Merge pull request #69 from rubenhensen/renovate/towfiqi-serpbear-3.x

chore(deps): update towfiqi/serpbear docker tag to v3.1.2

.github/workflows/validate.yaml

@@ -18,7 +18,7 @@ concurrency:
 
 jobs:
   validate:
-    runs-on: ubuntu-24.04
+    runs-on: ubuntu-26.04
     timeout-minutes: 20
     steps:
       - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1

actual-budget/helm-actual-budget.yaml

@@ -12,7 +12,7 @@ spec:
   project: default
   source:
     repoURL: https://community-charts.github.io/helm-charts
-    targetRevision: 1.8.9
+    targetRevision: 1.9.5
     chart: actualbudget
     helm:
       values: |

actual-budget/networkpolicy.yaml

@@ -55,3 +55,29 @@ spec:
               - 169.254.0.0/16
               - 172.16.0.0/12
               - 192.168.0.0/16
+---
+# cert-manager's HTTP-01 solver, which runs only while a certificate
+# renews: Let's Encrypt and cert-manager's own self check reach it on 8089
+# through Envoy. The policies above let Envoy reach the app's port only,
+# which left the solver unreachable (found 2026-10-26, when ynab's renewal
+# hung).
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: acme-http01-solver
+spec:
+  podSelector:
+    matchLabels:
+      acme.cert-manager.io/http01-solver: "true"
+  policyTypes:
+    - Ingress
+  ingress:
+    - from:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: envoy-gateway-system
+          podSelector:
+            matchLabels:
+              app.kubernetes.io/name: envoy
+      ports:
+        - port: 8089

apps/templates/authentik-helm.yaml

@@ -103,6 +103,8 @@ spec:
               requests:
                 cpu: 70m
                 memory: 560Mi
+              limits:
+                memory: 1536Mi
             volumes:
               - name: patched-ca-bundle
                 configMap:
@@ -192,6 +194,8 @@ spec:
               requests:
                 cpu: 100m
                 memory: 336Mi
+              limits:
+                memory: 1Gi
             volumes:
               - name: patched-ca-bundle
                 configMap:

apps/templates/cert-manager-helm.yaml

@@ -38,16 +38,22 @@ spec:
           requests:
             cpu: 10m
             memory: 48Mi
+          limits:
+            memory: 128Mi
         cainjector:
           resources:
             requests:
               cpu: 10m
               memory: 80Mi
+            limits:
+              memory: 256Mi
         webhook:
           resources:
             requests:
               cpu: 10m
               memory: 32Mi
+            limits:
+              memory: 128Mi
   syncPolicy:
     syncOptions:
     - CreateNamespace=true

apps/templates/cloudnative-pg-folder.yaml

@@ -0,0 +1,19 @@
+apiVersion: argoproj.io/v1alpha1
+kind: Application
+metadata:
+  name: cloudnative-pg-folder
+  namespace: argocd
+  finalizers:
+  - resources-finalizer.argocd.argoproj.io
+spec:
+  project: default
+  source:
+    repoURL: https://github.com/rubenhensen/k8scd.git
+    targetRevision: HEAD
+    path: cloudnative-pg
+  destination:
+    server: https://kubernetes.default.svc
+    namespace: cnpg-system
+  syncPolicy:
+    automated:
+      selfHeal: true

apps/templates/cloudnativepg-helm.yaml

@@ -28,6 +28,8 @@ spec:
           requests:
             cpu: 10m
             memory: 48Mi
+          limits:
+            memory: 128Mi
   destination:
     server: https://kubernetes.default.svc
     namespace: cnpg-system

apps/templates/default-folder.yaml

@@ -0,0 +1,19 @@
+apiVersion: argoproj.io/v1alpha1
+kind: Application
+metadata:
+  name: default-folder
+  namespace: argocd
+  finalizers:
+  - resources-finalizer.argocd.argoproj.io
+spec:
+  project: default
+  source:
+    repoURL: https://github.com/rubenhensen/k8scd.git
+    targetRevision: HEAD
+    path: default
+  destination:
+    server: https://kubernetes.default.svc
+    namespace: default
+  syncPolicy:
+    automated:
+      selfHeal: true

apps/templates/envoy-gateway-helm.yaml

@@ -12,7 +12,7 @@ spec:
     repoURL: docker.io/envoyproxy
     chart: gateway-helm
     # v1.9 supports Kubernetes 1.33 to 1.36
-    targetRevision: v1.9.2
+    targetRevision: 1.9.3
     helm:
       values: |
         config:

apps/templates/external-secrets-folder.yaml

@@ -0,0 +1,19 @@
+apiVersion: argoproj.io/v1alpha1
+kind: Application
+metadata:
+  name: external-secrets-folder
+  namespace: argocd
+  finalizers:
+  - resources-finalizer.argocd.argoproj.io
+spec:
+  project: default
+  source:
+    repoURL: https://github.com/rubenhensen/k8scd.git
+    targetRevision: HEAD
+    path: external-secrets
+  destination:
+    server: https://kubernetes.default.svc
+    namespace: external-secrets
+  syncPolicy:
+    automated:
+      selfHeal: true

apps/templates/external-secrets-helm.yaml

@@ -24,16 +24,22 @@ spec:
             requests:
               cpu: 20m
               memory: 64Mi
+            limits:
+              memory: 128Mi
           certController:
             resources:
               requests:
                 cpu: 10m
                 memory: 64Mi
+              limits:
+                memory: 128Mi
           webhook:
             resources:
               requests:
                 cpu: 10m
                 memory: 32Mi
+              limits:
+                memory: 128Mi
   destination:
     server: https://kubernetes.default.svc
     namespace: external-secrets

apps/templates/mbgwp-official-helm.yaml

@@ -12,7 +12,7 @@ spec:
   project: default
   source:
     repoURL: https://groundhog2k.github.io/helm-charts/
-    targetRevision: 0.16.7
+    targetRevision: 0.16.8
     chart: wordpress
     helm:
       values: |

apps/templates/plugin-barman-cloud-helm.yaml

@@ -0,0 +1,34 @@
+apiVersion: argoproj.io/v1alpha1
+kind: Application
+metadata:
+  name: plugin-barman-cloud
+  namespace: argocd
+  finalizers:
+  - resources-finalizer.argocd.argoproj.io
+spec:
+  project: default
+  source:
+    chart: plugin-barman-cloud
+    repoURL: https://cloudnative-pg.github.io/charts
+    targetRevision: 0.8.1
+    helm:
+      values: |
+        # CNPG's barman cloud plugin, which backs the PostgreSQL clusters
+        # up to Garage on the NAS (authentik/postgresql-backup.yaml,
+        # immich/postgresql-backup.yaml). It has to run in the operator's
+        # namespace; the chart gets its TLS certificates from cert-manager.
+        resources:
+          requests:
+            cpu: 10m
+            memory: 64Mi
+          limits:
+            memory: 128Mi
+  destination:
+    server: https://kubernetes.default.svc
+    namespace: cnpg-system
+  syncPolicy:
+    syncOptions:
+    - ServerSideApply=true
+    automated:
+      prune: true
+      selfHeal: true

argocd/overlays/resources.patch.yaml

@@ -1,6 +1,6 @@
 # Requests from what the components used on 2026-10-25 (kubectl top every
-# 5 minutes; highest value plus 10%). No limits until there are daytime
+# 5 minutes; highest value plus 10%). Memory limits at about twice the
-# numbers, see docs/roadmap.md.
+# highest value seen then or on 2026-10-27, rounded up; no CPU limits.
 apiVersion: apps/v1
 kind: StatefulSet
 metadata:
@@ -14,6 +14,8 @@ spec:
             requests:
               cpu: 150m
               memory: 704Mi
+            limits:
+              memory: 2560Mi
 ---
 apiVersion: apps/v1
 kind: Deployment
@@ -28,6 +30,8 @@ spec:
             requests:
               cpu: 100m
               memory: 128Mi
+            limits:
+              memory: 512Mi
 ---
 apiVersion: apps/v1
 kind: Deployment
@@ -42,6 +46,8 @@ spec:
             requests:
               cpu: 10m
               memory: 64Mi
+            limits:
+              memory: 256Mi
 ---
 apiVersion: apps/v1
 kind: Deployment
@@ -56,6 +62,8 @@ spec:
             requests:
               cpu: 20m
               memory: 96Mi
+            limits:
+              memory: 256Mi
 ---
 apiVersion: apps/v1
 kind: Deployment
@@ -70,6 +78,8 @@ spec:
             requests:
               cpu: 10m
               memory: 48Mi
+            limits:
+              memory: 128Mi
 ---
 apiVersion: apps/v1
 kind: Deployment
@@ -84,6 +94,8 @@ spec:
             requests:
               cpu: 10m
               memory: 48Mi
+            limits:
+              memory: 128Mi
 ---
 apiVersion: apps/v1
 kind: Deployment
@@ -98,3 +110,5 @@ spec:
             requests:
               cpu: 10m
               memory: 32Mi
+            limits:
+              memory: 128Mi

authentik/networkpolicy.yaml

@@ -108,7 +108,8 @@ spec:
 ---
 # The CNPG operator checks each instance on its status port and connects to
 # PostgreSQL; the instance manager inside each pod watches its Cluster
-# through the API server.
+# through the API server. The barman cloud sidecar sends the WAL and the
+# base backups to Garage on the NAS (postgresql-backup.yaml).
 apiVersion: networking.k8s.io/v1
 kind: NetworkPolicy
 metadata:
@@ -135,3 +136,34 @@ spec:
             cidr: 192.168.1.204/32
       ports:
         - port: 16443
+    - to:
+        - ipBlock:
+            cidr: 192.168.1.99/32
+      ports:
+        - port: 3900
+---
+# cert-manager's HTTP-01 solver, which runs only while a certificate
+# renews: Let's Encrypt and cert-manager's own self check reach it on 8089
+# through Envoy. The policies above let Envoy reach the app's port only,
+# which left the solver unreachable (found 2026-10-26, when ynab's renewal
+# hung).
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: acme-http01-solver
+spec:
+  podSelector:
+    matchLabels:
+      acme.cert-manager.io/http01-solver: "true"
+  policyTypes:
+    - Ingress
+  ingress:
+    - from:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: envoy-gateway-system
+          podSelector:
+            matchLabels:
+              app.kubernetes.io/name: envoy
+      ports:
+        - port: 8089

authentik/postgresql-backup.yaml

@@ -0,0 +1,94 @@
+# Backups of authentik's database by CNPG itself, through the barman cloud
+# plugin (apps/templates/plugin-barman-cloud-helm.yaml): a base backup every
+# night, and every WAL segment as soon as it is full or 5 minutes old
+# (archive_timeout). Both go to the bucket cnpg-authentik in Garage on the
+# NAS (nas/garage), so the database can be restored to any moment of the
+# last 30 days. Velero copies the data directory while PostgreSQL runs,
+# which is not a consistent copy.
+# The key is Garage's cnpg-authentik, in Vault as kv/cnpg-authentik-s3; it
+# can only reach this bucket.
+apiVersion: external-secrets.io/v1
+kind: ExternalSecret
+metadata:
+  name: authentik-postgres-s3
+spec:
+  secretStoreRef:
+    name: vault-backend
+    kind: ClusterSecretStore
+  refreshInterval: 15m
+  target:
+    name: authentik-postgres-s3
+    creationPolicy: Owner
+    template:
+      engineVersion: v2
+      data:
+        ACCESS_KEY_ID: "{{ .access_key_id }}"
+        ACCESS_SECRET_KEY: "{{ .secret_access_key }}"
+        # Garage rejects requests signed for another region
+        REGION: garage
+  data:
+    - secretKey: access_key_id
+      remoteRef:
+        key: kv/cnpg-authentik-s3
+        property: access_key_id
+    - secretKey: secret_access_key
+      remoteRef:
+        key: kv/cnpg-authentik-s3
+        property: secret_access_key
+---
+apiVersion: barmancloud.cnpg.io/v1
+kind: ObjectStore
+metadata:
+  name: authentik-postgres
+  annotations:
+    # The CRD comes with the plugin's own app
+    argocd.argoproj.io/sync-options: SkipDryRunOnMissingResource=true
+spec:
+  retentionPolicy: 30d
+  configuration:
+    destinationPath: s3://cnpg-authentik/
+    endpointURL: http://192.168.1.99:3900
+    s3Credentials:
+      accessKeyId:
+        name: authentik-postgres-s3
+        key: ACCESS_KEY_ID
+      secretAccessKey:
+        name: authentik-postgres-s3
+        key: ACCESS_SECRET_KEY
+      region:
+        name: authentik-postgres-s3
+        key: REGION
+    wal:
+      compression: gzip
+    data:
+      compression: gzip
+  instanceSidecarConfiguration:
+    resources:
+      requests:
+        cpu: 10m
+        memory: 64Mi
+      limits:
+        memory: 512Mi
+    env:
+      # Newer boto3 adds checksums that S3-compatible stores can reject;
+      # the plugin's docs suggest sending them only when required.
+      - name: AWS_REQUEST_CHECKSUM_CALCULATION
+        value: when_required
+      - name: AWS_RESPONSE_CHECKSUM_VALIDATION
+        value: when_required
+---
+apiVersion: postgresql.cnpg.io/v1
+kind: ScheduledBackup
+metadata:
+  name: authentik-postgres-daily
+spec:
+  cluster:
+    name: authentik-postgres
+  # 03:00 UTC, after Longhorn's and Velero's backups; immediate: one right
+  # away, so WAL archiving has a base backup to start from.
+  schedule: "0 0 3 * * *"
+  immediate: true
+  backupOwnerReference: self
+  method: plugin
+  pluginConfiguration:
+    name: barman-cloud.cloudnative-pg.io

authentik/postgresql-cluster.yaml

@@ -6,12 +6,20 @@ spec:
   instances: 1
   # Pinned so a new operator default (PostgreSQL 18 since CNPG 1.26.3) can't
   # start an in-place major upgrade. CNPG filled this in at creation.
-  imageName: ghcr.io/cloudnative-pg/postgresql:17.4
+  imageName: ghcr.io/cloudnative-pg/postgresql:17.11
   primaryUpdateStrategy: unsupervised
+  # WAL archiving and base backups to Garage, see postgresql-backup.yaml
+  plugins:
+    - name: barman-cloud.cloudnative-pg.io
+      isWALArchiver: true
+      parameters:
+        barmanObjectName: authentik-postgres
   resources:
     requests:
       cpu: 50m
       memory: 256Mi
+    limits:
+      memory: 1Gi
   bootstrap:
     initdb:
       database: authentik

bolletjestrui/networkpolicy.yaml

@@ -54,3 +54,29 @@ spec:
               - 169.254.0.0/16
               - 172.16.0.0/12
               - 192.168.0.0/16
+---
+# cert-manager's HTTP-01 solver, which runs only while a certificate
+# renews: Let's Encrypt and cert-manager's own self check reach it on 8089
+# through Envoy. The policies above let Envoy reach the app's port only,
+# which left the solver unreachable (found 2026-10-26, when ynab's renewal
+# hung).
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: acme-http01-solver
+spec:
+  podSelector:
+    matchLabels:
+      acme.cert-manager.io/http01-solver: "true"
+  policyTypes:
+    - Ingress
+  ingress:
+    - from:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: envoy-gateway-system
+          podSelector:
+            matchLabels:
+              app.kubernetes.io/name: envoy
+      ports:
+        - port: 8089

cert-manager/networkpolicy.yaml

@@ -0,0 +1,73 @@
+# cert-manager's three pods reach each other, DNS and the API server (the
+# node, port 16443), and take connections only from each other and the
+# node: the API server calls the webhook from there and the kubelet probes
+# come from there too (10.1.7.64 is the node's Calico VXLAN address,
+# projectcalico.org/IPv4VXLANTunnelAddr). Nothing scrapes the metrics on
+# 9402. The HTTP-01 solver pods run in each Certificate's own namespace
+# (acme-http01-solver in <namespace>/networkpolicy.yaml), not here.
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: default
+spec:
+  podSelector: {}
+  policyTypes:
+    - Ingress
+    - Egress
+  ingress:
+    - from:
+        - podSelector: {}
+    - from:
+        - ipBlock:
+            cidr: 192.168.1.204/32
+        - ipBlock:
+            cidr: 10.1.7.64/32
+  egress:
+    - to:
+        - podSelector: {}
+    - to:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: kube-system
+          podSelector:
+            matchLabels:
+              k8s-app: kube-dns
+      ports:
+        - protocol: UDP
+          port: 53
+        - protocol: TCP
+          port: 53
+    - to:
+        - ipBlock:
+            cidr: 192.168.1.204/32
+      ports:
+        - port: 16443
+---
+# The controller talks to Let's Encrypt (443), and before it lets Let's
+# Encrypt check a challenge it fetches http://<host>/.well-known/... itself
+# (80). Every certificate's host has a public address, node001's or the
+# home router's (checked 2026-10-26), so the LAN stays out.
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: controller-to-internet
+spec:
+  podSelector:
+    matchLabels:
+      app.kubernetes.io/name: cert-manager
+      app.kubernetes.io/component: controller
+  policyTypes:
+    - Egress
+  egress:
+    - to:
+        - ipBlock:
+            cidr: 0.0.0.0/0
+            except:
+              - 10.0.0.0/8
+              - 100.64.0.0/10
+              - 169.254.0.0/16
+              - 172.16.0.0/12
+              - 192.168.0.0/16
+      ports:
+        - port: 80
+        - port: 443

cloudnative-pg/networkpolicy.yaml

@@ -0,0 +1,56 @@
+# The CloudNativePG operator reaches DNS, the API server (the node, port
+# 16443) and the PostgreSQL instances it runs in other namespaces: the
+# instance manager's status port (8000) and PostgreSQL (5432), which the
+# instances' own policies let in (postgres-cnpg in authentik and immich).
+# Besides its own namespace it takes connections only from the node: the
+# API server calls its webhook (9443) from there and the kubelet probes
+# come from there too (10.1.7.64 is the node's Calico VXLAN address,
+# projectcalico.org/IPv4VXLANTunnelAddr). The chart's Application
+# (apps/templates/cloudnativepg-helm.yaml) only installs the chart, so
+# this folder gets an Application of its own, cloudnative-pg-folder.
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: default
+spec:
+  podSelector: {}
+  policyTypes:
+    - Ingress
+    - Egress
+  ingress:
+    - from:
+        - podSelector: {}
+    - from:
+        - ipBlock:
+            cidr: 192.168.1.204/32
+        - ipBlock:
+            cidr: 10.1.7.64/32
+  egress:
+    - to:
+        - podSelector: {}
+    - to:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: kube-system
+          podSelector:
+            matchLabels:
+              k8s-app: kube-dns
+      ports:
+        - protocol: UDP
+          port: 53
+        - protocol: TCP
+          port: 53
+    - to:
+        - ipBlock:
+            cidr: 192.168.1.204/32
+      ports:
+        - port: 16443
+    - to:
+        - namespaceSelector: {}
+          podSelector:
+            matchExpressions:
+              - key: cnpg.io/cluster
+                operator: Exists
+      ports:
+        - port: 8000
+        - port: 5432

default/networkpolicy.yaml

@@ -0,0 +1,14 @@
+# Nothing runs in default (2026-10-26: only the kubernetes Service, whose
+# endpoint is the API server on the node, which this doesn't touch). A pod
+# started here, by `kubectl run` or a manifest without a namespace, gets
+# no network at all, instead of reaching the LAN, Vault and every app.
+# Anything that needs a network gets a namespace and a policy of its own.
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: default-deny
+spec:
+  podSelector: {}
+  policyTypes:
+    - Ingress
+    - Egress

dns/cronjob.yaml

@@ -13,7 +13,7 @@ spec:
       restartPolicy: Never
       containers:
         - name: sync
-          image: python:3.12-alpine
+          image: python:3.15-alpine
           command:
             - sh
             - -c

dns/domains/rubenhensen.nl.yaml

@@ -18,7 +18,7 @@ records:
   - name: "ldap"
     expire: 300
     type: A
-    content: "62.41.86.27"
+    content: "62.41.86.53"
   - name: "argocd"
     expire: 300
     type: A
@@ -60,7 +60,7 @@ records:
   - name: "s2"
     expire: 300
     type: A
-    content: "62.41.86.27"
+    content: "62.41.86.53"
   # ntfy for Gatus alerts, one next to each Gatus. Same reasoning as s2.
   - name: "ntfy1"
     expire: 300
@@ -69,7 +69,7 @@ records:
   - name: "ntfy2"
     expire: 300
     type: A
-    content: "62.41.86.27"
+    content: "62.41.86.53"
   - name: "@"
     expire: 300
     type: MX

dns/kustomization.yaml

@@ -7,6 +7,7 @@ resources:
   - external-secret.yaml
   - rbac.yaml
   - cronjob.yaml
+  - networkpolicy.yaml
 
 configMapGenerator:
   - name: dns-sync-script

dns/networkpolicy.yaml

@@ -0,0 +1,37 @@
+# The only pod here is the dns-sync Job (a PostSync hook): it installs its
+# Python packages from PyPI and pushes domains/ to TransIP's API, so it
+# reaches DNS and the internet (not the LAN, the API server or other
+# namespaces). Nothing connects to it.
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: default
+spec:
+  podSelector: {}
+  policyTypes:
+    - Ingress
+    - Egress
+  egress:
+    - to:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: kube-system
+          podSelector:
+            matchLabels:
+              k8s-app: kube-dns
+      ports:
+        - protocol: UDP
+          port: 53
+        - protocol: TCP
+          port: 53
+    - to:
+        - ipBlock:
+            cidr: 0.0.0.0/0
+            except:
+              - 10.0.0.0/8
+              - 100.64.0.0/10
+              - 169.254.0.0/16
+              - 172.16.0.0/12
+              - 192.168.0.0/16
+      ports:
+        - port: 443

docs/immich-scouts-online.md

@@ -27,7 +27,9 @@ view, and everything is shared with everyone automatically.
 - **immich has no user groups**; albums are shared per user, as viewer or
   editor. The owner and editors of an album may add users and change
   their roles. Search only covers your own photos and those of partners
-  you show in your timeline.
+  you show in your timeline. People (face recognition) are per user too,
+  and a partner's library doesn't bring its people along; since 3.3 a
+  user can share people with the users in their cluster group.
 
 ## Design
 
@@ -46,7 +48,10 @@ view, and everything is shared with everyone automatically.
   every user into every album of the group account (`fotoupload@`), as
   editor or viewer by quota; the group account's library as a partner of
   everyone, shown in the timeline; album invitations for albums older than
-  a week hidden; sessions of OAuth users older than 30 days ended. Albums
+  a week hidden; sessions of OAuth users older than 30 days ended;
+  everyone in the group account's cluster group with its people shared
+  (names editable by those who may upload, added 2026-10-28, when nobody
+  but the group account could search by face). Albums
   that kaderleden create get the group account as editor first (in the
   database, since the API only lets an album's owner or editors share
   it), so they go to everyone too; names starting with "Privé"/"Prive"

docs/roadmap.md

@@ -1,22 +1,27 @@
 # Roadmap
 
-Where the cluster stands and what is left, as of 2026-10-25. Takes over
+Where the cluster stands and what is left, as of 2026-10-28. Takes over
 from wishlist.txt (January 2026, removed); its items are sorted in at the
 bottom.
 
 ## State
 
 One node: opti1, an OptiPlex 5060 with an i5-8600T, 32 GB and one 1 TB
-SATA SSD that is 72% full. microk8s v1.36.5 from 1.36/edge, snap held.
+SATA SSD that is 54% full. microk8s v1.36.5 from 1.36/edge, snap held.
-35 Argo CD apps (with metrics-server), all Synced/Healthy; every Longhorn
+39 Argo CD apps (with metrics-server), all Synced/Healthy; every Longhorn
-volume has one replica. authentik 2026.8.5, Vault 2.2.1. Backups go to the
+volume has one replica. authentik 2026.8.5, Vault 2.2.1, immich 3.3.5 on
-NAS (192.168.1.99): Longhorn to NFS at 00:00 (7 kept), Velero to Garage
+PostgreSQL 17, MariaDB 11.4.13. Memory requests and limits on everything
-(MinIO until 2026-10-26) at 01:00 (daily kept 30 days, weekly 90 days).
+but Longhorn, the Vault servers, calico and the LDAP outpost. Backups go
+to the NAS (192.168.1.99): Longhorn to NFS at 00:00 (7 kept), Velero to
+Garage (MinIO until 2026-10-26) at 01:00 (daily kept 30 days, weekly 90
+days), and CNPG to Garage (a base backup at 03:00, the WAL every 5
+minutes, 30 days).
 Gatus on s1 (node001, Hetzner) and s2 (home) watch each other and alert
 through ntfy; s2 also gets an hourly report from inside the cluster on the
 backups, Vault's seal, the Argo CD apps and the CNPG clusters.
-Every app namespace has NetworkPolicies, and CI renders and validates all
+Every namespace with pods but kube-system has NetworkPolicies, and CI
-apps on every PR. node001 runs mail and tunnels the public hosts home.
+renders and validates all apps on every PR. node001 runs mail and tunnels
+the public hosts home.
 
 ## Follow-ups from this week
 
@@ -26,19 +31,28 @@ docs/longhorn-second-ssd.md:
 - [x] Read `~/iosample.log` on opti1: the worst stretch was not the
   backups but 8.5 hours of Longhorn merging immich-claim's deleted
   snapshots (2026-10-24 14:30-23:15), see docs/upgrade-k8s-1.33.md
-- [ ] From 2026-10-27, if immich runs without problems: delete
+- [x] From 2026-10-27, if immich runs without problems: delete
-  `immich/postgresql-cluster.yaml` (the PG14 cluster still runs, 20 GiB),
+  `immich/postgresql-cluster.yaml` (the PG14 cluster, 20 GiB), the
-  the Longhorn snapshots and backups `immich-pg14-pre-vchord` and
+  Longhorn snapshots and backups `immich-pg14-pre-vchord` and
-  `immich-pg17-pre-v3`, and both dumps in `~` on opti1. PostgreSQL 14
+  `immich-pg17-pre-v3`, and both dumps in `~` on opti1. Done 2026-10-28
-  reaches end of life on 2026-11-12
+  (PR #61), with the PG14 volume's other Longhorn backups.
+  `immich-pg17-pre-v3` only ever was a snapshot
 - [ ] The upgrade's dqlite and backend copies in `/root/` on opti1
   (`backend-*.tar`, `dqlite-pre-*`)
-- [ ] Log in to immich as a jeugdlid (quota 0)
+- [ ] Log in to immich as a jeugdlid (quota 0): no uploads, albums as
+  viewer, people searchable but not renamable
+- [x] Searching by face: people are per user in immich, and a partner's
+  library doesn't bring its people along, so since the personal accounts
+  only the group account could. share-sync now puts everyone in the
+  group account's cluster group and shares its people (immich 3.3's
+  person sharing; PR #59, 2026-10-28): 4430 people, 169 named, for each
+  of the 6 accounts, with write access for those who may upload
 - [ ] Back to 1.36/stable once it is at v1.36.5 or later (still v1.36.2 on
   2026-10-25). The snap is held, so microk8s gets no updates at all until
   someone refreshes it
 - [ ] Second SSD, docs/longhorn-second-ssd.md
-- [ ] Optional: `MACHINE_LEARNING_MODEL_REVISION=v2` for immich ML
+- [x] `MACHINE_LEARNING_MODEL_REVISION=v2` for immich ML (PR #39,
+  2026-10-25; the models are cached on a volume)
 
 ## Next, in order
 
@@ -82,7 +96,9 @@ friend.
   certificate; the admin API stays on the LAN. The cluster side (DNS,
   node001 tunnel, Certificate, HTTPRoute) goes in git. Needed: how much
   space he wants, free space on the NAS
-- [ ] Decide which copy of the immich library goes off-site (see 5)
+- [x] Which copy of the immich library goes off-site: Velero's, with the
+  rest of the bucket (decided 2026-10-26). It is most of the ~190 GB, so
+  ask him for room for it
 - [ ] A restore test that runs on a schedule. `backup-testing/` predates
   Envoy Gateway and still rewrites `*.hensen.io` and bypasses sol-auth.
   Planned on 2026-10-25, not run: a second Velero on the real bucket is
@@ -202,9 +218,23 @@ Findings from 2026-10-25:
   maintenance job pods) was fixed that afternoon by
   `velero/skip-emptydir-policy.yaml`.
 
-- [ ] Whether the immich library stays in Velero. It stays for now
+- [x] Whether the immich library stays in Velero: it stays, since
-  (2026-10-25); decide again with the off-site copy (2)
+  Velero's bucket is what goes off-site (2; decided 2026-10-26)
-- [ ] CNPG backups through the barman cloud plugin
+- [x] CNPG backups through the barman cloud plugin (PR #56, merged
+  2026-10-26 22:43 UTC): authentik's and immich's PostgreSQL 17 back up
+  to their own bucket in Garage (`cnpg-authentik`, `cnpg-immich`), a base
+  backup at 03:00 UTC and the WAL every 5 minutes, kept 30 days. Adding
+  the sidecar restarted both instances; authentik was down from 22:44 to
+  22:48, since the old pod waited for its connections to close. The first
+  base backups: 46 MB in 29 s and 776 MB in 3.5 minutes. immich's
+  `immediate` backup ran before the plugin was up and failed; a manual one
+  (`immich-postgres-17-first`) went through. The 03:00 base backups of
+  10-27 and 10-28 completed for both; the failed backup is deleted
+- [ ] Restore one of them into a new Cluster (`bootstrap.recovery` from
+  the ObjectStore, to a point in time) to see that it works. After that,
+  decide whether Velero still needs to copy the CNPG data directories
+- [ ] Take `cnpg-authentik` and `cnpg-immich` along in the off-site copy
+  (2)
 - [x] Velero's target, MinIO on the NAS, is no longer maintained: no
   community images since October 2025, maintenance mode since December
   2025, and the repository is archived. Since 2026-10-26 Velero backs up
@@ -225,11 +255,13 @@ Findings from 2026-10-25:
   MinIO still has the kopia repositories of namespaces that are gone
   (~156 GB: nextcloud, ttrss, whisper-priv, whisper-webui, blog, ghost,
   mbg-wordpress, sogo, sogo-mail, outrun-game, owntracks); decide first
-  whether any of them is worth keeping
+  whether any of them is worth keeping. The dailies of 10-27 and 10-28
+  on Garage completed without errors
 - [ ] Garage's SQLite WAL (`meta/db.sqlite-wal`) was 1.7 GB on
   2026-10-26 next to a 270 MB database, and going by the folders in
   `meta/snapshots`, each 6-hourly metadata snapshot takes about 70
-  minutes. Not growing; check again after a week of normal use
+  minutes. Not growing; check again after a week of normal use. 1.8 GB
+  on 2026-10-28
 
 ### 6. Monitoring
 
@@ -298,10 +330,24 @@ Counted 2026-10-25 over the 73 running containers:
   OIDC logins of Vault and FreshRSS reach authentik, all 24
   ExternalSecrets refresh, raft intact; Gitea no longer reaches the NAS
   or Vault, and a pod in `default` no longer reaches Vault or Gitea
-- [ ] The infrastructure namespaces (cert-manager, cnpg-system,
+- [x] Certificates couldn't renew in those 11 namespaces: Envoy may only
+  reach each app's own port, and cert-manager's HTTP-01 solver listens on
+  8089. ynab's renewal hung from 2026-10-26 17:35 UTC; a policy for the
+  solver pods, applied by hand in actual-budget that evening, let it
+  through within 20 seconds (now valid until 2027-01-24). PR #51 adds it
+  to all 11 (merged 2026-10-26 22:27 UTC). Without it the next renewals
+  would hang too; the first real test is gitea's on 11-15, then
+  bolletjestrui on 11-18 and the rest from 12-03
+- [x] The infrastructure namespaces (cert-manager, cnpg-system,
   external-secrets, envoy-gateway-system, velero, mariadb-system, dns,
-  default) still have none; `default` can reach the LAN and everything
+  default), merged 2026-10-26: PR #54 for envoy-gateway-system (22:31
-  without a policy
+  UTC), PR #53 for the rest (22:42 UTC; one commit per namespace, and
+  `default` denies everything). Checked afterwards: all 16 public hosts
+  answer as before; server-side dry runs reach the CNPG, ESO,
+  cert-manager and MariaDB webhooks (all fail closed); Velero's location
+  Available, every ExternalSecret synced, the dns-sync job ran, no errors
+  from the operators after the barman plugin had started. Velero's policy
+  only lets it reach Garage, not MinIO's port 9000
 - [x] Requests: PR #33, merged 2026-10-25 11:23 UTC, for everything that
   had none except the Vault server, from `kubectl top` every 5 minutes
   from 03:13 to 08:40 UTC (highest memory plus 10%, CPU at p95, 10m
@@ -312,8 +358,19 @@ Counted 2026-10-25 over the 73 running containers:
   requested. Still without requests: Longhorn (22 containers, its chart),
   the Vault servers, the LDAP outpost, immich's PG14 cluster and three in
   kube-system
-- [ ] Memory limits, once there are daytime numbers: the night ones miss
+- [x] Memory limits: PR #58, for everything that got requests in #33.
-  immich's machine learning at work (it idles at 234Mi)
+  About twice the highest memory from `kubectl top` every 5 minutes from
+  2026-10-26 22:29 to 2026-10-27 21:55 UTC or the night of #33, rounded
+  up; no CPU limits. No OOM kills that day. immich's machine learning
+  sat idle all day (240Mi), so on 2026-10-28 a test image went through
+  CLIP, face recognition and OCR inside the container: 884Mi with all
+  three loaded, limit 2Gi
+- [ ] authentik's server fails its liveness probe while its database is
+  away (`/-/health/live/` answers 500 when PostgreSQL is down), so each
+  database restart gets the server killed too, as on 2026-10-27 for
+  PostgreSQL 17.11. The login only comes back once the server has
+  started again. A liveness probe that doesn't touch the database would
+  leave it at not ready
 - [x] WordPress (scoutingmbg.nl) was OOMKilled at 08:27:59 UTC on
   2026-10-25, its third restart: a crawler asked for 88 pages and assets
   within a few seconds (normally ~6 a minute), Apache started a PHP worker
@@ -334,10 +391,14 @@ Counted 2026-10-25 over the 73 running containers:
   `litespeed-cache` only caches pages on a LiteSpeed server, so it does
   little here, `wordpress-importer` is usually for a one-off import, and
   `code-snippets` runs PHP from the dashboard
-- [ ] Rate limiting per visitor in Envoy needs the client's address, which
+- [ ] Rate limiting per visitor in Envoy, on hold. Since PR #50
-  node001 loses: it passes TLS through, so every request arrives from
+  (2026-10-26) node001 passes the visitor's address on with the PROXY
-  46.224.26.65. That needs the PROXY protocol from node001's nginx to Envoy
+  protocol, and the apps see it: authentik logs it, Apache logs it for
-  (ClientTrafficPolicy `enableProxyProtocol`)
+  WordPress, so Wordfence's lockout after 20 failed logins now hits one
+  visitor instead of everyone. Envoy Gateway's local rate limit can't
+  count per address (no `Distinct`); that takes its global rate limit,
+  with a Redis and a ratelimit service next to it. Not worth two more
+  components on the one node while the apps' own limits work
 - [x] Probes for gitea (`/api/healthz`), freshrss (TCP, since
   mod_auth_openidc answers 401 to non-browsers) and serpbear (`/login`)
   (2026-10-25). Liveness only after 5 failures 30s apart, to ride out the
@@ -375,7 +436,15 @@ every PR.
   fail the run; tested with a planted Secret and an untagged image), plus
   PR annotations for containers without requests or probes. freshrss's
   `alpine/git` init container had no tag; now `v2.52.0`
-- [ ] Automerge patch versions of low-risk apps now that CI runs
+- [x] Automerge patch versions of low-risk apps now that CI runs: PR #52,
+  merged 2026-10-26 (ntfy, gatus, freshrss, serpbear, actual-budget and
+  CI's own tools; three days after the release, only on weekday evenings
+  and at the weekend, when the cluster can be reached). Nothing has
+  merged itself yet by 2026-10-28
+- [x] MariaDB stays on 11.4 (PR #60, 2026-10-28): Renovate offered 11.8.9
+  as a minor update (#42, closed), but 11.8 can't be rolled back and has
+  community support until June 2028, against May 2029 for 11.4. Renovate
+  now only offers 11.4 patches; the next LTS step, 12.3, is by hand
 - [x] Renovate listed external-snapshotter twice (#15 and #24): its
   `argocd` manager and a custom `github-tags` regex both matched. The
   regex is gone; Renovate should close #15 by itself
@@ -432,11 +501,10 @@ step after that, if downtime starts to matter.
   there and the volumes hold nothing (0 bytes in every backup). Drop both
   mounts and the two PVCs, or make them ConfigMap files if custom config
   is wanted
-- [ ] `kubectl delete namespace secrets`, empty since its app went
+- [x] The namespace `secrets` was already gone on 2026-10-26
-- [ ] 41 Failed Velero backups: the dailies from 2026-09-26 to 10-21 and
+- [x] The Failed Velero backups (the dailies from 2026-09-26 to 10-21 and
-  the weeklies since 08-02. Velero doesn't restore from a Failed backup;
+  the weeklies since 08-02; 40 by then, one had expired), deleted through
-  left alone they expire by 2027-01-16.
+  Velero on 2026-10-26. Left: 13 Completed, 3 PartiallyFailed
-  `kubectl -n velero exec deploy/velero -c velero -- /velero backup delete $(kubectl -n velero get backups.velero.io -o json | jq -r '.items[] | select(.status.phase=="Failed") | .metadata.name' | xargs) --confirm`
 - [ ] The mariadb-operator Backup `mbgwp-pre-migration-backup`, a dump
   from 2026-02-22 on its own 10 GiB volume (which has a Longhorn backup
   from 2026-10-22). The PVC has no owner, so it goes separately:
@@ -468,8 +536,8 @@ critical / high:
 
 | Image | Crit | High | What to do |
 |---|---|---|---|
-| tensorchord/cloudnative-pgvecto.rs:14-v0.2.1 | 38 | 342 | immich's old PG14 cluster; goes with the cleanup from 2026-10-27 |
+| tensorchord/cloudnative-pgvecto.rs:14-v0.2.1 | 38 | 342 | immich's old PG14 cluster; deleted 2026-10-28 |
-| cloudnative-pg/postgresql:17.4 | 27 | 260 | authentik's database, a Debian 11 image. Renovate now offers 17.7; a `17.x-standard-trixie` image is more current |
+| cloudnative-pg/postgresql:17.4 | 27 | 260 | authentik's database, a Debian 11 image. 17.11 since PR #40 (2026-10-27); not scanned again, a `17.x-standard-trixie` image is more current |
 | tensorchord/cloudnative-vectorchord:17.11-1.1.1 | 17 | 136 | immich's database; look for a newer tag |
 | longhornio/longhorn-instance-manager:v1.7.3 | 11 | 159 | still serves 16 volumes, see below |
 | actualbudget/actual-server:26.6.0 | 8 | 83 | chart 1.8.9; Renovate now covers it |
@@ -479,7 +547,7 @@ critical / high:
 | rubenhensen/bolletjestrui:main-10268f1 | 5 | 76 | rebuild on a current base image |
 | coredns/coredns:1.10.1 | 5 | 70 | see below |
 | dexidp/dex:v2.45.1, argocd v3.5.6, redis 8.2.3 | 4, 1, 2 | | come with Argo CD |
-| mariadb:11.4.5 | 4 | 65 | the newest 11.4 patch; Renovate now covers it |
+| mariadb:11.4.5 | 4 | 65 | 11.4.13 since PR #60; Renovate stays on 11.4 (it offered 11.8, PR #42) |
 | calico v3.29.3 | 2-3 | ~50 | comes with microk8s |
 | goauthentik ldap / server 2026.8.5 | 3, 1 | 74, 65 | the next authentik patch |
 

docs/upgrade-k8s-1.33.md

@@ -169,8 +169,9 @@ pgvecto.rs. Done 2026-10-24, about an hour of immich downtime:
   v2.4.1 refused to start (it only accepts VectorChord <0.6); v2.5.0 and up
   accept <2, so immich went to v2.7.5, the last 2.x. It rebuilt both
   indexes as `vchordrq` in about 1.5 minutes
-- [ ] After a few days without problems: delete the old cluster
+- [x] After a few days without problems: delete the old cluster
   (`immich/postgresql-cluster.yaml`), the snapshot/backup and the dump.
+  Done 2026-10-28 (PR #61), with the PG14 volume's Longhorn backups.
   Rolling back means reverting the old volume to the snapshot, since its
   embeddings are now `real[]` without pgvecto.rs
 - [x] immich 3.x: v2.7.5 → v3.3.5 with chart 0.10.3 → 0.13.4 on
@@ -189,8 +190,8 @@ pgvecto.rs. Done 2026-10-24, about an hour of immich downtime:
   (v3.3.0), for faster ML with less memory; numerically the same models,
   so no jobs to rerun. The model cache is an emptyDir, so the one-off
   preparation repeats on every restart
-- [ ] After a few days without problems: delete the `immich-pg17-pre-v3`
+- [x] After a few days without problems: delete the `immich-pg17-pre-v3`
-  snapshot and dump together with the PG14 leftovers above
+  snapshot and dump together with the PG14 leftovers above (2026-10-28)
 
 ## k8s-dqlite write stalls (found 2026-10-22)
 

external-secrets/networkpolicy.yaml

@@ -0,0 +1,68 @@
+# External Secrets' three pods reach each other, DNS and the API server
+# (the node, port 16443); the controller also reads Vault
+# (vault-active.vault.svc:8200, which vault/networkpolicy.yaml lets it
+# into). Besides their own namespace they take connections only from the
+# node: the API server calls the webhook from there and the kubelet probes
+# come from there too (10.1.7.64 is the node's Calico VXLAN address,
+# projectcalico.org/IPv4VXLANTunnelAddr). The Password generator behind
+# gatus/backup-check-token needs no network. The chart's Application
+# (apps/templates/external-secrets-helm.yaml) only installs the chart, so
+# this folder gets an Application of its own, external-secrets-folder.
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: default
+spec:
+  podSelector: {}
+  policyTypes:
+    - Ingress
+    - Egress
+  ingress:
+    - from:
+        - podSelector: {}
+    - from:
+        - ipBlock:
+            cidr: 192.168.1.204/32
+        - ipBlock:
+            cidr: 10.1.7.64/32
+  egress:
+    - to:
+        - podSelector: {}
+    - to:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: kube-system
+          podSelector:
+            matchLabels:
+              k8s-app: kube-dns
+      ports:
+        - protocol: UDP
+          port: 53
+        - protocol: TCP
+          port: 53
+    - to:
+        - ipBlock:
+            cidr: 192.168.1.204/32
+      ports:
+        - port: 16443
+---
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: controller-to-vault
+spec:
+  podSelector:
+    matchLabels:
+      app.kubernetes.io/name: external-secrets
+  policyTypes:
+    - Egress
+  egress:
+    - to:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: vault
+          podSelector:
+            matchLabels:
+              app.kubernetes.io/name: vault
+      ports:
+        - port: 8200

freshrss/freshrss-deployment.yaml

@@ -23,7 +23,7 @@ spec:
         io.kompose.service: freshrss
     spec:
       initContainers:
-        - image: alpine/git:v2.52.0
+        - image: alpine/git:v2.54.0
           name: git
           command:
           - git
@@ -74,7 +74,7 @@ spec:
               value: Europe/Amsterdam
           # 1.30.3 ships PHP 8.4.27, which fixes CVE-2026-109618 and
           # CVE-2026-109619.
-          image: freshrss/freshrss:1.30.3
+          image: freshrss/freshrss:1.30.4
           name: freshrss
           ports:
             - containerPort: 80
@@ -83,6 +83,8 @@ spec:
             requests:
               cpu: 50m
               memory: 176Mi
+            limits:
+              memory: 512Mi
           # TCP, because mod_auth_openidc answers 401 to anything that isn't a
           # browser.
           readinessProbe:

freshrss/networkpolicy.yaml

@@ -56,3 +56,29 @@ spec:
               - 169.254.0.0/16
               - 172.16.0.0/12
               - 192.168.0.0/16
+---
+# cert-manager's HTTP-01 solver, which runs only while a certificate
+# renews: Let's Encrypt and cert-manager's own self check reach it on 8089
+# through Envoy. The policies above let Envoy reach the app's port only,
+# which left the solver unreachable (found 2026-10-26, when ynab's renewal
+# hung).
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: acme-http01-solver
+spec:
+  podSelector:
+    matchLabels:
+      acme.cert-manager.io/http01-solver: "true"
+  policyTypes:
+    - Ingress
+  ingress:
+    - from:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: envoy-gateway-system
+          podSelector:
+            matchLabels:
+              app.kubernetes.io/name: envoy
+      ports:
+        - port: 8089

gateway/networkpolicy.yaml

@@ -0,0 +1,120 @@
+# Envoy Gateway: the controller and the proxy it runs (envoyproxy.yaml).
+# Both reach each other (the proxy gets its config from the controller on
+# 18000) and DNS, and take connections from each other and the node: the
+# API server calls the controller's topology webhook (9443) from there and
+# the kubelet probes come from there too (10.1.7.64 is the node's Calico
+# VXLAN address, projectcalico.org/IPv4VXLANTunnelAddr). Nothing scrapes
+# the metrics on 19001.
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: default
+  namespace: envoy-gateway-system
+spec:
+  podSelector: {}
+  policyTypes:
+    - Ingress
+    - Egress
+  ingress:
+    - from:
+        - podSelector: {}
+    - from:
+        - ipBlock:
+            cidr: 192.168.1.204/32
+        - ipBlock:
+            cidr: 10.1.7.64/32
+  egress:
+    - to:
+        - podSelector: {}
+    - to:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: kube-system
+          podSelector:
+            matchLabels:
+              k8s-app: kube-dns
+      ports:
+        - protocol: UDP
+          port: 53
+        - protocol: TCP
+          port: 53
+---
+# The proxy holds the node's ports 80 and 443 (10080 and 10443 inside) and
+# takes connections there from anywhere: the internet through node001 and
+# the router, the LAN, and pods that call a public hostname. It reaches
+# every app it routes to (pods in any namespace; their own policies decide
+# which port), Home Assistant on 192.168.1.202:8123 (outside the cluster,
+# home-assistant/), and the internet on 443 for Longhorn's OIDC login
+# (authentik's token endpoint, at its public address).
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: proxy
+  namespace: envoy-gateway-system
+spec:
+  podSelector:
+    matchLabels:
+      app.kubernetes.io/name: envoy
+      app.kubernetes.io/component: proxy
+  policyTypes:
+    - Ingress
+    - Egress
+  ingress:
+    - ports:
+        - port: 10080
+        - port: 10443
+  egress:
+    - to:
+        - namespaceSelector: {}
+    - to:
+        - ipBlock:
+            cidr: 192.168.1.202/32
+      ports:
+        - port: 8123
+    - to:
+        - ipBlock:
+            cidr: 0.0.0.0/0
+            except:
+              - 10.0.0.0/8
+              - 100.64.0.0/10
+              - 169.254.0.0/16
+              - 172.16.0.0/12
+              - 192.168.0.0/16
+      ports:
+        - port: 443
+---
+# Everything but the proxy (the controller, and the chart's certgen job on
+# upgrades) reaches the API server (the node, port 16443). The controller
+# also fetches the OIDC discovery document of Longhorn's SecurityPolicy
+# from authentik's public address.
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: control-plane
+  namespace: envoy-gateway-system
+spec:
+  podSelector:
+    matchExpressions:
+      - key: app.kubernetes.io/name
+        operator: NotIn
+        values:
+          - envoy
+  policyTypes:
+    - Egress
+  egress:
+    - to:
+        - ipBlock:
+            cidr: 192.168.1.204/32
+      ports:
+        - port: 16443
+    - to:
+        - ipBlock:
+            cidr: 0.0.0.0/0
+            except:
+              - 10.0.0.0/8
+              - 100.64.0.0/10
+              - 169.254.0.0/16
+              - 172.16.0.0/12
+              - 192.168.0.0/16
+      ports:
+        - port: 443

gateway/proxy-protocol.yaml

@@ -0,0 +1,23 @@
+# node001 passes TLS for the tunneled hosts through without opening it, so
+# without this every request arrives from 46.224.26.65. Wordfence locks that
+# address out after 20 failed logins (2026-10-26), and with it everyone who
+# logs in to scoutingmbg.nl. node001 puts a PROXY header with the visitor's
+# address in front of each connection; Envoy takes the address from there
+# and adds it to X-Forwarded-For.
+#
+# Optional, because s2 and ntfy2 point straight at home and their clients
+# send no header, and neither does node001's proxy on port 80. The price:
+# whoever connects to the home address directly can send a header of their
+# own and claim any address.
+apiVersion: gateway.envoyproxy.io/v1alpha1
+kind: ClientTrafficPolicy
+metadata:
+  name: proxy-protocol
+  namespace: envoy-gateway-system
+spec:
+  targetRefs:
+  - group: gateway.networking.k8s.io
+    kind: Gateway
+    name: homelab
+  proxyProtocol:
+    optional: true

gatus/README.md

@@ -62,7 +62,7 @@ node, both PowerDNS Authoritative. PowerDNS LUA records health-check
 addresses and only return the healthy ones:
 
 ```
-@  30  IN  LUA  A  "ifurlup('https://status.rubenhensen.nl/health', {{'46.224.26.65'}, {'62.41.86.27'}})"
+@  30  IN  LUA  A  "ifurlup('https://status.rubenhensen.nl/health', {{'46.224.26.65'}, {'62.41.86.53'}})"
 ```
 
 The double braces make two groups. PowerDNS returns the first group that has
@@ -85,7 +85,7 @@ Steps:
 
 1. Parent zone (`dns/domains/rubenhensen.nl.yaml`, synced to TransIP): add
    `status NS ns-hetzner.rubenhensen.nl.` and `status NS ns-home.rubenhensen.nl.`,
-   plus A records for `ns-hetzner` (46.224.26.65) and `ns-home` (62.41.86.27).
+   plus A records for `ns-hetzner` (46.224.26.65) and `ns-home` (62.41.86.53).
    First check that TransIP accepts NS records for a subdomain.
 2. Hetzner: run PowerDNS from node001.nix with `enable-lua-records=yes` and
    the bind backend. Use the same zone file as home. Open 53/udp and 53/tcp.

gatus/backup-check.yaml

@@ -256,7 +256,7 @@ spec:
               type: RuntimeDefault
           containers:
             - name: check
-              image: python:3.15.1-alpine3.24
+              image: python:3.15.2-alpine3.24
               command: ["python", "/script/check.py"]
               env:
                 - name: PYTHONDONTWRITEBYTECODE

gatus/gatus-deployment.yaml

@@ -16,7 +16,7 @@ spec:
     spec:
       containers:
         - name: gatus
-          image: twinproduction/gatus:v5.37.0
+          image: twinproduction/gatus:v5.38.0
           env:
             - name: GATUS_CONFIG_PATH
               value: /config/config.yaml

gatus/networkpolicy.yaml

@@ -83,3 +83,29 @@ spec:
             cidr: 192.168.1.204/32
       ports:
         - port: 16443
+---
+# cert-manager's HTTP-01 solver, which runs only while a certificate
+# renews: Let's Encrypt and cert-manager's own self check reach it on 8089
+# through Envoy. The policies above let Envoy reach the app's port only,
+# which left the solver unreachable (found 2026-10-26, when ynab's renewal
+# hung).
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: acme-http01-solver
+spec:
+  podSelector:
+    matchLabels:
+      acme.cert-manager.io/http01-solver: "true"
+  policyTypes:
+    - Ingress
+  ingress:
+    - from:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: envoy-gateway-system
+          podSelector:
+            matchLabels:
+              app.kubernetes.io/name: envoy
+      ports:
+        - port: 8089

gitea/gitea-deployment.yaml

@@ -20,7 +20,7 @@ spec:
         fsGroup: 1000
       containers:
         - name: gitea
-          image: gitea/gitea:28.2.0
+          image: gitea/gitea:28.3.0
           env:
             - name: USER_UID
               value: "1000"
@@ -54,6 +54,8 @@ spec:
             requests:
               cpu: 20m
               memory: 336Mi
+            limits:
+              memory: 1Gi
           # /api/healthz checks the database and the cache.
           readinessProbe:
             httpGet:

gitea/networkpolicy.yaml

@@ -55,3 +55,29 @@ spec:
               - 169.254.0.0/16
               - 172.16.0.0/12
               - 192.168.0.0/16
+---
+# cert-manager's HTTP-01 solver, which runs only while a certificate
+# renews: Let's Encrypt and cert-manager's own self check reach it on 8089
+# through Envoy. The policies above let Envoy reach the app's port only,
+# which left the solver unreachable (found 2026-10-26, when ynab's renewal
+# hung).
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: acme-http01-solver
+spec:
+  podSelector:
+    matchLabels:
+      acme.cert-manager.io/http01-solver: "true"
+  policyTypes:
+    - Ingress
+  ingress:
+    - from:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: envoy-gateway-system
+          podSelector:
+            matchLabels:
+              app.kubernetes.io/name: envoy
+      ports:
+        - port: 8089

immich/helm-immich.yaml

@@ -67,6 +67,8 @@ spec:
                       requests:
                         cpu: 10m
                         memory: 64Mi
+                      limits:
+                        memory: 256Mi
                     image:
                       repository: docker.io/valkey/valkey
                       tag: 9.1.2-alpine@sha256:48332870af354a799964c0012ae1194a0bf2bf894eb508f945810596dc2d8d11
@@ -93,6 +95,8 @@ spec:
                       requests:
                         cpu: 100m
                         memory: 1280Mi
+                      limits:
+                        memory: 3Gi
                     # Node trusts its built-in roots, which still include the
                     # ones above the compromised eMudhra intermediates. Use
                     # patched-ca-bundle through OpenSSL until Node ships a fix.
@@ -117,12 +121,14 @@ spec:
                     image:
                       repository: ghcr.io/immich-app/immich-machine-learning
                       pullPolicy: IfNotPresent
-                    # Idle at night (234Mi); the models take over a GiB when
+                    # Idle at 240Mi; 884Mi with CLIP, face recognition and
-                    # it works.
+                    # OCR loaded (2026-10-28, a test image through each).
                     resources:
                       requests:
                         cpu: 50m
                         memory: 1Gi
+                      limits:
+                        memory: 2Gi
                     env:
                       TRANSFORMERS_CACHE: /cache
                       HF_XET_CACHE: /cache/huggingface-xet

immich/networkpolicy.yaml

@@ -73,7 +73,9 @@ spec:
 ---
 # The CNPG operator checks each instance on its status port and connects to
 # PostgreSQL; the instance manager inside each pod watches its Cluster
-# through the API server (the node, port 16443).
+# through the API server (the node, port 16443). The barman cloud sidecar
+# sends the WAL and the base backups to Garage on the NAS
+# (postgresql-backup.yaml).
 apiVersion: networking.k8s.io/v1
 kind: NetworkPolicy
 metadata:
@@ -100,3 +102,34 @@ spec:
             cidr: 192.168.1.204/32
       ports:
         - port: 16443
+    - to:
+        - ipBlock:
+            cidr: 192.168.1.99/32
+      ports:
+        - port: 3900
+---
+# cert-manager's HTTP-01 solver, which runs only while a certificate
+# renews: Let's Encrypt and cert-manager's own self check reach it on 8089
+# through Envoy. The policies above let Envoy reach the app's port only,
+# which left the solver unreachable (found 2026-10-26, when ynab's renewal
+# hung).
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: acme-http01-solver
+spec:
+  podSelector:
+    matchLabels:
+      acme.cert-manager.io/http01-solver: "true"
+  policyTypes:
+    - Ingress
+  ingress:
+    - from:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: envoy-gateway-system
+          podSelector:
+            matchLabels:
+              app.kubernetes.io/name: envoy
+      ports:
+        - port: 8089

immich/postgresql-backup.yaml

@@ -0,0 +1,99 @@
+# Backups of immich's database (albums, faces, the photos' metadata; the
+# photos themselves are files on immich-claim) by CNPG itself, through the
+# barman cloud plugin (apps/templates/plugin-barman-cloud-helm.yaml): a base
+# backup every night, and every WAL segment as soon as it is full or 5
+# minutes old (archive_timeout). Both go to the bucket cnpg-immich in Garage
+# on the NAS (nas/garage), so the database can be restored to any moment of
+# the last 30 days. Velero copies the data directory while PostgreSQL runs,
+# which is not a consistent copy. Only the PostgreSQL 17 cluster; the 14
+# one is about to go.
+# The key is Garage's cnpg-immich, in Vault as kv/cnpg-immich-s3; it can
+# only reach this bucket.
+apiVersion: external-secrets.io/v1
+kind: ExternalSecret
+metadata:
+  name: immich-postgres-17-s3
+  namespace: immich
+spec:
+  secretStoreRef:
+    name: vault-backend
+    kind: ClusterSecretStore
+  refreshInterval: 15m
+  target:
+    name: immich-postgres-17-s3
+    creationPolicy: Owner
+    template:
+      engineVersion: v2
+      data:
+        ACCESS_KEY_ID: "{{ .access_key_id }}"
+        ACCESS_SECRET_KEY: "{{ .secret_access_key }}"
+        # Garage rejects requests signed for another region
+        REGION: garage
+  data:
+    - secretKey: access_key_id
+      remoteRef:
+        key: kv/cnpg-immich-s3
+        property: access_key_id
+    - secretKey: secret_access_key
+      remoteRef:
+        key: kv/cnpg-immich-s3
+        property: secret_access_key
+---
+apiVersion: barmancloud.cnpg.io/v1
+kind: ObjectStore
+metadata:
+  name: immich-postgres-17
+  namespace: immich
+  annotations:
+    # The CRD comes with the plugin's own app
+    argocd.argoproj.io/sync-options: SkipDryRunOnMissingResource=true
+spec:
+  retentionPolicy: 30d
+  configuration:
+    destinationPath: s3://cnpg-immich/
+    endpointURL: http://192.168.1.99:3900
+    s3Credentials:
+      accessKeyId:
+        name: immich-postgres-17-s3
+        key: ACCESS_KEY_ID
+      secretAccessKey:
+        name: immich-postgres-17-s3
+        key: ACCESS_SECRET_KEY
+      region:
+        name: immich-postgres-17-s3
+        key: REGION
+    wal:
+      compression: gzip
+    data:
+      compression: gzip
+  instanceSidecarConfiguration:
+    resources:
+      requests:
+        cpu: 10m
+        memory: 64Mi
+      limits:
+        memory: 512Mi
+    env:
+      # Newer boto3 adds checksums that S3-compatible stores can reject;
+      # the plugin's docs suggest sending them only when required.
+      - name: AWS_REQUEST_CHECKSUM_CALCULATION
+        value: when_required
+      - name: AWS_RESPONSE_CHECKSUM_VALIDATION
+        value: when_required
+---
+apiVersion: postgresql.cnpg.io/v1
+kind: ScheduledBackup
+metadata:
+  name: immich-postgres-17-daily
+  namespace: immich
+spec:
+  cluster:
+    name: immich-postgres-17
+  # 03:00 UTC, after Longhorn's and Velero's backups; immediate: one right
+  # away, so WAL archiving has a base backup to start from.
+  schedule: "0 0 3 * * *"
+  immediate: true
+  backupOwnerReference: self
+  method: plugin
+  pluginConfiguration:
+    name: barman-cloud.cloudnative-pg.io

immich/postgresql-cluster-17.yaml

@@ -13,14 +13,26 @@ spec:
     requests:
       cpu: 20m
       memory: 320Mi
+    limits:
+      memory: 1Gi
 
   postgresql:
     shared_preload_libraries:
       - "vchord.so"
 
+  # WAL archiving and base backups to Garage, see postgresql-backup.yaml
+  plugins:
+    - name: barman-cloud.cloudnative-pg.io
+      isWALArchiver: true
+      parameters:
+        barmanObjectName: immich-postgres-17
+
   # Filled once from the PostgreSQL 14 cluster (immich-postgres) after its
   # embedding columns were turned into real[] and pgvecto.rs was dropped,
-  # see docs/upgrade-k8s-1.33.md.
+  # see docs/upgrade-k8s-1.33.md. That cluster is gone since 2026-10-28, so
+  # creating this Cluster again fails here, on purpose: an empty database
+  # would start immich with an empty library. Restore from Garage instead
+  # (bootstrap.recovery, see postgresql-backup.yaml).
   bootstrap:
     initdb:
       database: immich

immich/postgresql-cluster.yaml

@@ -1,28 +0,0 @@
-apiVersion: postgresql.cnpg.io/v1
-kind: Cluster
-metadata:
-  name: immich-postgres
-  namespace: immich
-spec:
-  instances: 1
-
-  # Use TensorChord's CNPG-compatible image with pgvecto.rs v0.2.1
-  imageName: ghcr.io/tensorchord/cloudnative-pgvecto.rs:14-v0.2.1
-
-  postgresql:
-    shared_preload_libraries:
-      - "vectors.so"
-
-  bootstrap:
-    initdb:
-      database: immich
-      owner: immich
-      postInitSQL:
-        - CREATE EXTENSION IF NOT EXISTS vectors;
-        - CREATE EXTENSION IF NOT EXISTS cube;
-        - CREATE EXTENSION IF NOT EXISTS earthdistance CASCADE;
-        - ALTER USER immich WITH SUPERUSER;
-
-  storage:
-    size: 20Gi
-    storageClass: longhorn

immich/share-sync.yaml

@@ -8,6 +8,10 @@
 # - the group account's library is shared as a partner and shown in
 #   everyone's timeline, so search covers it (immich only searches your own
 #   and your timeline partners' photos);
+# - everyone is put in the group account's cluster group and gets its
+#   people (person sharing, immich 3.3): people are per user, and a partner
+#   library doesn't bring its people along. Those who may upload can name
+#   them, for everyone;
 # - the invitation notifications for albums older than a week are hidden,
 #   or a new member would get one per album;
 # - sessions of OAuth users older than SESSION_MAX_AGE_DAYS are ended, so
@@ -15,8 +19,11 @@
 #   Online.
 # Albums and partners go through the API. Joining other people's albums
 # (the API only lets an album's owner or editors share it), the timeline
-# flag (which only the receiving user can set), the notifications and the
+# flag and the cluster group (which only the receiving user can set or
-# sessions go straight to the database.
+# accept), sharing the people (the API wants person.update, which the
+# group's key lacks, and with password login off nobody can log in as the
+# group to add it), the notifications and the sessions go straight to the
+# database.
 apiVersion: v1
 kind: ConfigMap
 metadata:
@@ -24,6 +31,7 @@ metadata:
   namespace: immich
 data:
   sync.py: |
+    import collections
     import json
     import os
     import subprocess
@@ -129,6 +137,62 @@ data:
         'UPDATE partner SET "inTimeline" = true WHERE "sharedById" = :\'group\' AND NOT "inTimeline"',
         group=group_id,
     )
+    # What accepting an invitation to the group account's cluster group does
+    # (ClusterGroupService.acceptRequest): the person groups only the member
+    # has people in move along. Members with people in groups that others
+    # also have people in are left alone; that only happens after sharing.
+    for (email,) in sql(
+        """
+        BEGIN;
+        CREATE TEMP TABLE moving ON COMMIT DROP AS
+        SELECT u.id FROM "user" u, "user" g
+        WHERE g.id = :'group' AND u.id = ANY (:'members'::uuid[])
+          AND u."clusterGroupId" <> g."clusterGroupId"
+          AND NOT EXISTS (
+            SELECT 1 FROM person p
+            JOIN person o ON o."personGroupId" = p."personGroupId" AND o."ownerId" <> p."ownerId"
+            WHERE p."ownerId" = u.id
+          );
+        UPDATE person_group SET "clusterGroupId" = (SELECT "clusterGroupId" FROM "user" WHERE id = :'group')
+        WHERE id IN (SELECT p."personGroupId" FROM person p JOIN moving m ON m.id = p."ownerId");
+        DELETE FROM cluster_group_request
+        WHERE "userId" IN (SELECT id FROM moving)
+          AND "clusterGroupId" = (SELECT "clusterGroupId" FROM "user" WHERE id = :'group');
+        UPDATE "user" SET "clusterGroupId" = (SELECT "clusterGroupId" FROM "user" WHERE id = :'group')
+        WHERE id IN (SELECT id FROM moving)
+        RETURNING email;
+        COMMIT;
+        """,
+        group=group_id,
+        members="{" + ",".join(members) + "}",
+    ):
+        print(f"put {email} in the group's cluster group", flush=True)
+    # What PUT /people/users does for the group account: its people, with
+    # their names, in every member's people list and search filters. Write
+    # access (naming people, for everyone) for those who may upload.
+    shared = sql(
+        """
+        INSERT INTO person_user ("personGroupId", "sharedById", "sharedWithId", role)
+        SELECT p."personGroupId", p."ownerId", m.id, m.role::person_user_role_enum
+        FROM person p
+        JOIN "user" g ON g.id = p."ownerId"
+        CROSS JOIN unnest(:'ids'::uuid[], :'roles'::text[]) AS m(id, role)
+        JOIN "user" u ON u.id = m.id AND u."clusterGroupId" = g."clusterGroupId"
+        WHERE p."ownerId" = :'group' AND NOT p."isHidden"
+          AND NOT EXISTS (
+            SELECT 1 FROM person_user pu
+            WHERE pu."personGroupId" = p."personGroupId" AND pu."sharedById" = p."ownerId"
+              AND pu."sharedWithId" = m.id AND pu.role = m.role::person_user_role_enum
+          )
+        ON CONFLICT ("personGroupId", "sharedById", "sharedWithId") DO UPDATE SET role = EXCLUDED.role
+        RETURNING "sharedWithId", role
+        """,
+        group=group_id,
+        ids="{" + ",".join(wanted) + "}",
+        roles="{" + ",".join("read" if role == "viewer" else "write" for role in wanted.values()) + "}",
+    )
+    for (uid, role), count in sorted(collections.Counter(map(tuple, shared)).items()):
+        print(f"share {count} people with {members[uid]} ({role})", flush=True)
     sql(
         """
         UPDATE notification SET "deletedAt" = now()

mariadb-operator/kustomization.yaml

@@ -3,3 +3,4 @@ kind: Kustomization
 resources:
   - operator-crds-helm.yaml
   - operator-helm.yaml
+  - networkpolicy.yaml

mariadb-operator/networkpolicy.yaml

@@ -0,0 +1,54 @@
+# The mariadb-operator's three pods reach each other, DNS and the API
+# server (the node, port 16443); the operator also connects to MariaDB on
+# 3306 for the databases, users and grants it manages (mbgwp's
+# mariadb-from-operator lets it in). Besides their own namespace they take
+# connections only from the node: the API server calls the webhook from
+# there and the kubelet probes come from there too (10.1.7.64 is the node's
+# Calico VXLAN address, projectcalico.org/IPv4VXLANTunnelAddr). This
+# folder's Application deploys into argocd (the operator's own
+# Applications), hence the namespace here.
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: default
+  namespace: mariadb-system
+spec:
+  podSelector: {}
+  policyTypes:
+    - Ingress
+    - Egress
+  ingress:
+    - from:
+        - podSelector: {}
+    - from:
+        - ipBlock:
+            cidr: 192.168.1.204/32
+        - ipBlock:
+            cidr: 10.1.7.64/32
+  egress:
+    - to:
+        - podSelector: {}
+    - to:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: kube-system
+          podSelector:
+            matchLabels:
+              k8s-app: kube-dns
+      ports:
+        - protocol: UDP
+          port: 53
+        - protocol: TCP
+          port: 53
+    - to:
+        - ipBlock:
+            cidr: 192.168.1.204/32
+      ports:
+        - port: 16443
+    - to:
+        - namespaceSelector: {}
+          podSelector:
+            matchLabels:
+              app.kubernetes.io/name: mariadb
+      ports:
+        - port: 3306

mariadb-operator/operator-crds-helm.yaml

@@ -13,7 +13,7 @@ spec:
   source:
     chart: mariadb-operator-crds
     repoURL: https://mariadb-operator.github.io/mariadb-operator
-    targetRevision: 25.10.4
+    targetRevision: 26.10.2
   syncPolicy:
     automated:
       prune: true

mariadb-operator/operator-helm.yaml

@@ -13,13 +13,15 @@ spec:
   source:
     chart: mariadb-operator
     repoURL: https://mariadb-operator.github.io/mariadb-operator
-    targetRevision: 25.10.4
+    targetRevision: 26.10.2
     helm:
       values: |
         resources:
           requests:
             cpu: 10m
             memory: 48Mi
+          limits:
+            memory: 128Mi
         metrics:
           enabled: true
           serviceMonitor:
@@ -29,6 +31,8 @@ spec:
             requests:
               cpu: 10m
               memory: 32Mi
+            limits:
+              memory: 128Mi
           serviceMonitor:
             enabled: false
         certController:
@@ -36,6 +40,8 @@ spec:
             requests:
               cpu: 10m
               memory: 32Mi
+            limits:
+              memory: 128Mi
           serviceMonitor:
             enabled: false
   syncPolicy:

mbgwp/mariadb-cluster.yaml

@@ -8,7 +8,7 @@ spec:
     name: mbgwp-mariadb-credentials
     key: root-password
 
-  image: mariadb:11.4.5
+  image: mariadb:11.4.13
   imagePullPolicy: IfNotPresent
 
   port: 3306

mbgwp/networkpolicy.yaml

@@ -85,3 +85,29 @@ spec:
               kubernetes.io/metadata.name: mariadb-system
       ports:
         - port: 3306
+---
+# cert-manager's HTTP-01 solver, which runs only while a certificate
+# renews: Let's Encrypt and cert-manager's own self check reach it on 8089
+# through Envoy. The policies above let Envoy reach the app's port only,
+# which left the solver unreachable (found 2026-10-26, when ynab's renewal
+# hung).
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: acme-http01-solver
+spec:
+  podSelector:
+    matchLabels:
+      acme.cert-manager.io/http01-solver: "true"
+  policyTypes:
+    - Ingress
+  ingress:
+    - from:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: envoy-gateway-system
+          podSelector:
+            matchLabels:
+              app.kubernetes.io/name: envoy
+      ports:
+        - port: 8089

nas/garage/README.md

@@ -31,5 +31,12 @@ Velero's key is in Vault as `kv/velero-s3` (`access_key_id`,
 `velero/external-secret.yaml`. Without the cluster, `$g key info velero
 --show-secret` on the NAS shows it again.
 
+CNPG's backups (`authentik/postgresql-backup.yaml`,
+`immich/postgresql-backup.yaml`) have a bucket and a key per cluster, so
+neither database can touch the other's backups. Made on 2026-10-26 the
+same way as Velero's, with `cnpg-authentik` and `cnpg-immich` as both
+bucket and key name; the keys are in Vault as `kv/cnpg-authentik-s3` and
+`kv/cnpg-immich-s3`.
+
 Renovate updates the image tag in `compose.yaml`, but nothing deploys it:
 after merging, `docker compose pull && docker compose up -d` on the NAS.

nix-infra-machine/nodes/node001.nix

@@ -2,7 +2,7 @@
 let
   # Public IP of the home network where the k8s cluster ingress lives.
   # Keep in sync with dns/domains/rubenhensen.nl.yaml.
-  homeIP = "62.41.86.27";
+  homeIP = "62.41.86.53";
 
   # Hosts tunneled to the home k8s cluster. Each entry gets:
   #   * an SNI map entry for TCP passthrough on :443
@@ -173,12 +173,18 @@ in
     #   - mail.rubenhensen.nl (and anything else) → local stalwart on 8443
     #   - tunneled hosts → home cluster ingress on 443
     #   - s1 / ntfy1.rubenhensen.nl → local nginx on 8444 (Gatus, ntfy)
+    #
+    # Each connection starts with a PROXY header that carries the visitor's
+    # address, so the home cluster sees who is asking instead of this box
+    # (gateway/proxy-protocol.yaml in k8scd). nginx sends the header to
+    # every upstream or none, so the local ones go through a hop on
+    # 8445/8446 that takes it off again.
     streamConfig = ''
       map $ssl_preread_server_name $tunnel_upstream {
 ${sniMapEntries}
-        s1.rubenhensen.nl 127.0.0.1:8444;
+        s1.rubenhensen.nl 127.0.0.1:8446;
-        ntfy1.rubenhensen.nl 127.0.0.1:8444;
+        ntfy1.rubenhensen.nl 127.0.0.1:8446;
-        default 127.0.0.1:8443;
+        default 127.0.0.1:8445;
       }
 
       server {
@@ -186,6 +192,17 @@ ${sniMapEntries}
         listen [::]:443;
         proxy_pass $tunnel_upstream;
         ssl_preread on;
+        proxy_protocol on;
+      }
+
+      server {
+        listen 127.0.0.1:8445 proxy_protocol;
+        proxy_pass 127.0.0.1:8443;
+      }
+
+      server {
+        listen 127.0.0.1:8446 proxy_protocol;
+        proxy_pass 127.0.0.1:8444;
       }
     '';
 

ntfy/networkpolicy.yaml

@@ -65,3 +65,29 @@ spec:
               - 169.254.0.0/16
               - 172.16.0.0/12
               - 192.168.0.0/16
+---
+# cert-manager's HTTP-01 solver, which runs only while a certificate
+# renews: Let's Encrypt and cert-manager's own self check reach it on 8089
+# through Envoy. The policies above let Envoy reach the app's port only,
+# which left the solver unreachable (found 2026-10-26, when ynab's renewal
+# hung).
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: acme-http01-solver
+spec:
+  podSelector:
+    matchLabels:
+      acme.cert-manager.io/http01-solver: "true"
+  policyTypes:
+    - Ingress
+  ingress:
+    - from:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: envoy-gateway-system
+          podSelector:
+            matchLabels:
+              app.kubernetes.io/name: envoy
+      ports:
+        - port: 8089

ntfy/ntfy-deployment.yaml

@@ -16,7 +16,7 @@ spec:
     spec:
       containers:
         - name: ntfy
-          image: binwiederhier/ntfy:v2.34.0
+          image: binwiederhier/ntfy:v2.37.2
           args: ["serve"]
           env:
             - name: NTFY_BASE_URL

renovate.json

@@ -4,6 +4,7 @@
     "config:recommended",
     "customManagers:githubActionsVersions"
   ],
+  "timezone": "Europe/Amsterdam",
   "argocd": {
     "description": "Every app folder; Renovate's regexes are RE2, which has no lookahead, so the other folders are skipped through ignorePaths",
     "managerFilePatterns": [
@@ -81,13 +82,40 @@
       "description": "A PostgreSQL major upgrade needs pg_upgrade or a dump, not a tag bump",
       "matchDepNames": [
         "ghcr.io/cloudnative-pg/postgresql",
-        "ghcr.io/tensorchord/cloudnative-vectorchord",
+        "ghcr.io/tensorchord/cloudnative-vectorchord"
-        "ghcr.io/tensorchord/cloudnative-pgvecto.rs"
       ],
       "matchUpdateTypes": [
         "major"
       ],
       "enabled": false
+    },
+    {
+      "description": "MariaDB goes from one LTS to the next by hand, after a dump: there is no way back. 11.4 has community support until May 2029, longer than 11.8 (June 2028); the next step is 12.3 (June 2029). Within 11.4 the patches come through",
+      "matchDepNames": [
+        "mariadb"
+      ],
+      "allowedVersions": "<11.5"
+    },
+    {
+      "description": "Patch updates of apps that hold little state and are easy to roll back, and of CI's tools, merge themselves once CI passes: three days after the release, in the evening or at the weekend, when the cluster can be reached. Not the databases, operators, Vault, authentik, Longhorn, Velero, immich, Gitea or WordPress. Renovate merges itself; the repo has no branch protection for GitHub's auto-merge to wait on",
+      "matchFileNames": [
+        "ntfy/**",
+        "gatus/**",
+        "freshrss/**",
+        "serpbear/**",
+        "actual-budget/**",
+        ".github/workflows/**"
+      ],
+      "matchUpdateTypes": [
+        "patch"
+      ],
+      "minimumReleaseAge": "3 days",
+      "automerge": true,
+      "platformAutomerge": false,
+      "automergeSchedule": [
+        "* 17-21 * * 1-5",
+        "* 10-21 * * 0,6"
+      ]
     }
   ]
 }

serpbear/networkpolicy.yaml

@@ -54,3 +54,29 @@ spec:
               - 169.254.0.0/16
               - 172.16.0.0/12
               - 192.168.0.0/16
+---
+# cert-manager's HTTP-01 solver, which runs only while a certificate
+# renews: Let's Encrypt and cert-manager's own self check reach it on 8089
+# through Envoy. The policies above let Envoy reach the app's port only,
+# which left the solver unreachable (found 2026-10-26, when ynab's renewal
+# hung).
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: acme-http01-solver
+spec:
+  podSelector:
+    matchLabels:
+      acme.cert-manager.io/http01-solver: "true"
+  policyTypes:
+    - Ingress
+  ingress:
+    - from:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: envoy-gateway-system
+          podSelector:
+            matchLabels:
+              app.kubernetes.io/name: envoy
+      ports:
+        - port: 8089

serpbear/serpbear-deployment.yaml

@@ -51,7 +51,7 @@ spec:
               value: --use-openssl-ca
             - name: SSL_CERT_FILE
               value: /etc/ssl/certs/ca-certificates.crt
-          image: towfiqi/serpbear:3.1.0
+          image: towfiqi/serpbear:3.1.2
           name: serpbear
           ports:
             - containerPort: 3000
@@ -60,6 +60,8 @@ spec:
             requests:
               cpu: 10m
               memory: 160Mi
+            limits:
+              memory: 512Mi
           readinessProbe:
             httpGet:
               path: /login

vault/networkpolicy.yaml

@@ -66,3 +66,29 @@ spec:
               - 169.254.0.0/16
               - 172.16.0.0/12
               - 192.168.0.0/16
+---
+# cert-manager's HTTP-01 solver, which runs only while a certificate
+# renews: Let's Encrypt and cert-manager's own self check reach it on 8089
+# through Envoy. The policies above let Envoy reach the app's port only,
+# which left the solver unreachable (found 2026-10-26, when ynab's renewal
+# hung).
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: acme-http01-solver
+spec:
+  podSelector:
+    matchLabels:
+      acme.cert-manager.io/http01-solver: "true"
+  policyTypes:
+    - Ingress
+  ingress:
+    - from:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: envoy-gateway-system
+          podSelector:
+            matchLabels:
+              app.kubernetes.io/name: envoy
+      ports:
+        - port: 8089

vault/overrides.yaml

@@ -7,6 +7,8 @@ injector:
       requests:
          cpu: 10m
          memory: 32Mi
+      limits:
+         memory: 128Mi
    # The chart's default anti-affinity forbids two injector pods on one
    # node, so on this single node a rolling update can never start the new
    # pod while the old one runs. Recreate removes the old pod first, which

velero/networkpolicy.yaml

@@ -0,0 +1,52 @@
+# Velero, its node agent, and the pods it starts here (the data movers of
+# each pod volume backup or restore, kopia's maintenance jobs, the chart's
+# CRD upgrade job) reach each other, DNS, the API server (the node, port
+# 16443) and Garage on the NAS (192.168.1.99:3900, nas/garage). The node
+# agent reads volumes through its hostPath mount, not the network. Besides
+# their own namespace they take connections only from the node (kubelet
+# probes; 10.1.7.64 is the node's Calico VXLAN address,
+# projectcalico.org/IPv4VXLANTunnelAddr); nothing scrapes the metrics on
+# 8085. MinIO (the NAS, port 9000) is not allowed: going back to it means
+# adding its port here as well.
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: default
+spec:
+  podSelector: {}
+  policyTypes:
+    - Ingress
+    - Egress
+  ingress:
+    - from:
+        - podSelector: {}
+    - from:
+        - ipBlock:
+            cidr: 192.168.1.204/32
+        - ipBlock:
+            cidr: 10.1.7.64/32
+  egress:
+    - to:
+        - podSelector: {}
+    - to:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: kube-system
+          podSelector:
+            matchLabels:
+              k8s-app: kube-dns
+      ports:
+        - protocol: UDP
+          port: 53
+        - protocol: TCP
+          port: 53
+    - to:
+        - ipBlock:
+            cidr: 192.168.1.204/32
+      ports:
+        - port: 16443
+    - to:
+        - ipBlock:
+            cidr: 192.168.1.99/32
+      ports:
+        - port: 3900

velero/velero-helm.yaml

@@ -8,7 +8,7 @@ spec:
   source:
     chart: velero
     repoURL: https://vmware-tanzu.github.io/helm-charts
-    targetRevision: 12.2.1
+    targetRevision: 12.2.3
     helm:
       values: |
         # Velero builds its repository maintenance jobs from a Deployment
@@ -71,6 +71,8 @@ spec:
             requests:
               cpu: 10m
               memory: 64Mi
+            limits:
+              memory: 128Mi
         # 512Mi got the server OOMKilled 314 times, including halfway
         # through a backup of ~14.7k items.
         resources: