Imported from qwedsazxc78/devops-ai-skill (skills/nginx-ingress-retire/SKILL.md). Install upstream with npx skills add qwedsazxc78/devops-ai-skill --skill nginx-ingress-retire. Copyright stays with the author.

/gitops/retire-nginx — Nginx Ingress Retirement

Retires the nginx ingress controller (common.ingress) and all nginx Ingress resources (common.service) for one or all environments after Gateway API / Traefik migration is complete.

Run this skill only after:

• All services have active HTTPRoutes or Traefik Ingresses
• DNS has been cut over to the Traefik/Gateway LB
• nginx is confirmed idle (no live traffic)

Usage

/gitops/retire-nginx dev          # Retire dev only
/gitops/retire-nginx stg          # Retire stg only
/gitops/retire-nginx prd          # Retire prd only
/gitops/retire-nginx all          # Retire all envs sequentially (dev → stg → prd)

$ARGUMENTS

Core Rules

Step 1 — Parse Arguments

env = $ARGUMENTS   # dev | stg | prd | all

If no argument, enter interactive mode: detect available environments from common.ingress/overlays/ and common.service/overlays/, list them, and ask which to retire.

If all, build the queue [dev, stg, prd] and process each sequentially.

Step 2 — Pre-flight Safety Check (per env)

Run all checks before touching any files. Abort on hard failures.

2a. Verify kustomize build baseline

kustomize build common.service/overlays/<env> > /tmp/retire-nginx-<env>.yaml 2>&1
echo $?   # must be 0

Hard abort if non-zero — do not retire a broken overlay.

2b. Ingress inventory — nginx vs non-nginx breakdown

Build once, reuse for all checks:

# Count nginx-class Ingresses (the ones we're retiring)
grep -c "ingress.class: nginx\|ingressClassName: nginx" /tmp/retire-nginx-<env>.yaml || echo 0

# Count all Ingresses (nginx + Traefik + other)
grep -c "^kind: Ingress" /tmp/retire-nginx-<env>.yaml || echo 0

# Count replacement resources (HTTPRoutes + Traefik Ingresses)
grep -c "^kind: HTTPRoute" /tmp/retire-nginx-<env>.yaml || echo 0
traefik_count=$(grep -A5 "^kind: Ingress" /tmp/retire-nginx-<env>.yaml | grep -c "ingressClassName: traefik\|ingress.class: traefik" || echo 0)

Show the user this breakdown before proceeding:

Pre-flight inventory for <env>:
  nginx Ingresses (to retire):    N
  Traefik Ingresses (to keep):    M
  HTTPRoutes (to keep):           K
  Total Ingresses after retire:   M  (Traefik only)

Hard abort if nginx_count == 0 AND (traefik_count == 0 AND httproute_count == 0): — no replacements exist at all, migration is not done.

Soft warn + confirm if nginx_count == 0 AND (traefik_count > 0 OR httproute_count > 0): — already retired; ask if user wants to continue (for cleanup of controller/archive).

Hard abort if nginx_count > 0 AND traefik_count == 0 AND httproute_count == 0: — nginx is the only routing path; retiring would cause an outage.

2c. Check for active nginx entries in service overlay kustomization

Check both resources: and patches: sections for uncommented nginx files:

# Resources and patches that are active (not commented out, not $patch: delete)
grep -E "nginx" common.service/overlays/<env>/kustomization.yaml \
  | grep -v "^#" \
  | grep -v "patch: delete" \
  | grep -v "^\s*#"

If any active nginx entries found, cross-check the rendered manifest before warning — a file named *-nginx-* may patch the nginx base to a different ingress class (e.g., GCE or ALB):

# Get the rendered ingress.class for each suspected file
kustomize build common.service/overlays/<env> 2>/dev/null \
  | awk '/^kind: Ingress/{f=1} f && /^  name:/{n=$2} f && /ingress\.class:/{print n ": " $2} /^---/{f=0}'

If ingress.class is NOT nginx (e.g., gce, alb, traefik): this is a class-override patch — the file patches the nginx base resource to use a different LB. Do NOT add $patch: delete for this resource and do NOT delete this file. See Step 5c for the correct handling.

If ingress.class IS nginx: warn as usual:

⚠  WARNING: <env> overlay has active nginx ingress files not yet migrated:
     - <file>
   These services still depend on nginx for routing. Retire them first or
   they will lose routing when nginx controller is removed.
   Proceed anyway? (y/N)

Step 3 — Discovery (per env)

Scan for all nginx artifacts. Use the already-built manifest /tmp/retire-nginx-<env>.yaml.

3a. Controller artifacts

# ArgoCD Application for nginx controller
ls common.ingress/argocd/<env>.yaml 2>/dev/null && echo "EXISTS" || echo "ALREADY REMOVED"

# Controller overlay directory (files to delete)
ls common.ingress/overlays/<env>/ 2>/dev/null || echo "ALREADY REMOVED"

3b. Base nginx Ingress resources pulled into this env

Use kustomize build of the base (not grep — file names don't always match resource names):

kustomize build common.service/base 2>/dev/null \
  | awk '/^kind: Ingress/{f=1; name=""; ns=""} 
         f && /^  name:/{name=$2} 
         f && /^  namespace:/{ns=$2} 
         f && /ingress.class: nginx/{print name " (" ns ")"} 
         /^---/{f=0}'

This gives the exact metadata.name and metadata.namespace needed for $patch: delete targets. Cross-reference against the env build to confirm they're actually rendered in this env.

3c. Orphaned nginx files in env overlay

Files that exist in the overlay dir but are NOT referenced in kustomization.yaml:

# Find nginx-named files in the overlay dir
for f in common.service/overlays/<env>/*nginx*.yaml; do
  [ -f "$f" ] || continue
  fname=$(basename "$f")
  # Check if referenced in kustomization (resources or patches)
  if ! grep -qE "^\s+-\s+(path:\s+)?$fname" common.service/overlays/<env>/kustomization.yaml; then
    echo "ORPHAN: $f"
  fi
done

3d. Archive directory

ls common.service/overlays/<env>/archive/ 2>/dev/null | wc -l

Step 4 — Show Removal Plan + Confirm

Display a summary of exactly what will happen before touching any files. Only show sections that apply (e.g., omit ORPHAN FILES if none found):

Nginx Retirement Plan — <env>
══════════════════════════════════════════════════════════════════
PRE-FLIGHT
  nginx Ingresses to retire:    N
  Traefik Ingresses kept:       M
  HTTPRoutes kept:              K

CONTROLLER
  DELETE  common.ingress/argocd/<env>.yaml           (app-of-apps prunes Application → finalizer removes controller)
  KEEP    common.ingress/overlays/<env>/              (dead code — no ArgoCD app references it; clean up later)

SERVICE INGRESSES  ($patch: delete in overlay kustomization — base files stay)
  PATCH   common.service/overlays/<env>/kustomization.yaml
          → <resource-name>  (<namespace>)
          → <resource-name>  (<namespace>)
          ... (<N> resources)

[ORPHAN FILES]  (only if found)
  DELETE  common.service/overlays/<env>/<file>  (not in kustomization.yaml)

[ARCHIVE DIR]  (only if exists)
  DELETE  common.service/overlays/<env>/archive/   (<N> files — migration snapshots)

NOT TOUCHED
  KEEP    common.ingress/argocd/             (folder stays — other env deployment yamls still live here)
  KEEP    common.ingress/argocd/<other-envs>.yaml  (remaining envs still deployed via these apps)
  KEEP    common.ingress/base/                     (<remaining envs> still reference it)
  KEEP    common.service/base/*-nginx-ingress.yaml (<remaining envs> still render these)
  KEEP    common.service/argocd/<env>.yaml         (<env>-infra app manages secrets + active resources)
══════════════════════════════════════════════════════════════════
Proceed? (y/N)

Wait for explicit confirmation before executing.

Step 5 — Execute (per env)

5a. Delete the nginx controller ArgoCD Application yaml

rm common.ingress/argocd/<env>.yaml

Why this is sufficient — the cascade chain handles cluster cleanup automatically:

repo: dev.yaml deleted
  → app-of-apps (prune: true) syncs → prunes Application object from cluster
  → Application has resources-finalizer.argocd.argoproj.io
  → finalizer cascades → deletes nginx controller Deployment, Service, ConfigMap, RBAC

The overlay dir (common.ingress/overlays/<env>/) becomes dead code — no ArgoCD app references it anymore. Leave it in place (harmless) or clean it up in a follow-up commit. Skip this step if the file doesn't exist (already removed).

5c. Add $patch: delete for each base nginx Ingress

Append to the existing patches: section in common.service/overlays/<env>/kustomization.yaml. Use the exact metadata.name values discovered in Step 3b.

  # Nginx ingress retirement (<env>): delete all base nginx Ingress resources — replaced by Traefik/HTTPRoutes
  - patch: |-
      $patch: delete
      apiVersion: networking.k8s.io/v1
      kind: Ingress
      metadata:
        name: <resource-name>
    target:
      kind: Ingress
      name: <resource-name>

namePrefix note: Target names are the original names before namePrefix: <env>- is applied. Kustomize applies patches before namePrefix transformation. Always use the un-prefixed name.

configurations: section: If the kustomization has a configurations: block (e.g., for Gateway API nameref), add patches before configurations to preserve the section order.

Special case: class-override patches (e.g., GCE, ALB)

If a base nginx Ingress is patched by an env overlay to use a different ingress class (e.g., ingress.class: gce for a GCE external LB), do NOT add $patch: delete for it — the resource must stay in the rendered output as a non-nginx Ingress.

Correct handling:

1. Keep the existing patch file under patches: — it overrides the base nginx class to GCE/ALB
2. Do NOT add $patch: delete for this resource name
3. Rename the patch file to reflect its actual class (e.g., n8n-nginx-ingress.yaml → n8n-gce-ingress.yaml) for clarity — the metadata.name inside stays unchanged so the K8s resource name and GCE LB binding are undisturbed

Why not rename the K8s resource? Changing metadata.name causes the old Ingress to be deleted and a new one created — this triggers GCE/cloud LB reprovisioning. Keep the name to avoid downtime; rename only the file.

Cannot use $patch: delete + standalone resource with same name: Kustomize rejects duplicate resource IDs during accumulation, before patches are applied. If you try to delete the base resource AND add a same-name standalone resource, it will error.

5d. Delete orphaned nginx files

Only delete files confirmed as orphans in Step 3c:

rm common.service/overlays/<env>/<orphan-file>.yaml

5e. Delete archive directory

rm -rf common.service/overlays/<env>/archive/

Skip if the directory doesn't exist.

Step 6 — Validate (per env)

6a. Confirm zero nginx-class Ingress resources remain

kustomize build common.service/overlays/<env> 2>&1 \
  | grep -c "ingress.class: nginx\|ingressClassName: nginx"
# Expected: 0

Do NOT check total kind: Ingress count — envs with Traefik ingresses will legitimately have > 0. The retirement target is specifically nginx-class resources, not all Ingress kinds.

If non-zero, identify which nginx Ingress remains:

kustomize build common.service/overlays/<env> \
  | awk '/^kind: Ingress/{f=1} f && /^  name:/{n=$2} f && /ingress.class: nginx/{print "STILL PRESENT: " n} /^---/{f=0}'

Report and halt — do not proceed to commit.

6b. Verify remaining Ingresses are non-nginx class

kustomize build common.service/overlays/<env> \
  | awk '/^kind: Ingress/{f=1} f && /^  name:/{print $2} /^---/{f=0}'
# All names should be *-traefik-* or similar non-nginx resources

6c. Regression check — all other envs still build cleanly

for other_env in dev stg prd; do
  [ "$other_env" = "<env>" ] && continue
  kustomize build common.service/overlays/$other_env > /dev/null \
    && echo "$other_env: PASS" \
    || echo "$other_env: FAIL"
done

Halt and report if any other env fails — do not commit.

Step 7 — All-Env Cleanup Check (only after retire-nginx all)

After all three envs retired, check if common.ingress/ module is fully orphaned:

ls common.ingress/overlays/ 2>/dev/null   # should be empty
ls common.ingress/argocd/   2>/dev/null   # should show only README.md

If no overlay dirs remain, offer to archive the entire module:

⚠  All env overlays for common.ingress are removed.
   common.ingress/base/ is no longer referenced by any overlay.
   Recommend: archive or delete common.ingress/ entirely.
   Proceed? (y/N)

If confirmed, choose:

# Option A: archive in place (git history preserved + files accessible)
mkdir -p common.ingress/archive
git mv common.ingress/base  common.ingress/archive/base
git mv common.ingress/argocd common.ingress/archive/argocd

# Option B: delete entirely (git history preserves files)
rm -rf common.ingress/

Then check if base nginx Ingress files in common.service/base/ are still needed:

# Any remaining kustomization still rendering nginx ingresses from base?
for env in dev stg prd; do
  grep -E "nginx" common.service/overlays/$env/kustomization.yaml \
    | grep -v "^#" | grep -v "patch: delete" \
    && echo "  → $env still uses base nginx files"
done

If none remain, offer to remove nginx entries from common.service/base/kustomization.yaml and delete the corresponding base nginx Ingress files.

Step 8 — Summary Report

Nginx Retirement — Summary
══════════════════════════════════════════════════════════════════
Env   Controller  Overlay  nginx Ingress  Traefik kept  Validation
────────────────────────────────────────────────────────────────
<env> DELETED     DELETED  0              <N>            PASS
══════════════════════════════════════════════════════════════════
Files deleted:
  common.ingress/argocd/<env>.yaml
  common.ingress/overlays/<env>/  (<N> files)
  common.service/overlays/<env>/archive/  (<N> files)
  <orphan files if any>

Files modified:
  common.service/overlays/<env>/kustomization.yaml
  → added $patch: delete for <N> base nginx Ingress resources

Files kept (active routing / other envs):
  common.service/overlays/<env>/*-traefik-ingress.yaml  (<N> Traefik Ingresses)
  common.ingress/base/  (referenced by remaining envs)
  common.service/base/*-nginx-ingress.yaml  (referenced by remaining envs)

Regression:
  <other-env-1>  PASS
  <other-env-2>  PASS

Next steps:
  1. git add + commit:
     retire(nginx/<env>): remove controller app + exclude base nginx Ingress resources
  2. Push → ArgoCD auto-sync prunes nginx controller from <env> cluster
  3. kubectl get pods -n ingress-nginx  (expect: 0 pods)
  4. Monitor: no 502s — all traffic through Traefik/HTTPRoutes

Graceful Degradation