fix(eks): repair the EKS create flow — pinned modules, working CNI/CSI, honest errors and status - #289
Merged
Merged
Conversation
…surface real terraform errors
The root cause of three broken-create failure modes was one floating module
range: '~> 21.0' resolved to v21.24.1, whose changed defaults broke create
in three independent ways.
Template:
- pin terraform-aws-modules/eks to 21.24.1 and /vpc to 6.6.1 exactly; guard
test rejects ranges
- default kubernetes_version to a concrete 1.33 — module v21 cannot plan with
a null version (Invalid count argument on the documented no-flag create)
- declare vpc-cni (before_compute), kube-proxy, coredns: module v21 sets
bootstrap_self_managed_addons=false, so an undeclared CNI meant permanently
NotReady nodes
- set IMDS hop limit 2 on the node group: the v21 default of 1 cut the
(non-hostNetwork) EBS CSI controller off from the node role
- enable the CSI addon's defaultStorageClass: since EKS 1.30 AWS ships no
default class, so every PVC stayed Pending
- default instance type m7i-flex.large (Free-Tier-eligible drop-in for
m6i.large): new AWS accounts refuse non-eligible types after a ~30-minute
opaque ASG wait
Error surfacing:
- splitCause kept only the deepest error ('exit status 1') and discarded
the terraform stderr appended after it — the actual failure reason now
reaches the default, non-verbose panel
- verbose mode no longer tees terraform's machine-readable output (the
641 KB 'show -json' line) into the terminal; init/plan human output still
streams via a selective wrapper
Commands and docs:
- app status/access/uninstall now reject a stray positional cluster name
instead of silently running against the current kube-context
- --ref examples match the chart repo's real tag scheme; machine-type
defaults and the EKS out-of-the-box description updated
A typo'd --ref used to surface only at clone time — AFTER certificates were refreshed, the argo-cd helm release deployed and the API port awaited — leaving the cluster mutated with ArgoCD and no applications. - git.Repository.ValidateRef: one ls-remote round-trip (go-git remote list, same transport and auth as the clone) checks the ref exists as a branch or tag before anything touches the cluster - Installer runs the preflight first when app-of-apps is configured; the production validator is the git repository provider, injected through a ChartService seam so orchestration tests stay off the network - BranchNotFoundError now carries the refs the repository DOES offer; the error panel lists branches in full and the version-sorted top tags (1.0.9 < 1.0.48 — plain string sort would invert them), capped so hundreds of release tags cannot bury the branches
… list drifting from reality The creation box printed a hardcoded green 'Ready' and the configured node count on the provisioner's exit code alone; it would render identically over NotReady nodes or a cluster that cannot bind a single PVC. The box now verifies via client-go (brief poll, 45s cap): every node Ready and a default StorageClass present render 'Ready' with the observed fraction; anything else renders 'Provisioned — n/m nodes Ready' (title downgraded to⚠️ ) with a warning naming the specific gap; an unreachable API says 'health not verified' instead of guessing. The list's CONTEXT column was fabricated from the cluster name even when no kubeconfig entry existed (plan-stage failure). It is now derived from the kubeconfig via tfengine.KubeconfigHasContext. A resumed create also flips the record to 'creating' so list stops reporting the previous attempt's 'Failed' while an apply is running. On the transition to Ready the record's CreatedAt is stamped fresh, so a resumed create no longer reports the first failed attempt's timestamp forever.
|
Warning Review limit reached
Next review available in: 47 minutes You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository. How can I continue?After more reviews become available, a review can be triggered using the To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews. How do review limits work?CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability. For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window. Please refer docs for additional details. Review details⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Plus Run ID: 📒 Files selected for processing (2)
📝 WalkthroughWalkthroughChangesCluster provisioning and CLI behavior
Estimated code review effort: 5 (Critical) | ~90 minutes Sequence Diagram(s)sequenceDiagram
participant User
participant Installer
participant GitRepository
participant Cluster
User->>Installer: Start app installation
Installer->>GitRepository: Validate configured branch or tag
GitRepository-->>Installer: Return validation result
Installer->>Cluster: Install only when the reference is valid
sequenceDiagram
participant ClusterService
participant KubernetesAPI
participant ClusterSummary
ClusterService->>KubernetesAPI: Check nodes and default StorageClass
KubernetesAPI-->>ClusterService: Return observed cluster health
ClusterService->>ClusterSummary: Render verified status and warnings
Possibly related PRs
Suggested reviewers: 🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✨ Finishing Touches 💡 1📝 Generate docstrings 💡
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. 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: 4
🤖 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 `@cmd/app/app.go`:
- Around line 15-20: Update the app help example near noPositionalArgs to
replace the invalid positional-cluster form of the install command with the
--context option, while preserving the example’s intended targeting behavior.
In `@internal/cluster/providers/eks/template_guard_test.go`:
- Around line 55-60: Update the guard test around the add-on assertions to match
each add-on map assignment rather than merely searching the full template
source, so comments cannot satisfy the check. Require before_compute = true
specifically within the vpc-cni entry, while preserving validation that vpc-cni,
kube-proxy, and coredns are declared.
In `@internal/cluster/providers/terraform/engine_test.go`:
- Around line 241-244: Update the fixture setup around FindTerraform in the test
to detect runtime.GOOS == "windows" and create an additional executable
terraform.exe in binDir, while preserving the existing terraform fixture and
permissions for all platforms.
In `@internal/cluster/providers/terraform/kubecontext.go`:
- Around line 21-22: Update infoFor to check kubeconfig contexts using the same
candidates as EKS discovery: try rec.Name first, then the cluster ARN. Set
Context to the first matching candidate instead of only returning whether
rec.Name exists.
🪄 Autofix
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: ee445a6f-dd71-487c-8de3-16585aaf02d0
📒 Files selected for processing (31)
cmd/app/.install.mdcmd/app/access.gocmd/app/app.gocmd/app/install.gocmd/app/status.gocmd/app/status_access_test.gocmd/app/uninstall.godocs/getting-started/cloud-clusters.mddocs/reference/architecture/overview.mdinternal/chart/providers/git/repository.gointernal/chart/providers/git/repository_test.gointernal/chart/services/chart_service.gointernal/chart/services/install_orchestration_test.gointernal/chart/services/installer.gointernal/chart/utils/types/interfaces.gointernal/cluster/health_test.gointernal/cluster/models/flags.gointernal/cluster/providers/eks/provider.gointernal/cluster/providers/eks/provider_test.gointernal/cluster/providers/eks/template.gointernal/cluster/providers/eks/template_guard_test.gointernal/cluster/providers/eks/templates/main.tfinternal/cluster/providers/gke/provider.gointernal/cluster/providers/terraform/engine.gointernal/cluster/providers/terraform/engine_test.gointernal/cluster/providers/terraform/kubecontext.gointernal/cluster/providers/terraform/kubecontext_test.gointernal/cluster/service.gointernal/cluster/ui/wizard.gointernal/shared/errors/errors.gointernal/shared/errors/errors_test.go
…, context matching - engine.go: drop redundant .Terraform selectors on promoted methods (staticcheck QF1008); the Init/Plan overrides keep the explicit selector, which is what prevents self-recursion - template_guard_test: bind the addon guard to the actual map entries via regex — the addon names also appear in comments, which could satisfy a plain Contains; before_compute is now required inside the vpc-cni block specifically - engine_test: create a terraform.exe fixture on Windows — LookPath there resolves only PATHEXT extensions, so the bare stub is invisible - cluster list context: beyond the plain cluster name, recognize the conventional context names (EKS: the ARN 'aws eks update-kubeconfig' writes, matched by prefix/suffix since the record has no account id; GKE: gcloud's gke_<project>_<location>_<name>) — the same candidate shapes discovery matches
ivan-flamingo
previously approved these changes
Aug 6, 2026
The Windows unit-test leg dies at random with 'Exception 0xc0000005 / unknown pc' — a Go 1.26.0/1.26.1 runtime bug corrupting return addresses on the stack during GC (suspected Green Tea GC; golang/go#77975, release-blocking, fix targeted at a patch release). The argocd test binary happened to be the victim this run; any package can be next. Our code is not implicated: the suite is race-clean under -race and the crash is inside the runtime during test cleanup. Disable the new GC via GOEXPERIMENT=nogreenteagc on the Windows leg only; drop this once a fixed Go patch release is picked up.
…d in 1.26.2 golang/go#77975 (return addresses corrupted during GC stack scanning, 'Exception 0xc0000005 / unknown pc' at random on windows-amd64) was fixed in Go 1.26.2 (backport golang/go#78041, closed). CI and goreleaser both resolve the toolchain from go.mod via go-version-file, so the toolchain directive is the single switch that moves unit tests, e2e AND the shipped Windows release binaries off the buggy 1.26.0 runtime. Drop the GOEXPERIMENT=nogreenteagc workaround from the Windows unit-test leg — with a fixed toolchain it only kept a non-default GC configuration alive in CI.
There was a problem hiding this comment.
Actionable comments posted: 1
🤖 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 `@internal/cluster/providers/gke/provider.go`:
- Around line 504-506: Update the GKE context predicate passed to
tfengine.KubeconfigContextMatching so it also requires the location segment to
equal rec.Region or begin with rec.Region followed by a hyphen, while preserving
the existing project/name prefix and suffix checks. Add a regression case
covering same-project, same-name contexts in different regions and verify only
the recorded region matches.
🪄 Autofix
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: a459a59c-b30b-4625-9dee-0e450449f1b9
⛔ Files ignored due to path filters (1)
go.sumis excluded by!**/*.sum
📒 Files selected for processing (10)
go.modinternal/cluster/models/flags.gointernal/cluster/providers/eks/provider.gointernal/cluster/providers/eks/provider_test.gointernal/cluster/providers/eks/template_guard_test.gointernal/cluster/providers/gke/provider.gointernal/cluster/providers/gke/provider_test.gointernal/cluster/providers/terraform/engine.gointernal/cluster/providers/terraform/engine_test.gointernal/cluster/providers/terraform/kubecontext.go
🚧 Files skipped from review as they are similar to previous changes (5)
- internal/cluster/models/flags.go
- internal/cluster/providers/eks/provider.go
- internal/cluster/providers/terraform/engine_test.go
- internal/cluster/providers/eks/template_guard_test.go
- internal/cluster/providers/terraform/engine.go
The gke_<project>_<location>_<name> match checked only project and name: a same-name, same-project cluster in ANOTHER region satisfied the prefix/suffix too, and the lexically first kubeconfig context won regardless of the record. The location segment must now be the recorded region itself (regional cluster) or a zone inside it (region + "-" prefix, e.g. us-central1-a); us-east10 does not false-match us-east1 since its next rune is not a hyphen.
ivan-flamingo
approved these changes
Aug 6, 2026
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
2 participants
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 by CodeRabbit
New Features
Bug Fixes
--context.Documentation