Updates deploy-e2e-headlamp.sh to reference the consolidated RBAC manifest
at privilegedescalation/infra/base/rbac/e2e-ci-runner.yaml instead of the
non-existent local path deployment/e2e-ci-runner-rbac.yaml.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
The E2E workflow and deploy scripts were targeting the legacy
privilegedescalation-dev namespace, which is not managed by Flux GitOps
in privilegedescalation/infra.
The infra repo (PR #11) already provisions the headlamp-dev namespace
and corresponding RBAC (e2e-ci-runner-headlamp-rbac.yaml) that grants
the ARC runner SA (runners-privilegedescalation-gha-rs-no-permission in
arc-runners) the permissions needed to deploy/teardown the E2E
Headlamp instance.
This change aligns all E2E infrastructure to use headlamp-dev:
- .github/workflows/e2e.yaml: E2E_NAMESPACE=headlamp-dev
- scripts/deploy-e2e-headlamp.sh: default namespace and comments
- scripts/teardown-e2e-headlamp.sh: default namespace
- deployment/e2e-ci-runner-rbac.yaml: namespace and add missing events
permission (already present in infra copy)
Refs: PRI-423
Co-authored-by: Chris Farhood <chris@farhood.org>
Co-authored-by: Paperclip <noreply@paperclip.ing>
The :latest tag caused E2E flakiness when a newer Headlamp image was
pulled on some cluster nodes (IfNotPresent policy) but not others.
Concurrent E2E runs on main saw different image versions, and the newest
:latest (sha256:89c6c65) failed to pass the readiness probe within 120s.
Pin to v0.40.1 — the same version running in production (kube-system) —
so all nodes use the same cached digest and CI is deterministic. Update
this pin when Headlamp is upgraded in production.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
kubectl apply without prior deletion patches in place: if the pod spec is
unchanged between runs, no rollout is triggered and a potentially degraded
pod from a prior run keeps serving. This caused the auth.setup.ts timeout
(waiting for the "use a token" button) even when no concurrent runs were
present — the headlamp-e2e pod was in an inconsistent state from a previous
run that didn't tear down cleanly.
Changes:
- deploy-e2e-headlamp.sh: delete Deployment, Service, and ServiceAccount
(with --wait) before applying, guaranteeing a fresh pod each run
- auth.setup.ts: add explicit waitFor({ state: 'visible', timeout: 15_000 })
before the "use a token" button click, so failures surface at 15 s with a
clear locator error rather than silently timing out at 60 s
Fixes the pre-existing infra issue blocking PR#110.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
kubectl rollout status confirms the pod is ready per readinessProbe, but
Kubernetes Service DNS propagation to the runner pod may lag behind.
This caused intermittent E2E failures with ERR_NAME_NOT_RESOLVED.
Add a poll loop (max 120s) after rollout status that verifies the service
URL is reachable via HTTP before writing .env.e2e. This eliminates the
race condition between DNS propagation and Playwright launch.
Fixes: PRI-687 (intermittent E2E DNS failure)
The Helm chart deployment was consistently failing — the pod enters
CrashLoopBackOff despite identical kubectl manifests working. The Helm
chart also silently ignored extraVolumes/extraVolumeMounts (pnpm-style
keys not supported by the chart), meaning the plugin ConfigMap was
never actually mounted even when deploy appeared to succeed.
Replace with direct kubectl apply using a bash heredoc to render the
manifest with shell variable substitution. This removes the Helm
dependency, fixes the plugin volume mount, and uses the exact
configuration that was proven to work in the cluster.
Also adds explicit initialDelaySeconds/failureThreshold on readiness
and liveness probes to give Headlamp adequate startup time.
Note: .github/workflows/e2e.yaml still has a Setup Helm step that is
now unused — assigned to Hugh Hackman to remove.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Per org RBAC policy, development/testing Headlamp instances must run in
`privilegedescalation-dev`, not `default`. Agents only have read-write
access in `privilegedescalation` and `privilegedescalation-dev` — the
`default` namespace is outside our permitted scope.
Updated:
- deployment/e2e-ci-runner-rbac.yaml: Role/RoleBinding now targets privilegedescalation-dev
- deployment/headlamp-e2e-values.yaml: comment updated
- scripts/deploy-e2e-headlamp.sh: default namespace changed
- scripts/teardown-e2e-headlamp.sh: default namespace changed
Note: .github/workflows/e2e.yaml still sets E2E_NAMESPACE: default and
needs a separate update — delegated to Hugh Hackman (workflow owner).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Fails fast with a clear error and remediation hint if the runner SA
lacks configmap delete permission, instead of dying mid-deploy.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Board directive: E2E tests must run in the `default` namespace.
Nothing should persist beyond a test run; no dedicated namespace needed.
Changes:
- e2e-ci-runner-rbac.yaml: retarget Role/RoleBinding to `default`,
remove ClusterRole/ClusterRoleBinding (no longer needed since we
don't need cluster-scoped namespace read permission)
- e2e.yaml: set E2E_NAMESPACE=default
- deploy-e2e-headlamp.sh: default namespace to `default`, remove
namespace existence check (default always exists)
- teardown-e2e-headlamp.sh: default namespace to `default`, remove
namespace existence check guard
- headlamp-e2e-values.yaml: update usage comment
- e2e/README.md: remove namespace creation prerequisite
Closes#78#79
Co-Authored-By: Paperclip <noreply@paperclip.ing>
When the headlamp-e2e namespace does not exist, teardown now exits
early with a clear message instead of failing with a misleading RBAC
error. Addresses PRI-443.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
The CI runner SA only has namespace-scoped RBAC in headlamp-e2e — it
cannot create or delete namespaces at the cluster level. Deploy now
verifies the namespace exists (with a clear error if not), and teardown
cleans up resources without deleting the namespace itself.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Delete custom Docker image approach per board directive. Plugin is now
loaded into stock Headlamp via a ConfigMap volume mount:
- Delete Dockerfile.e2e
- deploy-e2e-headlamp.sh creates a ConfigMap from dist/ and mounts it
into the stock ghcr.io/headlamp-k8s/headlamp image
- Helm values use extraVolumes/extraVolumeMounts for the ConfigMap
- No custom images, no PVCs, no kubectl exec/cp
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Replace the PVC + kubectl-patch approach for E2E plugin deployment with
a custom Docker image that has the plugin pre-installed. This eliminates
all policy-violating operations:
- No PVCs in kube-system
- No kubectl exec/cp to Headlamp pods
- No deployment patching via kubectl
- No temporary pods or ConfigMap-based file transfers
The new approach builds a Headlamp image with the plugin baked in
(Dockerfile.e2e), deploys it as a dedicated instance in the headlamp-e2e
namespace via Helm, and tears it down after tests complete.
RBAC is scoped to the headlamp-e2e namespace instead of kube-system.
Note: .github/workflows/e2e.yaml still needs updating to use the new
scripts — that change is delegated to Hugh (CI/CD owner).
Closes: privilegedescalation/headlamp-polaris-plugin#72
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* e2e: shared volume plugin deployment replacing init container approach
Replace the init container plugin installation with a shared PVC volume
between the CI runner and Headlamp pod. The runner builds the plugin and
copies it to the shared mount; Headlamp reads from the same volume.
- Add deployment/headlamp-e2e-values.yaml (PVC-backed shared volume)
- Add deployment/headlamp-plugins-pvc.yaml (PVC manifest)
- Add scripts/deploy-plugin-via-volume.sh (build + copy + restart)
- Remove deployment/headlamp-static-plugin-values.yaml (init container)
This is CI-only test infrastructure — ArtifactHub remains the sole
user-facing distribution channel.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* ci: update e2e workflow for shared volume plugin deployment
Replace the old preflight-only approach with a build-and-deploy flow
that uses a shared volume (hostPath) between the CI runner and the
Headlamp pod. The workflow now builds the plugin from source, copies
the artifact to a shared volume path, and optionally calls Gandalf's
deploy script for Headlamp rollout coordination.
Removes kubectl exec/cp references and version-match preflight in
favor of deploying the PR's actual build artifact.
Refs: PRI-216, PRI-195
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* ci: align e2e workflow with Gandalf's deploy script interface
Simplify deploy step to call scripts/deploy-plugin-via-volume.sh
directly instead of duplicating copy logic. Align env var names
(PLUGIN_VOLUME_PATH, HEADLAMP_DEPLOY) with the deploy script's
expected interface from PR #59.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* fix: deploy plugin via temporary pod instead of assuming local PVC mount
The deploy script assumed the PVC was mounted on the CI runner at
/mnt/headlamp-plugins, but the runner pod doesn't have that mount.
Fix by using a temporary pod (kubectl run) that mounts the PVC,
receives the plugin tarball via stdin, and extracts it.
Also adds missing workflow steps to create the PVC and upgrade
Headlamp with the shared volume helm values before deploying.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* fix: add kubectl, helm, and helm repo setup steps to e2e workflow
The self-hosted runner doesn't have kubectl or helm pre-installed.
Add setup steps using azure/setup-kubectl and azure/setup-helm
actions, and add the Headlamp helm repo before the upgrade step.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* fix: update Headlamp Helm repo URL from headlamp-k8s to kubernetes-sigs
The Headlamp project moved to the kubernetes-sigs org. The old Helm chart
repository URL (headlamp-k8s.github.io) returns 404, causing E2E workflow
failure at the `helm repo add` step.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* chore: add RBAC manifest for E2E CI runner
Documents the Role and RoleBinding applied to the cluster for the ARC
runner service account. Grants permissions in kube-system needed for
shared volume plugin deployment (PVCs, pods, Helm resources).
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* fix: remove .github/workflows/e2e.yaml changes from PR
The workflow changes should be handled separately by Hugh Hackman
per PRI-215. This PR should only contain deployment manifests and
scripts, not CI workflow modifications.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* ci: add shared volume plugin deployment to E2E workflow
Adds the build, Helm, PVC, and plugin deploy steps needed for the
shared volume E2E approach. Uses the correct kubernetes-sigs Helm repo
URL and overrides config.sessionTTL=0 to avoid schema validation error.
This is the workflow counterpart to the deployment manifests and scripts
already in this PR (PVC, values overlay, deploy script).
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* fix(e2e): set sessionTTL=1 to satisfy Helm schema minimum
The Headlamp Helm chart schema enforces a minimum of 1 for
config.sessionTTL. Setting it to 0 caused helm upgrade to fail
with a schema validation error.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* fix(e2e): add cluster-scoped RBAC for CI runner
The Headlamp Helm chart manages ClusterRole and ClusterRoleBinding
resources. The CI runner SA needs cluster-level permissions to
get/update these during helm upgrade. Added ClusterRole and
ClusterRoleBinding alongside the existing namespace-scoped Role.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* fix(e2e): replace helm upgrade with kubectl patch to avoid cluster RBAC
The CI runner SA cannot access cluster-scoped resources (ClusterRole,
ClusterRoleBinding) needed by helm upgrade's 3-way merge. Replace the
helm upgrade step with kubectl patch commands that add the shared volume
mount directly to the Headlamp deployment.
This eliminates the need for cluster-admin intervention:
- kubectl patch adds PVC volume + volumeMount to the deployment
- kubectl set env configures the plugins directory
- kubectl rollout status waits for the update
Also removes the now-unnecessary ClusterRole/ClusterRoleBinding from the
RBAC manifest — only namespace-scoped Role/RoleBinding is needed.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* fix(e2e): improve volume mount idempotency check
Check for existing volume mount by mountPath and PVC claimName, not
just by volume name. A prior helm upgrade may have created mounts
with different names but the same path, causing kubectl patch to fail
with "mountPath must be unique".
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* fix(e2e): schedule deploy pod on same node as Headlamp
The headlamp-plugins PVC is ReadWriteOnce, so the temporary deploy
pod must run on the same node as the Headlamp pod to mount it.
Look up the Headlamp pod's node and set nodeName in the pod spec.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* fix(e2e): use Job with base64 tarball instead of kubectl run stdin
The kubectl run --rm -i stdin pipe times out in the ARC runner
environment. Replace with a Kubernetes Job that receives the plugin
tarball as base64-encoded data in the container command. This avoids
the unreliable attach/stdin mechanism entirely.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* fix(e2e): use ConfigMap for tarball instead of inline base64
Embedding base64 data in the YAML spec broke parsing. Store the plugin
tarball in a ConfigMap via --from-file and mount it in the deploy Job.
This avoids both the stdin pipe issue and the YAML escaping issue.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* fix(e2e): use temp file for Job YAML to avoid heredoc escaping
Variable expansion inside heredocs breaks YAML parsing when values
contain colons and quotes (like nodeName). Write the Job manifest to
a temp file with literal YAML, then sed-substitute the dynamic values.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* fix(e2e): use Pod instead of Job for plugin deploy
The CI runner SA has permission to create Pods but not Jobs in
kube-system. Switch from a Job to a plain Pod with restartPolicy:Never.
Use ConfigMap mount for tarball data (no stdin piping needed).
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* fix: align registerPluginSettings name with deployed plugin directory
The plugin is deployed to the 'polaris' directory but was registered with
'headlamp-polaris', causing Headlamp to not match the settings component
with the loaded plugin. This fixes all 5 failing E2E settings tests.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* fix: use package name for registerPluginSettings, not directory name
Headlamp identifies plugins by their package.json name (headlamp-polaris),
not the deploy directory name (polaris). The previous commit incorrectly
changed this to 'polaris', causing the settings component to never render
in the plugin settings page — breaking all 5 E2E settings tests.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* fix: align registerPluginSettings name with deploy directory 'polaris'
The shared volume deploy script places the plugin at /headlamp/plugins/polaris/,
so Headlamp matches settings by directory name 'polaris', not the package.json
name 'headlamp-polaris'. This reverts commit b9d718b which incorrectly changed
the registration name back to 'headlamp-polaris'.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* fix: align plugin deploy dir with package.json name, clean stale dirs
The PVC had a stale headlamp-polaris directory from a previous install.
Headlamp loads plugins by scanning the plugins dir and reading package.json
from each subdirectory — it was loading the old build from headlamp-polaris/
while the deploy script was writing to polaris/. The settings registration
name needs to match the plugin name Headlamp identifies.
Changes:
- Deploy script now uses headlamp-polaris as the directory name (matching
package.json name field)
- Deploy pod cleans up both polaris/ and headlamp-polaris/ before deploying
to ensure no stale copies remain
- registerPluginSettings uses headlamp-polaris to match Headlamp's plugin
identifier
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* fix: align registerPluginSettings and E2E test with package.json name
Headlamp identifies plugins by reading package.json from the plugin
directory. Since package.json name is 'headlamp-polaris', both the
registerPluginSettings call and the E2E settings test must use
'headlamp-polaris', not 'polaris'.
- registerPluginSettings('polaris') → registerPluginSettings('headlamp-polaris')
- E2E test locator: text=polaris → text=headlamp-polaris
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* fix(e2e): load main page before settings to ensure plugin list is populated
Headlamp's PluginSettings component initializes its state from
localStorage on mount and never syncs when props.plugins updates later.
If the settings page loads before fetchAndExecutePlugins completes,
the plugin list stays empty and the test can't find "headlamp-polaris".
Fix: navigate to the main page first, wait for the Polaris sidebar
entry to confirm the plugin is loaded (which populates localStorage),
then navigate to the settings page.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* fix(e2e): use client-side routing for settings navigation
The PluginSettings component reads the plugin registry once on mount
and never re-renders when new plugins register. Using page.goto() for
the settings URL re-initializes the SPA, causing PluginSettings to
mount before async plugin scripts finish calling registerPluginSettings().
Replace page.goto() with pushState + popstate to do client-side routing.
This preserves the already-loaded plugin registrations from the main
page, so PluginSettings sees the plugin immediately on mount.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
* fix(e2e): use correct HOME-context URL for plugin settings page
The settings page is at /settings/plugins (HOME sidebar context), not
/c/main/settings/plugins (in-cluster context). The in-cluster URL
doesn't match any route, so PluginSettings never mounted and the
plugin entry was never visible.
With the correct URL, no preloading or client-side routing hacks are
needed — PluginSettings uses useTypedSelector on the Redux plugin store,
so it re-renders automatically when registerPluginSettings() fires.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
---------
Co-authored-by: Gandalf the Greybeard <gandalf@privilegedescalation.dev>
Co-authored-by: Paperclip <noreply@paperclip.ing>
Co-authored-by: Hugh Hackman <hugh@privilegedescalation.com>
Co-authored-by: Hugh Hackman <hugh-hackman[bot]@users.noreply.github.com>
Chris Farhood
privilegedescalation-engineer[bot]
Hugh Hackman