.enabled and not $pv.existingClaim so the PVC creation logic is consistent
with deployment.yaml and configmap-dind-init.yaml.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro Plus
Run ID: 55872eae-55ee-4231-b11e-4dfa8a5999d9
📒 Files selected for processing (17)
.github/workflows/helm.ymlREADME.mddocs/DOCKER.mddocs/future-architecture/adr/0011-kata-as-first-class-dind-runtime.mddocs/kata-runtime.mddocs/kubernetes.mdexamples/helm/README.mdexamples/helm/kata/values.yamlhelm/computer-use-server/Chart.yamlhelm/computer-use-server/README.mdhelm/computer-use-server/templates/NOTES.txthelm/computer-use-server/templates/_helpers.tplhelm/computer-use-server/templates/configmap-dind-init.yamlhelm/computer-use-server/templates/deployment.yamlhelm/computer-use-server/templates/pvc-var-lib-docker.yamlhelm/computer-use-server/values.schema.jsonhelm/computer-use-server/values.yaml
Sysbox does not work on containerd 2.x and has no upstream fix. Rather than carry it as a second, broken-on-modern-clusters option, the chart now targets Kata Containers exclusively. BREAKING CHANGE: the chart defaults changed. Existing Sysbox installs must migrate to Kata (install kata-deploy; see docs/kata-runtime.md). Flipped defaults: - orchestrator.runtimeClassName: sysbox-runc -> kata-qemu - dind.privileged: null -> true - dind.storageDriver: overlay2 -> fuse-overlayfs - dind.kataInit.enabled: false -> true - persistence.varLibDocker.persistentVolume.enabled: false -> true `runtimeClassName: ""` (stock runc + privileged) is retained as an explicitly unsupported local-testing escape hatch, with a loud NOTES.txt warning. All sysbox references removed from the chart, templates, schema, NOTES, and CI. The helm.yml workflow now asserts the default render is the Kata path and checks the runc fallback opts out of every Kata knob. Chart version 0.2.0 -> 0.3.0. Closes #116 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Rewrites kubernetes.md, the chart README, kata-runtime.md, and the helm examples to present Kata Containers as the single supported DinD runtime instead of one of two options. - kubernetes.md: remove the containerd-1.x/2.x runtime-choice table; Kata is the only path. - chart README: collapse "Runtime selection" to Kata + the unsupported runc fallback. - kata-runtime.md: Sysbox kept only as a historical comparison row in the tradeoffs table, not as a chart option. - examples/helm/kata/ removed — its config is now the chart default, so examples/helm/standalone/ covers it. future-architecture/ docs are intentionally left as-is (historical context; ADR-0011 already records the move to Kata). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Yambr
changed the title
There was a problem hiding this comment.
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
helm/computer-use-server/templates/deployment.yaml (1)
208-223:⚠️ Potential issue | 🟡 Minor | ⚡ Quick winAdd schema validation to prevent Block PVC without kataInit.
The block-mode PVC logic depends on
kataInit.enabledbeing true: when enabled, the init wrapper formats the block device and mounts it at/var/lib/docker; when disabled, the block device is exposed but unmounted, causingdockerdto fail.While
kataInit.enableddefaults to true invalues.yaml(mitigating the issue in typical deployments), the schema allows independently enablingpersistence.varLibDocker.persistentVolumewithout enforcingkataInit.enabled: true. Add a schema validation to requirekataInit.enabled: truewhen the Block PVC is enabled, or update thevarLibDockerIsPVChelper to guard against this misconfiguration.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@helm/computer-use-server/templates/deployment.yaml` around lines 208 - 223, The deployment's block PVC path can be enabled via persistence.varLibDocker.persistentVolume while kataInit.enabled may be false, causing dockerd to fail; update validation or the helper so this misconfiguration is prevented: either add a chart schema rule that requires kataInit.enabled: true whenever persistence.varLibDocker.persistentVolume (or the varLibDocker block-mode condition used by the varLibDockerIsPVC helper) is true, or modify the varLibDockerIsPVC template helper to also check .Values.kataInit.enabled and return "false" unless both the PVC is block-mode and kataInit.enabled is true, ensuring the block device is not exposed without the init wrapper.
🧹 Nitpick comments (1)
helm/computer-use-server/templates/_helpers.tpl (1)
89-100: 💤 Low valueAuto-derive fallback for
nullreturnsfalsefor Kata, contradicting documented requirements.When
dind.privilegedisnullandruntimeClassNameis set (e.g.,kata-qemu), this helper returnsfalse. However, the values.yaml documentation states Kata requiresprivileged=truefor iptables NAT.This is currently safe because
values.yamldefaultsprivilegedtotrue(so the null branch is never hit with Kata defaults), but the auto-derive logic is misleading and could cause failures if someone explicitly setsprivileged: nullwith Kata.Consider inverting the fallback or documenting that
nullis only intended for the runc fallback path.Option: clarify the comment to match behavior
{{/* Whether the inner dind container must run privileged. Resolution order: 1. dind.privileged explicitly set (true/false) => use it verbatim. The Kata default is `true`: dockerd requires CAP_NET_ADMIN/RAW for iptables NAT, and the caps stay confined to the microVM. 2. dind.privileged is null => auto-derive from runtimeClassName: - runtimeClassName empty => true (runc fallback — dockerd needs it) - runtimeClassName set => false + runtimeClassName empty => true (runc fallback — dockerd needs privileged) + runtimeClassName set => false (legacy Sysbox behavior; Kata users must set explicit true) */}}🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@helm/computer-use-server/templates/_helpers.tpl` around lines 89 - 100, The helper computer-use-server.dindPrivileged currently auto-derives .Values.dind.privileged as false when .Values.dind.privileged is null and .Values.orchestrator.runtimeClassName is set, which breaks Kata (it requires privileged=true); update the helper so the null/default branch returns true (i.e., when .Values.dind.privileged is not set, default privileged to true regardless of runtimeClassName) by changing the conditional branches in the computer-use-server.dindPrivileged template to return true instead of false for the currently falling-through case that uses .Values.orchestrator.runtimeClassName.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Outside diff comments:
In `@helm/computer-use-server/templates/deployment.yaml`:
- Around line 208-223: The deployment's block PVC path can be enabled via
persistence.varLibDocker.persistentVolume while kataInit.enabled may be false,
causing dockerd to fail; update validation or the helper so this
misconfiguration is prevented: either add a chart schema rule that requires
kataInit.enabled: true whenever persistence.varLibDocker.persistentVolume (or
the varLibDocker block-mode condition used by the varLibDockerIsPVC helper) is
true, or modify the varLibDockerIsPVC template helper to also check
.Values.kataInit.enabled and return "false" unless both the PVC is block-mode
and kataInit.enabled is true, ensuring the block device is not exposed without
the init wrapper.
---
Nitpick comments:
In `@helm/computer-use-server/templates/_helpers.tpl`:
- Around line 89-100: The helper computer-use-server.dindPrivileged currently
auto-derives .Values.dind.privileged as false when .Values.dind.privileged is
null and .Values.orchestrator.runtimeClassName is set, which breaks Kata (it
requires privileged=true); update the helper so the null/default branch returns
true (i.e., when .Values.dind.privileged is not set, default privileged to true
regardless of runtimeClassName) by changing the conditional branches in the
computer-use-server.dindPrivileged template to return true instead of false for
the currently falling-through case that uses
.Values.orchestrator.runtimeClassName.
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro Plus
Run ID: 3dd79e2c-f54e-4b47-a2b3-d5fc793a34cc
📒 Files selected for processing (15)
.github/workflows/helm.yml.github/workflows/release.ymldocs/kata-runtime.mddocs/kubernetes.mdexamples/helm/README.mdexamples/helm/standalone/values.yamlexamples/helm/with-open-webui/README.mdhelm/computer-use-server/Chart.yamlhelm/computer-use-server/README.mdhelm/computer-use-server/templates/NOTES.txthelm/computer-use-server/templates/_helpers.tplhelm/computer-use-server/templates/deployment.yamlhelm/computer-use-server/templates/pvc-var-lib-docker.yamlhelm/computer-use-server/values.schema.jsonhelm/computer-use-server/values.yaml
✅ Files skipped from review due to trivial changes (3)
- .github/workflows/release.yml
- examples/helm/standalone/values.yaml
- docs/kata-runtime.md
…re not unsupported - Remove the last Sysbox references from the chart README and kata-runtime.md (the "unlike Sysbox" aside and the Sysbox column in the tradeoffs table). - The runc + privileged fallback is functional — it is not "unsupported", it is insecure (the inner daemon shares the host kernel). Reword values.yaml, README, NOTES.txt, schema, and kata-runtime.md accordingly so the wording describes the security tradeoff, not a non-working path. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
🧹 Nitpick comments (1)
helm/computer-use-server/values.yaml (1)
133-147: 💤 Low valueClarify the
nullauto-derivation comment — the described fallback contradicts Kata requirements.The comment on lines 145-146 states that
nullauto-derives tofalsewhenruntimeClassNameis set (i.e., Kata). However, lines 140-143 explicitly state thatprivileged: trueis required for Kata. If someone setsprivileged: nullexpecting auto-derivation to "do the right thing" for Kata, they'll getfalseand hit the iptables permission error.Consider rewording to make it clear that
nullauto-derivation is a legacy escape hatch that only works correctly for the runc fallback, not for Kata:Suggested clarification
# null => auto-derive from orchestrator.runtimeClassName (empty => true, -# set => false). Only useful for the runc fallback. +# set => false). WARNING: this derivation does NOT work for Kata — Kata +# needs true. Only use null with the runc fallback (empty runtimeClassName).🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@helm/computer-use-server/values.yaml` around lines 133 - 147, Update the comment for the privileged setting to clarify that null auto-derivation (based on orchestrator.runtimeClassName) is a legacy escape hatch intended only for the runc fallback and will not produce a safe value for Kata: explicitly state that for Kata users privileged must be true (required for dockerd to have CAP_NET_ADMIN/CAP_NET_RAW) and that setting privileged: null with a Kata runtimeClassName will result in false and cause iptables permission failures; reference the storageDriver and privileged keys and mention Kata/runc fallback so readers know when to avoid relying on null auto-derivation.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Nitpick comments:
In `@helm/computer-use-server/values.yaml`:
- Around line 133-147: Update the comment for the privileged setting to clarify
that null auto-derivation (based on orchestrator.runtimeClassName) is a legacy
escape hatch intended only for the runc fallback and will not produce a safe
value for Kata: explicitly state that for Kata users privileged must be true
(required for dockerd to have CAP_NET_ADMIN/CAP_NET_RAW) and that setting
privileged: null with a Kata runtimeClassName will result in false and cause
iptables permission failures; reference the storageDriver and privileged keys
and mention Kata/runc fallback so readers know when to avoid relying on null
auto-derivation.
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro Plus
Run ID: 9c22e46d-baad-4b1f-9629-2dd4b686777b
📒 Files selected for processing (5)
docs/kata-runtime.mdhelm/computer-use-server/README.mdhelm/computer-use-server/templates/NOTES.txthelm/computer-use-server/values.schema.jsonhelm/computer-use-server/values.yaml
✅ Files skipped from review due to trivial changes (2)
- helm/computer-use-server/README.md
- docs/kata-runtime.md
- values.schema.json: add an if/then guard so a Kata runtimeClassName (^kata) requires dind.privileged != false and dind.kataInit.enabled != false. Invalid Kata combos now fail schema validation instead of only failing later at render/runtime. The runc fallback (empty runtimeClassName) is unaffected. CodeRabbit's companion sysbox-runc guard is not applicable — Sysbox was removed in this PR. - docs/kata-runtime.md: fix the broken link fragment (#configure-the-helm-chart -> #step-2--configure-the-helm-chart), flagged by markdownlint MD051. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Collaborator
Author
|
Addressed both CodeRabbit findings in 7a3b30d:
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Makes Kata Containers (
kata-qemu) the single supported DinD runtime for thecomputer-use-serverHelm chart, and removes Sysbox entirely. Closes #116.Why: Sysbox no longer works on containerd 2.x (RKE2 / k3s / kubeadm ≥ 1.34). Docker acquired Nestybox; public Sysbox releases are frozen on containerd 1.x with no upstream fix. Rather than carry Sysbox as a second option that is broken on every modern cluster, the chart now targets Kata exclusively. This upstreams the configuration we run in production on a containerd 2.x cluster.
Changes
Chart — defaults flipped to Kata:
orchestrator.runtimeClassNamesysbox-runckata-qemudind.privilegednulltruedind.storageDriveroverlay2fuse-overlayfsdind.kataInit.enabledfalsetruepersistence.varLibDocker.persistentVolume.enabledfalsetruedind.kataInit— chart-templated entrypoint wrapper (ConfigMap, not a custom image): installsfuse-overlayfs, creates/dev/fuse, formats+mounts the Block PVC, runs the cgroup-v2 PID-1 evacuation shim./var/lib/dockeris a Block-mode PVC — required under Kata for xattr-dependent workloads (virtio-fs dropssecurity.capabilityxattrs, CVE-2021-20263).runtimeClassName: ""(stock runc + privileged) is retained as an explicitly unsupported local-testing escape hatch with a loudNOTES.txtwarning.0.1.0→0.3.0.helm.yml): the default render is asserted to be the Kata path; the runc fallback is asserted to opt out of every Kata knob.Docs:
docs/kata-runtime.md— install kata-deploy → configure → verify → troubleshoot, plus a tradeoffs table (Sysbox kept only as a historical comparison row).docs/future-architecture/adr/0011-kata-as-first-class-dind-runtime.md— ADR.kubernetes.md/ chart README / examples — Kata is the only documented runtime;examples/helm/kata/removed (it is now the chart default, covered bystandalone/).docs/future-architecture/left as-is (historical context; ADR-0011 records the move).Testing
helm lint— default (Kata),standalone,with-open-webuiall passhelm template— default render assertsruntimeClassName: kata-qemu,privileged: true, the init wrapper,volumeMode: Block,fuse-overlayfs; the runc fallback opts out of all of itkubeconform -strict— 21/21 resources valid (default + runc-fallback renders)shellcheckon the rendereddind-entrypoint.sh— cleantests/test-no-cyrillic.sh,tests/test-project-structure.shpassNot exercisable in CI: the runtime behavior inside a Kata guest (dockerd start with
fuse-overlayfs,mknod/mkfs/cgroup-evacuation, xattr preservation, the iptables-needs-privileged claim) requires a Kata-capable cluster. These are validated in a production deployment; CI here covers chart rendering only.ADRs reviewed
🤖 Generated with Claude Code
Summary by CodeRabbit
New Features
Documentation
Configuration Changes