feat(k8s): migrate pipelines to prod EKS as a private service - #31
Merged
Conversation
Deploy the Open WebUI Pipelines filter service onto the self-managed prod
EKS cluster (hub-and-spoke ArgoCD), mirroring the skillshield backend
pattern. Pipelines is called only in-cluster by ai-gateway /
ai-gateway-data at /v1/perform_filters, so it deploys PRIVATE: ClusterIP
service, no ingress, no DNS. It is an arbitrary-code plugin runtime, so
private is required.
Chart (k8s/chart): Deployment + Service (ClusterIP:8080) + HPA (min1/max2)
+ ExternalSecret (ESO, prod/pipelines -> {{ .Release.Name }}-secrets via
envFrom). No migrate Job (stateless — no DB/redis/migrations). No
securityContext (mirrors skillshield; the image runs as root and writes to
PIPELINES_DIR at runtime). Probes hit GET / (get_status -> unauthenticated
200), with a generous startup window for cold model load. PIPELINES_API_KEY
comes from Secrets Manager, never the app's insecure default.
prod overlay (k8s/prod/values.yaml): image :prod, ingress disabled,
externalSecret prod/pipelines, autoscaling min1/max2.
CI: deploy-eks.yml (SHA-pinned actions; PR=build-only, push main=push :sha,
dispatch=advance :prod + rollout with fail-fast; no migrate step) and
chart-validate.yml (helm lint + template on k8s/** changes).
Removed the unused configMap template (skillshield has none; pipelines ships
no gateway plugin config). Left dev/staging overlays untouched. The gateway
keeps calling Porter pipelines until the later cutover.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…imeouts - Container securityContext: drop ALL caps + allowPrivilegeEscalation:false + seccomp RuntimeDefault (NOT runAsNonRoot — image runs as root and writes PIPELINES_DIR). Restores hardening for the plugin-execution service (greptile + cursor HIGH). - externalSecret.enabled defaults TRUE — pod always gets PIPELINES_API_KEY from SM, never the 0p3n-w3bu! fallback (greptile). `required` guard on remoteKey (greptile "empty key renders") and on image.repository (greptile "empty image"). - id-token:write scoped to the build-and-deploy job, not workflow-global — PR runs / the resolve job never receive it (cursor HIGH: PR-branch cred theft). - dev overlay autoscaling.enabled=false so the held-off dev (PR #30, replicas 0) doesn't get scaled back up by the now-default HPA (cursor). - rollout status timeout 10m->15m and job timeout 30->60m for the heavy torch/transformers image + ML cold starts (cursor x2). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes using high effort and found 2 potential issues.
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit ee59818. Configure here.
…leak - ExternalSecret uses an explicit `data` mapping keyed on secretKeys ([PIPELINES_API_KEY]) via remoteRef.property, so a missing key FAILS the ESO sync (pod won't start) instead of silently letting the app use its 0p3n-w3bu! default (cursor HIGH). - dev overlay adds replicaCount: 0 (not just HPA off) so the held-off env stays at zero pods (cursor). - rollout-failure diagnostics no longer stream pod logs into the Actions log — this service handles LLM/guardrail content that may be sensitive; describe + events + pod status suffice (cursor MEDIUM). Also corrected the 10m->15m text. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
zeus-12
approved these changes
Jul 13, 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
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.

Migrates the Open WebUI Pipelines filter service off Porter onto the self-managed prod EKS cluster (hub-and-spoke ArgoCD), mirroring the skillshield backend pattern.
Private by design
Pipelines is called only in-cluster by
ai-gateway/ai-gateway-dataat/v1/perform_filters, and it is an arbitrary-code plugin runtime — so it deploys PRIVATE: ClusterIP service, no ingress, no DNS.What's here
k8s/chart/— Deployment + Service (ClusterIP:8080, targetPort http) + HPA (min1/max2, CPU/mem 75) + ExternalSecret (ESOprod/pipelines→<release>-secrets, injected viaenvFrom). Ingress template kept but gatedenabled: false.GET /(get_status→ unauthenticated200 {"status": true}); generous startup window for cold model load.PIPELINES_API_KEYcomes from Secrets Manager, never the app's insecure0p3n-w3bu!default.image: …/k8s/pipelines:prod,pullPolicy: Always;strategy: RollingUpdate maxSurge 1 maxUnavailable 0;revisionHistoryLimit: 3.k8s/prod/values.yaml— prod overlay (tagprod, ingress disabled,prod/pipelines, autoscaling min1/max2, env PORT/HOST/PIPELINES_DIR)..github/workflows/deploy-eks.yml— SHA-pinned actions. PR = build-only; push main = push:shaonly; dispatch = advance:prod+ rollout (fail-fast). Migrate step removed (stateless)..github/workflows/chart-validate.yml— helm lint + template onk8s/**.Notable deltas from the existing chart
<release>-env→<release>-secrets(skillshield/envFromconvention).securityContext(mirrors skillshield; the image runs as root and writes toPIPELINES_DIRat runtime — non-root uid 1000 would break pip installs / plugin writes).configMaptemplate and the hardcodedDD_AGENT_HOSTenv (neither in skillshield; pipelines ships no gateway plugin config).Validation
helm lint k8s/chart -f k8s/prod/values.yaml✅helm template pipelines-prod k8s/chart -f k8s/prod/values.yaml→ Deployment, Service (ClusterIP:8080), ExternalSecret, HPA (min1/max2); no Ingress ✅Companion
unbound-infraPR wires ArgoCD + the repo secret + OIDC trust. The gateway keeps calling Porter pipelines until the later cutover.🤖 Generated with Claude Code
Note
Medium Risk
Privileged CI (OIDC → ECR/EKS) and dispatch-only
:prodtagging affect prod rollout; chart changes harden secrets and network exposure but the workload remains a root plugin runtime executing arbitrary code.Overview
Adds prod EKS packaging for Pipelines (private ClusterIP, ArgoCD-managed manifests) plus CI that separates image build from prod rollout.
Helm chart is reshaped for the skillshield-style deploy: drops ConfigMap/
DD_AGENT_HOST, renames ExternalSecret output to<release>-secrets, switches ESO fromdataFromextract to pinnedsecretKeys(so missingPIPELINES_API_KEYfails sync instead of falling back to the app default), moves non-secret config intoenv, and uses values-driven seccomp/cap-drop hardening withoutrunAsNonRoot. Addsk8s/prod/values.yaml(tag:prod,prod/pipelines, ingress off, HPA 1–2); dev overlay scales to 0 with HPA off..github/workflows/deploy-eks.ymlbuilds/pushes immutable:shaon main;:prodadvances only onworkflow_dispatch(then rollout restart with 15m fail-fast)..github/workflows/chart-validate.ymlrunshelm lint+helm templateonk8s/**so chart-only changes still get CI.k8s/README.mddocuments private in-cluster cutover prerequisites.Reviewed by Cursor Bugbot for commit 0fed160. Bugbot is set up for automated code reviews on this repo. Configure here.
Greptile Summary
This PR moves the Pipelines service onto the private prod EKS path. The main changes are:
Confidence Score: 5/5
This looks safe to merge.
Important Files Changed
Reviews (3): Last reviewed commit: "fix(chart,ci): round 2 review — enforce ..." | Re-trigger Greptile
Context used:
Learned From
websentry-ai/ai-gateway-data#448