obs,loki: remove redundant values

renamed "home-assistant" folder

.github/dependabot.yml

@@ -13,6 +13,7 @@ updates:
       - "/docker/home/zigbee2mqtt"
       - "/docker/infrastructure/mail/roundcube"
       - "/docker/infrastructure/network/traefik"
+      - "/docker/infrastructure/observability"
       - "/docker/infrastructure/squid"
       - "/docker/infrastructure/sshportal"
       - "/docker/personal/fireshare"
@@ -43,7 +44,7 @@ updates:
     directories:
       - "/kubernetes/code/gitea"
       - "/kubernetes/code/harbor"
-      - "/kubernetes/home/home-assisant"
+      - "/kubernetes/home/home-assistant"
       - "/kubernetes/home/zigbee2mqtt"
       - "/kubernetes/infrastructure/authentik"
       - "/kubernetes/infrastructure/mail/stalwart"

.gitignore

@@ -9,3 +9,6 @@ terraform.tfstate.backup
 kubeconfig
 talosconfig
 
+# Ignore rendered Talos machine configs (may contain cluster secrets)
+p330.yaml
+

.swarmcd/stacks.yaml

@@ -38,6 +38,11 @@ traefik:
   branch: main
   compose_file: docker/infrastructure/network/traefik/docker-compose.yml
 
+observability:
+  repo: homeprod
+  branch: main
+  compose_file: docker/infrastructure/observability/docker-compose.yml
+
 squid:
   repo: homeprod
   branch: main

docker/infrastructure/observability/alloy.river

@@ -0,0 +1,71 @@
+// Grafana Alloy config for the docker host (replaces Promtail, EOL since
+// 2026-03-02). Tails Docker container logs via docker_sd discovery, attaches
+// container_name/compose_service/stream labels, and pushes them to the
+// in-cluster Loki at loki-internal.lan (LAN-only Traefik Ingress on port 80).
+// River config produced by `alloy convert -f promtail` from an equivalent
+// Promtail YAML (so component names and argument shapes are guaranteed
+// correct for Alloy v1.18).
+
+discovery.docker "docker" {
+  host             = "unix:///var/run/docker.sock"
+  refresh_interval = "5s"
+}
+
+loki.process "docker" {
+  forward_to = [loki.write.default.receiver]
+
+  // Parse docker json-file log lines
+  // ({"log":"...","stream":"stdout","time":"..."}).
+  stage.json {
+    expressions = {
+      log    = "log",
+      stream = "stream",
+    }
+  }
+  stage.labels {
+    values = {
+      stream = null,
+    }
+  }
+}
+
+discovery.relabel "docker" {
+  targets = []
+
+  // Strip leading / from container name.
+  rule {
+    source_labels = ["__meta_docker_container_name"]
+    regex         = "/(.*)"
+    target_label  = "container_name"
+  }
+  // Compose service name (if label is set).
+  rule {
+    source_labels = ["__meta_docker_container_label_com_docker_compose_service"]
+    target_label  = "compose_service"
+  }
+  // Log stream (stdout/stderr).
+  rule {
+    source_labels = ["__meta_docker_container_log_stream"]
+    target_label  = "stream"
+  }
+  // Mark the source so dashboards/alerts can distinguish the docker host.
+  rule {
+    target_label = "source"
+    replacement  = "docker-host"
+  }
+}
+
+loki.source.docker "docker" {
+  host             = "unix:///var/run/docker.sock"
+  targets          = discovery.docker.docker.targets
+  forward_to       = [loki.process.docker.receiver]
+  relabel_rules    = discovery.relabel.docker.rules
+  refresh_interval = "5s"
+}
+
+loki.write "default" {
+  endpoint {
+    url = "http://loki-internal.lan/loki/api/v1/push"
+  }
+  external_labels = {}
+}

docker/infrastructure/observability/docker-compose.yml

@@ -0,0 +1,81 @@
+# Docker-host collectors for the central observability stack on Kubernetes.
+# Managed by swarm-cd (see .swarmcd/stacks.yaml).
+#
+# node-exporter + cAdvisor run locally (scraped by vmagent below).
+# vmagent remote_writes metrics to vm-internal.lan (LAN-only Traefik Ingress) ->
+#   vmsingle-vm-victoria-metrics-k8s-stack.observability.svc.cluster.local:8428
+# alloy tails /var/lib/docker/containers + /var/log and pushes logs to
+#   loki-internal.lan -> loki.observability.svc.cluster.local:3100
+
+services:
+  node-exporter:
+    image: prom/node-exporter:v1.9.1
+    container_name: obs-node-exporter
+    network_mode: host
+    pid: host
+    command:
+      - "--path.rootfs=/host"
+    volumes:
+      - /:/host:ro,rslave
+    restart: unless-stopped
+    labels:
+      - "traefik.enable=false"
+
+  cadvisor:
+    image: gcr.io/cadvisor/cadvisor:v0.49.1
+    container_name: obs-cadvisor
+    ports:
+      - "8080"
+    volumes:
+      - /:/rootfs:ro
+      - /var/run:/var/run:ro
+      - /sys:/sys:ro
+      - /dev/disk/:/dev/disk:ro
+      - /var/lib/docker/:/var/lib/docker:ro
+    restart: unless-stopped
+    labels:
+      - "traefik.enable=false"
+
+  vmagent:
+    image: victoriametrics/vmagent:v1.131.0
+    container_name: obs-vmagent
+    # vmagent listens on 8429 (its own metrics), scrapes node-exporter:9100
+    # and cadvisor:8080, remote_writes to vm-internal.lan
+    command:
+      - "-promscrape.config=/etc/vmagent/vmagent.yml"
+      - "-remoteWrite.url=http://vm-internal.lan/api/v1/write"
+    volumes:
+      - type: bind
+        source: /root/homeprod/docker/infrastructure/observability/vmagent.yml
+        target: /etc/vmagent/vmagent.yml
+        read_only: true
+    depends_on:
+      - node-exporter
+      - cadvisor
+    restart: unless-stopped
+    labels:
+      - "traefik.enable=false"
+
+  alloy:
+    image: grafana/alloy:v1.18.0
+    container_name: obs-alloy
+    command:
+      - "run"
+      - "/etc/alloy/config.river"
+      - "--server.http.listen-addr=0.0.0.0:12345"
+    volumes:
+      - type: bind
+        source: /root/homeprod/docker/infrastructure/observability/alloy.river
+        target: /etc/alloy/config.river
+        read_only: true
+      - /var/lib/docker/containers:/var/lib/docker/containers:ro
+      - /var/log:/var/log:ro
+      - /var/run/docker.sock:/var/run/docker.sock:ro
+    restart: unless-stopped
+    labels:
+      - "traefik.enable=false"
+
+networks:
+  default:
+    name: proxy
+    external: true

docker/infrastructure/observability/vmagent.yml

@@ -0,0 +1,22 @@
+# vmagent scrape config for the docker host.
+# Scrapes the two local collectors, forwards everything to the central
+# VictoriaMetrics via the LAN-only Caddy route vm-internal.lan (Caddy proxies
+# it to vmsingle-vm-victoria-metrics-k8s-stack.observability.svc.cluster.local:8428).
+
+scrape_configs:
+  - job_name: docker-host-node
+    static_configs:
+      - targets:
+          - localhost:9100
+        labels:
+          # Distinguishes the docker host from K8s nodes in dashboards/alerts.
+          instance_source: docker-host
+          host: docker-r740
+
+  - job_name: docker-host-cadvisor
+    static_configs:
+      - targets:
+          - localhost:8080
+        labels:
+          instance_source: docker-host
+          host: docker-r740

infra/p330/kube/main.tf

@@ -0,0 +1,195 @@
+# Talos node for the P330 — joins the r740 "kube" cluster.
+terraform {
+  required_providers {
+    talos = {
+      source  = "siderolabs/talos"
+      version = "0.9.0"
+    }
+    null = {
+      source  = "hashicorp/null"
+      version = "3.2.3"
+    }
+  }
+}
+
+# Read the r740 kube module state to reuse the cluster secrets & endpoint.
+# The r740 module exposes: client_configuration, machine_secrets, cluster_name,
+# cluster_endpoint, kube_host.
+data "terraform_remote_state" "r740_kube" {
+  backend = var.r740_backend
+
+  config = var.r740_backend == "local" ? {
+    path = "${var.r740_state_path}/terraform.tfstate"
+  } : var.r740_backend_config
+}
+
+locals {
+  cluster_name     = data.terraform_remote_state.r740_kube.outputs.cluster_name
+  cluster_endpoint = data.terraform_remote_state.r740_kube.outputs.cluster_endpoint
+  machine_secrets  = data.terraform_remote_state.r740_kube.outputs.machine_secrets
+  client_config    = data.terraform_remote_state.r740_kube.outputs.client_configuration
+
+  # kubeconfig produced by the r740 kube module — used to wait for the node and
+  # apply labels/taints. There is no in-tree kubernetes provider here on
+  # purpose: managing a `kubernetes_node` resource conflicts with the node
+  # object that kubelet itself creates, so we use a null_resource with kubectl
+  # to wait + label + taint idempotently.
+  kubeconfig_path = "${var.r740_state_path}/kubeconfig"
+
+  # Network config: static if node_subnet is provided, otherwise Talos DHCPs.
+  static_network = var.node_subnet == null ? {} : {
+    interfaces = [{
+      interface = var.network_interface
+      addresses = [var.node_subnet]
+      routes    = var.node_gateway == null ? [] : [{ gateway = var.node_gateway }]
+    }]
+  }
+
+  network_patch = {
+    nameservers = var.nameservers
+  }
+  network_patch_merged = merge(local.network_patch, local.static_network)
+
+  machine_patch = {
+    install = {
+      image = var.installer_image
+      disk  = var.install_disk
+    }
+    network = merge(local.network_patch_merged, {
+      # Pin the Kubernetes node name. Talos otherwise auto-generates a hostname
+      # (e.g. "talos-8ec-vd1"), so the node registers with that random name
+      # instead of var.p330_node_name — and our label/taint null_resource waits
+      # for the wrong node. Setting machine.network.hostname fixes the node name.
+      hostname = var.p330_node_name
+    })
+    # Kernel modules required by Longhorn (iSCSI + ext4) — must match the
+    # control-plane nodes so Longhorn can schedule replicas on the failover node.
+    kernel = {
+      modules = [
+        { name = "iscsi_tcp" },
+        { name = "libiscsi" },
+        { name = "scsi_transport_iscsi" },
+        { name = "ext4" },
+      ]
+    }
+    sysctls = {
+      "fs.inotify.max_user_instances" = "1024"
+      "fs.inotify.max_user_watches"   = "1048576"
+    }
+    kubelet = {
+      # Keep the failover node from accumulating non-essential DaemonSet pods
+      # via the regular scheduler; the taint does the heavy lifting, this is
+      # belt-and-braces.
+      extraArgs = {
+        "register-with-taints" = "${var.failover_taint_key}=${var.failover_taint_value}:${var.failover_taint_effect}"
+      }
+    }
+  }
+}
+
+# Control-plane machine configuration. machine_type = "controlplane" makes
+# Talos generate a join config that runs the apiserver/controller-manager/
+# scheduler AND joins the existing etcd cluster as a new member (the cluster
+# was already bootstrapped by the r740 module's talos_machine_bootstrap).
+data "talos_machine_configuration" "p330" {
+  cluster_name     = local.cluster_name
+  machine_type     = "controlplane"
+  cluster_endpoint = local.cluster_endpoint
+  machine_secrets  = local.machine_secrets
+  config_patches = [
+    yamlencode({
+      machine = local.machine_patch
+    })
+  ]
+}
+
+# Rendered config is written to disk so it can also be applied manually with
+# `talosctl apply-config --nodes <p330_host> --file p330.yaml` if needed.
+resource "local_file" "p330_machine_config" {
+  filename = "${path.module}/p330.yaml"
+  content  = data.talos_machine_configuration.p330.machine_configuration
+}
+
+# Apply the machine config to the running (maintenance-mode) node over the
+# Talos API. Because the config patch contains a `machine.install` block, when
+# Talos receives this config on a node booted from the USB (maintenance) image
+# it installs itself to install.disk and reboots into the installed system.
+# For a controlplane node it then joins the existing etcd cluster as a new
+# member and runs the control-plane components; for a worker it just registers
+# via kubelet.
+resource "talos_machine_configuration_apply" "p330" {
+  client_configuration        = local.client_config
+  machine_configuration_input = data.talos_machine_configuration.p330.machine_configuration
+  node                        = var.p330_host
+  depends_on                  = [local_file.p330_machine_config]
+}
+
+# Emit a talosconfig scoped to this node for ad-hoc `talosctl` use.
+data "talos_client_configuration" "p330" {
+  cluster_name         = local.cluster_name
+  client_configuration = local.client_config
+  nodes                = [var.p330_host]
+}
+
+resource "local_file" "talosconfig" {
+  content    = data.talos_client_configuration.p330.talos_config
+  filename   = "${path.module}/talosconfig"
+  depends_on = [data.talos_client_configuration.p330]
+}
+
+# Wait for the node to register with Kubernetes (kubelet creates the Node
+# object after Talos installs and reboots), then label it and (re)apply the
+# failover taint. This is idempotent: kubectl exits 0 if the label/taint already
+# exists. The taint is also set via kubelet `register-with-taints`, so this
+# null_resource is a safety net for manual edits / drift.
+resource "null_resource" "p330_node_label_and_taint" {
+  triggers = {
+    node       = var.p330_node_name
+    key        = var.failover_taint_key
+    value      = var.failover_taint_value
+    effect     = var.failover_taint_effect
+    kubeconfig = local.kubeconfig_path
+  }
+
+  provisioner "local-exec" {
+    # Wait for the node to show up, then label + taint. The wait loop is bounded
+    # by kubectl --timeout; tune it via TF_LOG / re-run if the node is slow to
+    # join (a controlplane node must first complete the etcd join handshake).
+    command = <<-EOT
+      set -euo pipefail
+      KUBECONFIG="${local.kubeconfig_path}"
+      export KUBECONFIG
+      NODE="${var.p330_node_name}"
+
+      echo "Waiting for node $NODE to be registered (kubelet creates the Node object once Talos has installed, rebooted and joined etcd)..."
+      # kubectl wait --for=condition=Ready fails instantly with NotFound if the
+      # node object doesn't exist yet, so poll for existence first.
+      # /bin/sh (dash) has no $SECONDS, so count iterations with a bounded loop.
+      tries=240   # 240 * 5s = 20 minutes max
+      until kubectl get node "$NODE" >/dev/null 2>&1; do
+        tries=$((tries - 1))
+        if [ "$tries" -le 0 ]; then
+          echo "Timed out waiting for node $NODE to register." >&2
+          exit 1
+        fi
+        sleep 5
+      done
+      echo "Node $NODE registered. Waiting for it to become Ready..."
+
+      # Now wait for Ready (a controlplane node needs etcd joined + apiserver up).
+      kubectl wait --for=condition=Ready "node/$NODE" --timeout=20m || \
+        kubectl wait --for=jsonpath='{.status.conditions[?(@.reason=="KubeletReady")].status}'=True "node/$NODE" --timeout=20m
+
+      # Failover marker + taint (applied to both controlplane and worker nodes).
+      kubectl label --overwrite node "$NODE" homeprod.io/failover=true
+
+      # Apply the taint idempotently (kubectl taint --overwrite is a no-op if it exists).
+      kubectl taint --overwrite node "$NODE" \
+        "${var.failover_taint_key}=${var.failover_taint_value}:${var.failover_taint_effect}"
+
+      echo "Node $NODE ready, labeled and tainted for failover-only scheduling."
+    EOT
+  }
+
+  depends_on = [talos_machine_configuration_apply.p330]
+}

infra/p330/kube/variables.tf

@@ -0,0 +1,118 @@
+# Variables for the P330 Talos worker node that joins the r740 cluster.
+
+variable "p330_host" {
+  description = "Reachable IP/hostname of the P330 Talos node (for Talos API access)."
+  type        = string
+}
+
+variable "p330_node_name" {
+  description = "Kubernetes/Talos node name for the P330 (e.g. p330)."
+  type        = string
+  default     = "p330"
+}
+
+variable "r740_state_path" {
+  description = <<EOT
+Path to the Terraform state of the r740 kube module, used by terraform_remote_state
+to read the cluster secrets and endpoint so this node can join the existing cluster.
+Path is resolved by terraform_remote_state relative to the working directory where
+terraform runs (this module dir). The default points two levels up to the repo
+root and back down to the r740 kube module.
+EOT
+  type        = string
+  default     = "../../r740/kube"
+}
+
+variable "r740_backend" {
+  description = <<EOT
+Terraform backend type used by the r740 kube module.
+Set to "local" (default) when r740 uses a local tfstate file in its own directory,
+or the matching remote backend name ("s3", "remote", ...) if the r740 module uses
+a configured backend.
+EOT
+  type        = string
+  default     = "local"
+}
+
+variable "r740_backend_config" {
+  description = <<EOT
+Backend configuration map passed to terraform_remote_state when r740_backend is
+not "local". For a local backend this is ignored.
+EOT
+  type        = map(string)
+  default     = {}
+}
+
+variable "installer_image" {
+  description = <<EOT
+Talos installer image to use on the P330 (bare metal).
+Must be a **metal** Image Factory build that includes ixgbe.allow_unsupported_sfp=1
+in the kernel command line (sd-boot/UKI ignores machine.install.extraKernelArgs, so
+the param must be baked into the image). The default is a custom factory build
+(a18165114...).
+EOT
+  type        = string
+  default     = "factory.talos.dev/installer/a18165114f80c28601d05bc4ff1f6ea6d6b214882c5b9af7928aaf4d09741beb:v1.13.6"
+}
+
+variable "install_disk" {
+  description = "Block device path to install Talos on (e.g. /dev/sda, /dev/nvme0n1)."
+  type        = string
+  default     = "/dev/nvme0n1"
+}
+
+variable "node_subnet" {
+  description = <<EOT
+Static IPv4 address in CIDR notation for the P330 node (e.g. 10.1.2.132/24).
+Set to null to use DHCP. A static address is recommended for a failover node so
+DNS/affinity rules stay stable.
+EOT
+  type        = string
+  default     = "10.1.2.132/24"
+}
+
+variable "node_gateway" {
+  description = "IPv4 gateway for the P330 node. Ignored when node_subnet is null."
+  type        = string
+  default     = "10.1.2.1"
+}
+
+variable "network_interface" {
+  description = <<EOT
+Primary network interface name on the P330. Defaults to enp3s0f1 (the 10G Intel
+X520 NIC), which must be on the same L2/subnet as the r740 control plane so etcd
+peer traffic (TLS-verified against the r740's etcd cert SANs) doesn't cross a
+router. eno1 (1G) is left unconfigured.
+EOT
+  type        = string
+  default     = "enp3s0f1"
+}
+
+variable "nameservers" {
+  description = "DNS nameservers configured on the node (must work independently of kube)."
+  type        = list(string)
+  default     = ["10.1.2.148", "1.1.1.1"]
+}
+
+variable "failover_taint_key" {
+  description = "Taint key applied to the node to reserve it for failover workloads."
+  type        = string
+  default     = "dedicated"
+}
+
+variable "failover_taint_value" {
+  description = "Taint value applied to the node."
+  type        = string
+  default     = "failover"
+}
+
+variable "failover_taint_effect" {
+  description = "Taint effect applied to the node (NoSchedule / NoExecute)."
+  type        = string
+  default     = "NoSchedule"
+
+  validation {
+    condition     = contains(["NoSchedule", "PreferNoSchedule", "NoExecute"], var.failover_taint_effect)
+    error_message = "failover_taint_effect must be NoSchedule, PreferNoSchedule or NoExecute."
+  }
+}

infra/r740/kube/main.tf

@@ -40,7 +40,9 @@ data "talos_machine_configuration" "kube" {
     yamlencode({
       machine = {
         install = {
-          image = "factory.talos.dev/installer/ce4c980550dd2ab1b17bbf2b08801c7eb59418eafe8f279833297925d67c7515:v1.11.5"
+          # Image Factory image with iSCSI extension for Longhorn.
+          # Generated at https://factory.talos.dev — siderolabs/iscsi-tools + qemu-guest-agent
+          image = "factory.talos.dev/installer/dc7b152cb3ea99b821fcb7340ce7168313ce393d663740b791c36f6e95fc8586:v1.13.6"
         }
         network = {
           nameservers = [
@@ -53,6 +55,28 @@ data "talos_machine_configuration" "kube" {
         certSANs = [
           "${var.kube_host}", "${var.kube_hostname}"
         ]
+        # Kernel modules required by Longhorn (iSCSI + ext4)
+        kernel = {
+          modules = [
+            {
+              name = "iscsi_tcp"
+            },
+            {
+              name = "libiscsi"
+            },
+            {
+              name = "scsi_transport_iscsi"
+            },
+            {
+              name = "ext4"
+            },
+          ]
+        }
+        # Sysctls for Longhorn
+        sysctls = {
+          "fs.inotify.max_user_instances"  = "1024"
+          "fs.inotify.max_user_watches"    = "1048576"
+        }
       }
       cluster = {
         clusterName = "kube-${var.physical_hostname}"
@@ -106,6 +130,33 @@ output "kubeconfig" {
   value = talos_cluster_kubeconfig.kube.kubeconfig_raw
 }
 
+output "client_configuration" {
+  description = "Talos client configuration (sensitive) used to manage nodes."
+  sensitive   = true
+  value       = talos_machine_secrets.kube.client_configuration
+}
+
+output "machine_secrets" {
+  description = "Talos machine secrets (sensitive) used to generate node configs."
+  sensitive   = true
+  value       = talos_machine_secrets.kube.machine_secrets
+}
+
+output "cluster_name" {
+  description = "Name of the Talos cluster the worker joins."
+  value       = "kube-${var.physical_hostname}"
+}
+
+output "cluster_endpoint" {
+  description = "Endpoint (host:port) of the Talos/Kubernetes API on the cluster."
+  value       = "https://${var.kube_host}:6443"
+}
+
+output "kube_host" {
+  description = "Reachable IP/hostname of the control-plane node."
+  value       = var.kube_host
+}
+
 resource "local_file" "kubeconfig" {
     content     = "${talos_cluster_kubeconfig.kube.kubeconfig_raw}"
     filename = "${path.module}/kubeconfig"

infra/rpi4/kube/main.tf

@@ -0,0 +1,214 @@
+# Talos control-plane node for the Raspberry Pi 4 — joins the r740 "kube" cluster
+# as a third etcd member to restore quorum (2-of-3 majority). Unlike the p330
+# failover node, this node is tainted "quorum" so no user workloads are ever
+# scheduled on it; only essential DaemonSets (Cilium, etc.) that tolerate the
+# taint land here for cluster networking.
+#
+# Secret handling: the cluster machine secrets are provided via
+# var.machine_secrets_file (a local, gitignored JSON file in the provider's
+# machine_secrets format). They are consumed by EPHEMERAL resources and
+# WRITE-ONLY attributes so they never land in Terraform state. See
+# variables.tf and scripts/extract-talos-secrets.sh for how to produce the
+# file from the live r740 node.
+terraform {
+  required_providers {
+    talos = {
+      source  = "siderolabs/talos"
+      version = "0.11.0"
+    }
+    null = {
+      source  = "hashicorp/null"
+      version = "3.2.3"
+    }
+  }
+}
+
+locals {
+  # Load the machine secrets from the gitignored JSON file. This local is only
+  # ever referenced by ephemeral resources / write-only attributes, so the
+  # values are never persisted to state.
+  machine_secrets = jsondecode(file(var.machine_secrets_file))
+
+  # Network config: static if node_subnet is provided, otherwise Talos DHCPs.
+  # The rpi4 uses DHCP (node_subnet = null), so only nameservers are patched in.
+  static_network = var.node_subnet == null ? {} : {
+    interfaces = [{
+      interface = var.network_interface
+      addresses = [var.node_subnet]
+      routes    = var.node_gateway == null ? [] : [{ gateway = var.node_gateway }]
+    }]
+  }
+
+  network_patch = {
+    nameservers = var.nameservers
+  }
+  network_patch_merged = merge(local.network_patch, local.static_network)
+
+  machine_patch = {
+    install = {
+      image = var.installer_image
+      disk  = var.install_disk
+    }
+    network = local.network_patch_merged
+    # NOTE: no Longhorn iSCSI/ext4 kernel modules here. This is a quorum-only
+    # node: the quorum taint keeps user workloads (and Longhorn replicas) off
+    # it, so the storage stack is not needed. Essential DaemonSets such as
+    # Cilium still run here for cluster networking and tolerate the taint.
+    sysctls = {
+      "fs.inotify.max_user_instances" = "1024"
+      "fs.inotify.max_user_watches"   = "1048576"
+    }
+    kubelet = {
+      # Register the node already tainted so the scheduler never admits user
+      # workloads even before the null_resource below runs. NoSchedule is
+      # sufficient: essential DaemonSets (Cilium, etc.) tolerate it, but no
+      # user pods are admitted.
+      extraArgs = {
+        "register-with-taints" = "${var.quorum_taint_key}=${var.quorum_taint_value}:${var.quorum_taint_effect}"
+      }
+    }
+  }
+}
+
+# --- Ephemeral resources: secrets never stored in state ---------------------
+#
+# talos_machine_configuration generates the control-plane join config from the
+# provided machine_secrets. The output (machine_configuration) is an ephemeral
+# value — it can only flow into write-only attributes or provisioners, never
+# into a persisted resource attribute.
+ephemeral "talos_machine_configuration" "rpi4" {
+  cluster_name     = var.cluster_name
+  machine_type     = "controlplane"
+  cluster_endpoint = var.cluster_endpoint
+  machine_secrets  = local.machine_secrets
+  config_patches = [
+    yamlencode({
+      machine = local.machine_patch
+    }),
+    # Pin the Kubernetes node name via a HostnameConfig document (Talos v1.13+).
+    # The old machine.network.hostname field conflicts with the default
+    # HostnameConfig document ("static hostname is already set"), so we use the
+    # document-based config with auto: off + an explicit hostname instead.
+    yamlencode({
+      apiVersion = "v1alpha1"
+      kind       = "HostnameConfig"
+      hostname   = var.rpi4_node_name
+      auto       = "off"
+    })
+  ]
+}
+
+# talos_client_configuration generates a Talos client config (talosconfig) from
+# the machine_secrets, scoped to the rpi4 node. Also ephemeral — used only to
+# drive the write-only client_configuration_wo on the apply resource.
+ephemeral "talos_client_configuration" "rpi4" {
+  cluster_name    = var.cluster_name
+  machine_secrets = local.machine_secrets
+  nodes           = [var.rpi4_host]
+}
+
+# --- Apply the config to the node (write-only attrs → no secrets in state) --
+#
+# machine_configuration_input_wo and client_configuration_wo are write-only:
+# Terraform uses them during apply but does NOT persist them to state. Only a
+# hash of the machine config (machine_configuration_hash) is stored, for drift
+# detection. Because the config patch contains a `machine.install` block, when
+# Talos receives this config on a node booted from the SD card (maintenance)
+# image it installs itself to install.disk and reboots into the installed
+# system. As a controlplane node it then joins the existing etcd cluster as a
+# new member and runs the control-plane components. With r740 + p330 + rpi4 the
+# etcd cluster reaches 3 members → 2-of-3 quorum.
+resource "talos_machine_configuration_apply" "rpi4" {
+  node                           = var.rpi4_host
+  client_configuration_wo        = ephemeral.talos_client_configuration.rpi4.client_configuration
+  machine_configuration_input_wo = ephemeral.talos_machine_configuration.rpi4.machine_configuration
+}
+
+# --- Write the rendered config to disk for manual use ----------------------
+#
+# local_file.content cannot accept an ephemeral value (it would persist to
+# state), so we use a null_resource local-exec provisioner instead —
+# provisioners do not persist their arguments to state. This writes rpi4.yaml
+# so the config can also be applied manually with
+# `talosctl apply-config --nodes <rpi4_host> --file rpi4.yaml` if needed.
+resource "null_resource" "rpi4_machine_config_file" {
+  triggers = {
+    # Re-run only when the (non-secret) inputs that shape the config change.
+    node            = var.rpi4_node_name
+    install_disk    = var.install_disk
+    installer_image = var.installer_image
+    taint           = "${var.quorum_taint_key}=${var.quorum_taint_value}:${var.quorum_taint_effect}"
+  }
+
+  provisioner "local-exec" {
+    command = <<-EOT
+      set -euo pipefail
+      cat > "${path.module}/rpi4.yaml" <<'YAMLEOF'
+${ephemeral.talos_machine_configuration.rpi4.machine_configuration}
+YAMLEOF
+      echo "Wrote ${path.module}/rpi4.yaml"
+    EOT
+  }
+
+  depends_on = [talos_machine_configuration_apply.rpi4]
+}
+
+# --- Wait for the node, then label + taint ---------------------------------
+#
+# Wait for the node to register with Kubernetes (kubelet creates the Node
+# object after Talos installs and reboots), then label it and (re)apply the
+# quorum taint. This is idempotent: kubectl exits 0 if the label/taint already
+# exists. The taint is also set via kubelet `register-with-taints`, so this
+# null_resource is a safety net for manual edits / drift. The kubeconfig path
+# is only used inside the provisioner (not persisted to state).
+resource "null_resource" "rpi4_node_label_and_taint" {
+  triggers = {
+    node       = var.rpi4_node_name
+    key        = var.quorum_taint_key
+    value      = var.quorum_taint_value
+    effect     = var.quorum_taint_effect
+    kubeconfig = var.kubeconfig_path
+  }
+
+  provisioner "local-exec" {
+    # Wait for the node to show up, then label + taint. The wait loop is bounded
+    # by kubectl --timeout; tune it via TF_LOG / re-run if the node is slow to
+    # join (a controlplane node must first complete the etcd join handshake).
+    command = <<-EOT
+      set -euo pipefail
+      KUBECONFIG="${var.kubeconfig_path}"
+      export KUBECONFIG
+      NODE="${var.rpi4_node_name}"
+
+      echo "Waiting for node $NODE to be registered (kubelet creates the Node object once Talos has installed, rebooted and joined etcd)..."
+      # kubectl wait --for=condition=Ready fails instantly with NotFound if the
+      # node object doesn't exist yet, so poll for existence first.
+      # /bin/sh (dash) has no $SECONDS, so count iterations with a bounded loop.
+      tries=240   # 240 * 5s = 20 minutes max
+      until kubectl get node "$NODE" >/dev/null 2>&1; do
+        tries=$((tries - 1))
+        if [ "$tries" -le 0 ]; then
+          echo "Timed out waiting for node $NODE to register." >&2
+          exit 1
+        fi
+        sleep 5
+      done
+      echo "Node $NODE registered. Waiting for it to become Ready..."
+
+      # Now wait for Ready (a controlplane node needs etcd joined + apiserver up).
+      kubectl wait --for=condition=Ready "node/$NODE" --timeout=20m || \
+        kubectl wait --for=jsonpath='{.status.conditions[?(@.reason=="KubeletReady")].status}'=True "node/$NODE" --timeout=20m
+
+      # Quorum marker + taint (applied to the controlplane node).
+      kubectl label --overwrite node "$NODE" homeprod.io/quorum=true
+
+      # Apply the taint idempotently (kubectl taint --overwrite is a no-op if it exists).
+      kubectl taint --overwrite node "$NODE" \
+        "${var.quorum_taint_key}=${var.quorum_taint_value}:${var.quorum_taint_effect}"
+
+      echo "Node $NODE ready, labeled and tainted for quorum-only scheduling."
+    EOT
+  }
+
+  depends_on = [talos_machine_configuration_apply.rpi4]
+}

infra/rpi4/kube/variables.tf

@@ -0,0 +1,146 @@
+# Variables for the Raspberry Pi 4 Talos control-plane node that joins the r740
+# "kube" cluster as a third etcd member to restore quorum (2-of-3 majority).
+#
+# Secret handling: the cluster machine secrets (cluster id/secret, etcd/k8s
+# certs, bootstrap token) are NOT read from terraform state (the r740 state is
+# stale) and are NOT generated here (that would create a new, incompatible
+# cluster). Instead they are provided via `machine_secrets_file` — a local,
+# gitignored JSON file in the Talos provider's machine_secrets format. The
+# file is produced once from the live r740 node (see
+# scripts/extract-talos-secrets.sh) and stored in a real secret manager
+# (Bitwarden); you paste it back to disk when running this module. Ephemeral
+# resources + write-only attributes ensure the secrets never land in Terraform
+# state.
+
+variable "rpi4_host" {
+  description = "Reachable IP/hostname of the rpi4 Talos node (for Talos API access). With DHCP this is the leased IP (e.g. 10.1.2.135)."
+  type        = string
+}
+
+variable "rpi4_node_name" {
+  description = "Kubernetes/Talos node name for the rpi4 (e.g. rpi4). Pinned via machine.network.hostname so the node registers with this name regardless of DHCP."
+  type        = string
+  default     = "rpi4"
+}
+
+# --- Cluster identity (no terraform_remote_state — state is stale) ----------
+
+variable "cluster_name" {
+  description = "Name of the existing Talos cluster the rpi4 joins. Must match the cluster the r740 bootstrapped (kube-r740)."
+  type        = string
+  default     = "kube-r740"
+}
+
+variable "cluster_endpoint" {
+  description = "Endpoint (host:port) of the Talos/Kubernetes API on the cluster. Must match the r740 bootstrap endpoint."
+  type        = string
+  default     = "https://kube-r740.lan:6443"
+}
+
+# --- Secrets (provided manually, never in state) ---------------------------
+
+variable "machine_secrets_file" {
+  description = <<EOT
+Path to a local, gitignored JSON file containing the cluster machine secrets in
+the Talos provider's machine_secrets format (cluster.id, cluster.secret, certs,
+secrets.bootstrap_token, secrets.secretbox_encryption_secret, trustdinfo.token).
+Generate it once from the live r740 node with
+scripts/extract-talos-secrets.sh, store the contents in Bitwarden, and paste it
+back to this file when running this module. The file MUST be gitignored — it
+contains the cluster root of trust.
+EOT
+  type        = string
+  default     = "secrets.json"
+}
+
+# --- Install / network -----------------------------------------------------
+
+variable "installer_image" {
+  description = <<EOT
+Talos installer image to use on the rpi4 (bare metal, ARM64).
+Must be an ARM64 Image Factory build (schematic generated at
+https://factory.talos.dev) for the Raspberry Pi 4 platform. Unlike the x86
+control-plane nodes, this quorum node does NOT need the iSCSI/Longhorn
+extensions because no user workloads or Longhorn replicas are scheduled on it
+(the quorum taint keeps it empty); only essential DaemonSets (Cilium, etc.)
+land here.
+EOT
+  type        = string
+  default     = "factory.talos.dev/installer/ee21ef4a5ef808a9b7484cc0dda0f25075021691c8c09a276591eedb638ea1f9:v1.13.6"
+}
+
+variable "install_disk" {
+  description = "Block device path to install Talos on. For the rpi4 booting from the SD card this is /dev/mmcblk0."
+  type        = string
+  default     = "/dev/mmcblk0"
+}
+
+variable "node_subnet" {
+  description = <<EOT
+Static IPv4 address in CIDR notation for the rpi4 node (e.g. 10.1.2.135/24).
+Set to null (default) to use DHCP. The rpi4 uses DHCP, so a static address is
+not required; the node registers with Kubernetes under rpi4_node_name regardless
+of the leased IP.
+EOT
+  type        = string
+  default     = null
+}
+
+variable "node_gateway" {
+  description = "IPv4 gateway for the rpi4 node. Ignored when node_subnet is null (DHCP)."
+  type        = string
+  default     = null
+}
+
+variable "network_interface" {
+  description = <<EOT
+Primary network interface name on the rpi4. The built-in Ethernet port is eth0.
+EOT
+  type        = string
+  default     = "eth0"
+}
+
+variable "nameservers" {
+  description = "DNS nameservers configured on the node (must work independently of kube)."
+  type        = list(string)
+  default     = ["10.1.2.148", "1.1.1.1"]
+}
+
+# --- Quorum taint ----------------------------------------------------------
+
+variable "quorum_taint_key" {
+  description = "Taint key applied to the node to reserve it as a quorum-only member (no user workloads)."
+  type        = string
+  default     = "dedicated"
+}
+
+variable "quorum_taint_value" {
+  description = "Taint value applied to the node."
+  type        = string
+  default     = "quorum"
+}
+
+variable "quorum_taint_effect" {
+  description = "Taint effect applied to the node. NoSchedule is sufficient: essential DaemonSets (Cilium, etc.) tolerate it for networking, but no user workloads are admitted."
+  type        = string
+  default     = "NoSchedule"
+
+  validation {
+    condition     = contains(["NoSchedule", "PreferNoSchedule", "NoExecute"], var.quorum_taint_effect)
+    error_message = "quorum_taint_effect must be NoSchedule, PreferNoSchedule or NoExecute."
+  }
+}
+
+# --- Kubeconfig for the label/taint null_resource --------------------------
+
+variable "kubeconfig_path" {
+  description = <<EOT
+Path to a kubeconfig for the cluster, used by the null_resource that waits for
+the node and applies the quorum label/taint. This is NOT stored in state — it is
+only referenced inside a local-exec provisioner. Point it at the r740 kube
+module's kubeconfig (../../r740/kube/kubeconfig) or any valid kubeconfig for the
+cluster.
+EOT
+  type        = string
+  default     = "../../r740/kube/kubeconfig"
+}

kubernetes/code/gitea/values.yaml

@@ -24,15 +24,33 @@ valkey:
             limits:
                 cpu: 500m
                 memory: 256Mi
+        tolerations:
+            - key: ENC[AES256_GCM,data:o/sXxge7tAi4,iv:4at3gD2SjyjA1r1eVYE3bi0xEzFkwxKgJRJGagfzDjA=,tag:OxOcClpHU/Z3sSiMHikOsA==,type:str]
+              operator: Equal
+              value: ENC[AES256_GCM,data:5FY3Hmqh8Uw=,iv:8LGdTpFovI9lsfAaf65dHdJnNsm17ahm2/x+97LiTgA=,tag:b07oTtzzfU4FFjm9W+iz9A==,type:str]
+              effect: NoSchedule
+            - key: ENC[AES256_GCM,data:FPkNzrFpfqQI7h1D346c72Og0B2y5w+z5m25trSt4kIEjxJ4RA==,iv:Xjl+xrXs5m28siBqNXTTmtLKEiTxnLZAzvFXEXH8ZcM=,tag:VQoot4mURGhCmy+MSjr4BA==,type:str]
+              operator: Exists
+              effect: NoSchedule
+        affinity:
+            nodeAffinity:
+                preferredDuringSchedulingIgnoredDuringExecution:
+                    - weight: 100
+                      preference:
+                        matchExpressions:
+                            - key: ENC[AES256_GCM,data:O4Wtr+1YuNGmYknT35XPONTeqUk=,iv:VqTkPwG4bxvRgXDGrmpwVeT2zLQzOLrwKlhqYCv/GiA=,tag:2pntJGKlUF8cBKgKpyySHQ==,type:str]
+                              operator: NotIn
+                              values:
+                                - "true"
 postgresql:
     enabled: true
     global:
         postgresql:
             auth:
-                postgressPassword: ENC[AES256_GCM,data:5SoftEfh4w8hmA==,iv:QbbOmVM3Jp3gxAxL1OvnB6WKFX+OTHFENz1roMnPqyc=,tag:Gtn7H7d3O7Hweg5q6P6xUg==,type:str]
+                postgressPassword: ENC[AES256_GCM,data:/vgY7Q154znmpg==,iv:GqwNyccbeCSm/b+7PufHcoSrsrrIR3FVgRAI0WxrHow=,tag:o/kkueEMtJUVLeARdy5zpQ==,type:str]
-                password: ENC[AES256_GCM,data:0BMgFdvPvhB9oQ==,iv:Z0mkQBfhpc7y1nSLY6MOucxwuKZ8TfOk6NlQE9RwVQw=,tag:yGDAOH6TCJDCfhxNlC/eCg==,type:str]
+                password: ENC[AES256_GCM,data:gc4Ev4kn9KaSqw==,iv:Bd/qGIaobKewB/rtJf426pzS4qPe/WAMi2AB7T7HtCE=,tag:XlDtRF1tWv+9YmTo1H/Lvw==,type:str]
                 database: gitea
-                username: ENC[AES256_GCM,data:F7+91aE=,iv:tEGgJ6GoXVkNoD6D4ZnyClrOQLGlqzFIUqqwDrVjd9U=,tag:H1+f/EKf/KptooRf8g0qcQ==,type:str]
+                username: ENC[AES256_GCM,data:6mlFP+E=,iv:+QEgLA6reecmpvvtmdOnwkpqOa7fOV06ZS8VTDQo7MY=,tag:dREH7uvHk8ZS6HcA7Vo0zw==,type:str]
     volumePermissions:
         enabled: true
     primary:
@@ -43,6 +61,24 @@ postgresql:
             limits:
                 cpu: 1000m
                 memory: 1Gi
+        tolerations:
+            - key: ENC[AES256_GCM,data:xd4ugBSV+lcC,iv:3dUqomveQf0/rMwOrlNnTk1mJoF5sW2RQ+UrPLlczkI=,tag:16IzYY0mqB8LlwRGXxUy8w==,type:str]
+              operator: Equal
+              value: ENC[AES256_GCM,data:rK9ACxsGlEk=,iv:afDm0KWC0HAHw4AJnlpPw5oGcfEe3oG+RgiRCTpjb7g=,tag:sDOIWCiqxi04YYxalZ6DbQ==,type:str]
+              effect: NoSchedule
+            - key: ENC[AES256_GCM,data:83c8V9jJ+d5zpFFFhbo//O/opfgAKNpoqcsRw8W9oGbTu9pBMg==,iv:uw+6IhfkixBxBo7Gs2VSvW+q+JpW35rE9REkb6SQDTc=,tag:donRKnK6NbgBKhDgEOR0Eg==,type:str]
+              operator: Exists
+              effect: NoSchedule
+        affinity:
+            nodeAffinity:
+                preferredDuringSchedulingIgnoredDuringExecution:
+                    - weight: 100
+                      preference:
+                        matchExpressions:
+                            - key: ENC[AES256_GCM,data:UKDLBJ1yfJ2gcyu/Ms3jVnBTfDE=,iv:fK73/46Jh3nfPMnWmMHr+SK3k3l16IsT9Ah45+yCz0w=,tag:A+PvsR9YpWwHrVpHOYXMGQ==,type:str]
+                              operator: NotIn
+                              values:
+                                - "true"
         livenessProbe:
             enabled: true
             initialDelaySeconds: 30
@@ -75,8 +111,8 @@ gitea:
     oauth:
         - name: Authentik
           provider: openidConnect
-          key: ENC[AES256_GCM,data:oq+UsuiNEGWphxXbwtbBs3d27bB2cVPoR617TWhcPOk67D7XWUPkmg==,iv:hODOA45FFex/b5ghHScnDoc5osu1d8bzD969oFVH3wE=,tag:67TMl55I6zgJWNGMqco6fg==,type:str]
+          key: ENC[AES256_GCM,data:ctukTzFfQAZSAIaXonIu6nVsTHnTNAlqObVhGkKt/aLa9o3KxDHdBw==,iv:QoudL4RRdwpPfz6h6MWoaBOztALRa8PEi5fRFkjFzQ8=,tag:+TgP8KrYQNCBmTaqscOn8Q==,type:str]
-          secret: ENC[AES256_GCM,data:o9BZSzy61ouycNgtgRg/Jn8Wz+hLBBqgX46iHbaYtPbUVPVFc9tpdCsTdHCpfSGfiS1B8kWcoSzIDIAIIofRSCS/QqEQeuTwsyakkXhgn/LtLn+TH7bdLLS8qqusy5MZ9U18IJ/UBPGZ5+16Y/glbhpfHu9NNv0AI07nPciTq1w=,iv:7dn7eW2oEK/EU4xIETwVSekZUmktrs7U9KTJC486nHs=,tag:Otl2KuySu9bI/cZfCyjiKA==,type:str]
+          secret: ENC[AES256_GCM,data:mxNRVXfX0X777xh5bB7PwMEeaxRPszmvmpoQt2lMtvkTSIQMJOllzY27PAShJ/y6JBPRslg6us9TpKd3zt5KcVY4Jne3hkpkQCKqE8HN3XnjHDHDzAm4ugYib5J97BeLw9pvdurLxw5iXAp3XFxMytzU4xCrYPsS61FWJRTLFRg=,iv:NCGMyWJT575CGnnFM4mcLDxsoMkggsxiNkXNp/aU0es=,tag:5uudyC1IIxorgepR7HnPMA==,type:str]
           autoDiscoverUrl: https://authentik.vhaudiquet.fr/application/o/gitea/.well-known/openid-configuration
     config:
         APP_NAME: Gitea
@@ -112,28 +148,48 @@ resources:
         memory: 1Gi
 # Increase replica count is not useful for 1-node cluster
 replicaCount: 1
+# Failover tolerations + affinity — allow scheduling on p330 during r740 outage
+tolerations:
+    - key: ENC[AES256_GCM,data:RBMIdZLIV79D,iv:TTa1CovTGm91r/lYCKOKjnhM+i9hoLVsmDi2kmki9c4=,tag:zuhnRKxy3dtkYJr3TuLilw==,type:str]
+      operator: Equal
+      value: ENC[AES256_GCM,data:LLOg34s1mrw=,iv:rjOyHPWBGrpNl1YGi2cYN6JRm7m28HiRibQoK0JZ3CE=,tag:oSyz0nISr3plqdXQ5UJxzw==,type:str]
+      effect: NoSchedule
+    - key: ENC[AES256_GCM,data:jVh0MTlCs7rVPSYvulsw8bIGwYPw5soaOzMlzVzVbFy13qqKLA==,iv:JfZ+JWxHBiTD/rG8xHYKogkyryNQIUIXyf9chxpwlEo=,tag:hwlZ+k1L0uPe+UGBOeMlYg==,type:str]
+      operator: Exists
+      effect: NoSchedule
+affinity:
+    nodeAffinity:
+        preferredDuringSchedulingIgnoredDuringExecution:
+            - weight: 100
+              preference:
+                matchExpressions:
+                    - key: ENC[AES256_GCM,data:Sn7FuYqGMbTjx/EpqRIDx5GofR8=,iv:Wkyz+RmyTGswohAZmyeQ1lWT8hTaYw7qAVmXTnSIT7c=,tag:WAiU90Wo8Av0TsdMtyPF9w==,type:str]
+                      operator: NotIn
+                      values:
+                        - "true"
+# (postgresql and valkey tolerations are set in their respective blocks above)
 sops:
-    lastmodified: "2026-06-27T10:20:08Z"
+    lastmodified: "2026-07-25T12:44:33Z"
-    mac: ENC[AES256_GCM,data:VZY4oK2PCMBJd4VozL5qS7jlZtOu1JNsOn7aqJgPZ8uQR7kG939By7oVELMH9us0Sb8AZIYmzx7nLwnlp6IUNnacOffhKNl4stTFrb5Vry/5zem0CNp08eEDEt9VoxNudfQYC2axuGSwbAIahMzZsOYVRqOGgF1gEG17+hgd7ww=,iv:MiHKGUfzaSAppXLLxJSQD5loGkToFLzVQetWoRHq074=,tag:FduCPbHmy6A0mYdCrnRxDw==,type:str]
+    mac: ENC[AES256_GCM,data:V9ZOpVvIi9PzScyasqQJb0yZDrqQ0RSXGyAY1+WaHCIa4/YRKn9d9V4wy+CwWaeEiIO7+fkp2ROInl7eVmETm5JppQ899G33GTmHCZLcBvVR2M4xEkx8eXy3saKCbC2DN4bHRuSCvvqw2XR+QIdemLL7ETuRo2cepUMhz/93peY=,iv:3wm/iCVfAYUUa7VfBXsx200LypYYWgJYCEm5X5Wwsnk=,tag:OzGw1PQUuVRVtQ5slH3E0Q==,type:str]
     pgp:
-        - created_at: "2026-06-27T10:20:08Z"
+        - created_at: "2026-07-25T12:44:32Z"
           enc: |-
             -----BEGIN PGP MESSAGE-----
 
-            hQIMA7uy4qQr71wiAQ//VDYGwZupOfefdEtJkD9SPOW3tHcOy2XR6ULhVQjXTGGh
+            hQIMA7uy4qQr71wiAQ//RIr73tcloQQg2OqSLn7W1DI6T1dYCh+ntTP3Y8KeB1WC
-            APuONtgUZKFCViUqYZB7l6fazM8U/fv3cHsIpZfALFGE9wSZ+4So6D5dmBj13egh
+            hZy2JOdTwHLkd9MiDimgjH0E1036+Pta9xhmoi9lnY3cvwOH/1oHxjd6ixhPxYzV
-            yznV55xjinoaKPzaYYZoVmlt1ZO22SxmC0vld3yfAs89KD1Aow5wNZGTBELzyAWF
+            Xt38fQ3BkFvf3fHbhcbzpMIFL8tfvuVDAxRV1fHHOdCGgXqeC5GP0KtDGmnddHui
-            gMYGMkCsLlH3gqBaJTV1m5LKHHQAtoyzvJbKcpiynNRySfN/ogNX67GmgEMH9+u8
+            w+ms/1bS/EiwS4C7IW0yYSD3nGIF3GyLwMPOnbmJSz+ealEWjuBi8R0S1gX40QFU
-            UexPwUHL6skIzHLmugYcrs4FAABot4WzoJsR794wOq/DEqHBE4SfQg48y/DROu2O
+            EV0MBrVrtFtnA2R8UZGCjU/xpgoQFE1U8dArNt0DPqNsOeL17WQZrZ58DurmnmiE
-            mgo6KqiXXe8qPGCIvjLiudb7MdSLpoLqgZ0vpExzdS/wMXAXczodrRRJmzEQyeDE
+            DNjndlR8Est1y0gSO9ujPtGXbbuhAOIXuEsYMfnBgQ5Y+dShs2G1EHC0OexYB83W
-            j08pquhfUzQ1SA/9VRQkgl4r0dj0CTuc+EhvHD4M9nj3JIjvuJ0AdyKye5S8rdNM
+            FSHxM69hjQkvRLL0/lX+5NH5h+IJqZEUsLd0/J4h2Iz/EsDRxO6pKt0HzZbJd/y8
-            rm9eyX8V+o6X811qUM8bobKEi36tdnSMiB1yVHTtdW7hKctQwRMNQ8A9QEtOxS2F
+            qY0Y2K5cGUzG0y0TWuUznJyxmTSCgrZAE8sgBhU607+UnXU678cqbkRgSuHTNdMe
-            +9Owub4LXZwBejSLgh9jaKEnilJQKVYN5Aw2lNAmZZZKtJ+iGHHkq9H6k+k5oEoG
+            yNxQas/PA0ltezOe3dBG34Cwf0oQZ/NCt2TfSFciFEgf6ze+fu+T+gCFfftWPiy6
-            xkQ/IYIEGLhfjH0PDZ9m2G3Ren3EkEkNjYvhwENhuUcyuW5Abl62v9FMY8c9NYyK
+            pQ/3Fl6yAaVguu6UCR27wRMhXETZ+YZ8rw32XPbLE6BOvQ1LcGg9E7wJlw5eeir8
-            +uf5VS1ceHOBcelQcdSx9Kuox5UuyUtbwUcv+HbTmymZtmYdw+dnOTtRHR5CNY7S
+            0aWvKyUu/ClK6+/s7Ytxv1RoL72gn6WuOgK5q6Sf3mqmltuvJxtDVAnqCHNRfHXS
-            XgEvPWAEagJpsmqC2n7Vc9d7b7TVp+SLSiL/w0v9JD6ko54Rwj0x21Odx3FLe7P6
+            XgFH8gPqy47Gqg9qggCHjDIUbQeAu1IAznJAQ368WXkTqqnq887sSSgcs6JVduJf
-            obKU6XNNX9PPThJw2W0E295i2+DmOTH0AFio3rsg5wM3JFUokY/t8590sWnFR8g=
+            2YV5NQFOcoI6OF66FXXmY+sTpJaHcgVfKt3EHAqXxHCFZUSBHMwDJm9ulRBuYsc=
-            =Gypa
+            =raBM
             -----END PGP MESSAGE-----
           fp: DC6910268E657FF70BA7EC289974494E76938DDC
     encrypted_regex: ^(password|value|ssh-key|api-key|user|username|privateKey|clientSecret|clientId|apiKey|extraArgs.*|.*Secret.*|extraEnvVars|.*SECRET.*|.*secret.*|key|.*Password|.*\.ya?ml)$

kubernetes/home/home-assisant/values.yaml

@@ -1,42 +0,0 @@
-persistence:
-    enabled: true
-ingress:
-    enabled: true
-    external: false
-    hosts:
-        - host: homeassistant.lan
-          paths:
-            - path: /
-              pathType: Prefix
-configurations:
-    trusted_proxies:
-        - 10.0.0.0/8
-        - 172.16.0.0/12
-        - 192.168.0.0/16
-        - 127.0.0.0/8
-sops:
-    lastmodified: "2025-12-04T18:23:43Z"
-    mac: ENC[AES256_GCM,data:JSaTCBXqpiP0R3rIEa/EZaDPbBInY2a6UMhFD9Gcw/mJVq/wF72KsFRfJl2hZ3d1FtsQ6Z0l3GrbiV41348jxzWM4XnE0e0hywdc2Ryqw0B3GS0EoNk1IoPR8ieZtxWNtFvglR2gVWWUnN7caYf/acE04NnzIDxl0tPEFckS+Nc=,iv:C784uqYb7KGhn2wGWZnCA7rhGocJWP6mfBhjNLWbimE=,tag:A7JLOmSVJ+ZrNA36DJfRCw==,type:str]
-    pgp:
-        - created_at: "2025-12-04T18:23:43Z"
-          enc: |-
-            -----BEGIN PGP MESSAGE-----
-
-            hQIMA7uy4qQr71wiAQ/+J1e3KfHaAkFS+lYz9jKzPAffkiDVCeecEf31Lk9DJxFd
-            7agtTWvcXtIsiJwF4sv4NzxY3pQrNRKwiq8pd3iXWHSqflIaQ2M9+dUGHykkjA5W
-            R+y/wgM+K0MBaFMxQU/xC0lM7YFwypV1JBp4jmpNn3McGba2yZAoKo1uAQ9LMP4L
-            Fd5wqNUrLNfwbh5Ts1rBVAGwStVlkRUvdUVCXjngdWVmE5xB3IBypdrFX2oclkl8
-            0zSNJwknC4bi4FeKBxWC90HT0n8zSMybdZOltcl8J3YVCL6l0F9hw6Ub5j1ej7qN
-            Igyll2/T4FEOExdo9oRXbEjC6GSZZBD6NMlrgJqVVUWoqDm4WQlKc7JRYAjtRgbS
-            Us8qCD8CMb57E6AZHD+GRZDK48OeevM21YxRvpSQlQfyeGbMrxF05zzw5InHSMhw
-            VslHQRmUMQqEzTsSeFbNxTeqy1YGGDYlnBpEjl8TPIZXE+mjwHm9QJbKwQFUK4Rh
-            YuEZDlUONdaKQpNX9OS6kwiV493sSruC2yWNwYIlgERrkcjBggKYPiV7lDiFnl1W
-            RVcsqOGnsrrwFHT7m6/MTIzH0k4fDt6Vv/FCToaoadvLPYNZb7xoKMahzT1z3SP9
-            FMuGus2r8h6ShEZVyLAt6o+BuuYyzOrVo10k1yTKDqEDD662rbRvTrBZlOk1vUPS
-            XgEtw6w6fFkESJKetHiN6TxG2+WyJw5PILm2sjak5AcR3qWiux/ENPb8kE0BIB0S
-            1Z/YTHYNP2mtEegDsp1DoT9tOSTWfrllL+p3kccvizEB5hq9+oMEzPGBstjWRjg=
-            =5PDx
-            -----END PGP MESSAGE-----
-          fp: DC6910268E657FF70BA7EC289974494E76938DDC
-    encrypted_regex: ^(password|value|ssh-key|api-key|user|username|privateKey|clientSecret|clientId|apiKey|extraArgs.*|.*Secret.*|extraEnvVars|.*SECRET.*|.*secret.*|key|.*Password|.*\.ya?ml)$
-    version: 3.10.2

kubernetes/home/home-assistant/values.yaml

@@ -0,0 +1,61 @@
+persistence:
+    enabled: true
+ingress:
+    enabled: true
+    external: false
+    hosts:
+        - host: homeassistant.lan
+          paths:
+            - path: /
+              pathType: Prefix
+configurations:
+    trusted_proxies:
+        - 10.0.0.0/8
+        - 172.16.0.0/12
+        - 192.168.0.0/16
+        - 127.0.0.0/8
+# Failover tolerations + affinity — allow scheduling on p330 during r740 outage
+tolerations:
+    - key: ENC[AES256_GCM,data:SMOTaAtzpNtv,iv:rsMiAMaugnRKQjtp67nU0pW+MWXgolluiE5qoIWJkRA=,tag:RLUTotQG3XUQfDiA7CE7vA==,type:str]
+      operator: Equal
+      value: ENC[AES256_GCM,data:DhbXHMsUJ7E=,iv:lQQTpWpDGi7JEQV24jdICHMnC8yV0EsNQ/nauBsqk/8=,tag:RKripNRdBwqnt1sCY72eYA==,type:str]
+      effect: NoSchedule
+    - key: ENC[AES256_GCM,data:YApCSh00O5YoRqzHUNSoPbYWOsNVwFKOlln0C2+mYpDfKqLU5g==,iv:eOv/Yri400RHpm6Bnxqw4IYp2IiYAKXTd06sduW3Ku0=,tag:9GR1Q8lWHG7Uplevd1hbYg==,type:str]
+      operator: Exists
+      effect: NoSchedule
+affinity:
+    nodeAffinity:
+        preferredDuringSchedulingIgnoredDuringExecution:
+            - weight: 100
+              preference:
+                matchExpressions:
+                    - key: ENC[AES256_GCM,data:0rESi0B+l5uEwmPQPuFzMylc6JE=,iv:dGx39YEojvQ/J2y6SlP54I5L7Zr+yecwJwNQQ53RV9k=,tag:icAw2r/q4+UcHzYClfOBSg==,type:str]
+                      operator: NotIn
+                      values:
+                        - "true"
+sops:
+    lastmodified: "2026-07-25T12:44:32Z"
+    mac: ENC[AES256_GCM,data:cNqULRYPz0Es3az9v9HWlEf1pvPFWfZ67OofRWaDgFx3KVKVj6VUY6HQgN589xbbCoBKLSKXUoag2zOku1P3yBkb39WrdUtfSJw2t7BvZIo0l9VG0bRAYZgJYgnQA1NrjxixzgZZ7AHst1ljB+ZC6qq5DgEJcuCiue9zkfySTMY=,iv:rNDmiWS1U01xgRHty/sp/RJ5iz78sOdBGP0mdjUX7TA=,tag:BVKsb/HQchTaLo+2Tlrn+A==,type:str]
+    pgp:
+        - created_at: "2026-07-25T12:44:32Z"
+          enc: |-
+            -----BEGIN PGP MESSAGE-----
+
+            hQIMA7uy4qQr71wiARAAgxstgt71NySIfZBgF0Nb78qkkFV7xOEfv+VWO3EaSp7Q
+            8jhARPxkAVMwl1MQIdCxB6UuUeQi8fOrLwjYzTN1tWG/VV1P3KV3klm6jNcZ4Zxv
+            OdYt6kRLNfI4J7KWFhZg8A7/FfXSOtZEWLjD60hb4ie02ALEC9o2rUcdxWH2ZdHU
+            69GhP3VU9uiChOxEk6eLyQr22IF9IfQ7BVUckVzU13neZ2PEYXCV3QaHMBrvutMz
+            IFUGpi+drrXxXq9KmJQ4Yr4L0+mHwXYrFVi5rHK/ibgRDNT7T9nK56WCsgRd0MiK
+            7+1vK/q73p5ZdozmalYB2lKuehRapf/xdXbn1SrWGXbsHubdrE2AQgRxBHVy3k8I
+            spN2ZTPzl29mUsVIWKE/CdF1GsUGYrKwKL7S6aHZBtRXncbKBEFLbrloAjs+j83+
+            aU5ybyQWnZxh7f4+nQyDerSJXZJ17eEJRjqMDhJ6VmA8iOBWxgudM57d4Umt53cf
+            s4vOikRSO0cd+/OMvQQBW1IPMkwRSi4/jD8BZTXisWVUQ3/98NQPXQE5EYk2w7aM
+            5T1orWWWMz5EdQNptqz5GL6qPoH4UpVSGCNuCf/oyLR6MCt24JDR9KobOgXKUrfM
+            Qig6/+4WbvxBJ9zevi14FL+K00MeWd3pQT2Bsg8l3fWPIRe+p9lSB6yrsf6xHu/S
+            XgFWycRMj+U+y7B1kyXT5W6WshFHHTOlyAVgk9lnB+kKEVno4pAQBNuNlW4gwOvr
+            iuc9EaiuHat+CW+WdxLdH3i/jcNmvXfXRXdE3449aXv+haNBjxz93nRo1vbAFuw=
+            =O+AO
+            -----END PGP MESSAGE-----
+          fp: DC6910268E657FF70BA7EC289974494E76938DDC
+    encrypted_regex: ^(password|value|ssh-key|api-key|user|username|privateKey|clientSecret|clientId|apiKey|extraArgs.*|.*Secret.*|extraEnvVars|.*SECRET.*|.*secret.*|key|.*Password|.*\.ya?ml)$
+    version: 3.10.2

kubernetes/infrastructure/authentik/values.yaml

@@ -1,14 +1,14 @@
 authentik:
-    secret_key: ENC[AES256_GCM,data:wnDvn9EdX2SYDoWn9tTrbKvBx11fGfGlTk+9MSz8cBzwMDaY5W1bCwjtH7TmAnwYhbUQxXdVIxgPpj7k0FVqM5fuMHZwcb2AgtDMGbxRXtM=,iv:jwUq7dbl+V1t+Ek7vyBEfwX1B/fVmE0TIFMVGj40kVI=,tag:G1bDdR9OUqOtZQmvUcAfRA==,type:str]
+    secret_key: ENC[AES256_GCM,data:P5e7cSOa0Uw0NEmWuDzIv1QxIkbvLE9vjzULmS8yS8b62fS5jHuFWUF/8VkW+pXiJo51mZ9b3QWDBdD732hpSrR7hUpK19WdZvFfXMKsqVc=,iv:AuuvJtXdlQ2SzeiFdxv96cEVyeMTNW/fG8bPS4ItFtY=,tag:AJC/T+eXGwOdQ/U7rebIdQ==,type:str]
     postgresql:
-        password: ENC[AES256_GCM,data:G0KPHz7zZ5ivtA==,iv:w4m21LqGEtvEGAp7NlsoV9bwRkE/Mga4Nxp5mqBIJHs=,tag:kKHC3yXqJKZX9jiYuYXt9Q==,type:str]
+        password: ENC[AES256_GCM,data:wmidNe0ODeH8FQ==,iv:zh38F/lIKdFBBIQufjLORtfusPWwDnsDIh61b6ltbQE=,tag:SSMCicRKoPBXwPLFoh+uoQ==,type:str]
     email:
         from: webbot@vhaudiquet.fr
         host: mail.vhaudiquet.fr
         port: 587
-        username: ENC[AES256_GCM,data:s0L1fCp8aObv5j6jlBiGDfj0aIM=,iv:cbvXxI3DIjnhnfHmUpeDF5KKRxUC2+YdvwU5FPcEe0s=,tag:rR8HpXUEabNZ36xabVkSIA==,type:str]
+        username: ENC[AES256_GCM,data:VV9x5UWZ2F/p/l4pOVdDB3ZeK98=,iv:H7YXijpo2oJEO96LJOERL4lyQtV7QJq86H33wDpKObU=,tag:q+KiivjpV3kXY0AYGKhzFQ==,type:str]
         use_tls: true
-        password: ENC[AES256_GCM,data:rb6o5MGQWK7LIT7gass=,iv:NjI0iqRk2usrqcH6kSo72T337y6avHz5whg9bLfAYDA=,tag:DnRJaixjmNY31V1h9Et5PA==,type:str]
+        password: ENC[AES256_GCM,data:pco+hLWairs2UFNFTSY=,iv:OMrxrzKDo91UIMgEhPdN4dTpRxl5A1LKHYk5h4eFD/A=,tag:ecjAP9lv6FPxPppyurJ5cA==,type:str]
 postgresql:
     enabled: true
     volumePermissions:
@@ -17,12 +17,30 @@ postgresql:
             repository: bitnamilegacy/os-shell
             tag: 12-debian-12-r51
     auth:
-        password: ENC[AES256_GCM,data:9PI//cgxRlmjNw==,iv:q0SLpmaTIC3OUulPDxiDWie0+oI7w17V2fPmJ52o8ao=,tag:2lwG95Remzqt1dj8H4CG0w==,type:str]
+        password: ENC[AES256_GCM,data:cRaZ23HWjZ41ig==,iv:HABmXhmHvMhjUJ989xJOFPWW+EnLTCot1is80vMTBUE=,tag:r+/iRkQIlXvALv/JQbGFRg==,type:str]
     image:
         repository: bitnamilegacy/postgresql
         tag: 15.9.0
     primary:
         args: []
+        tolerations:
+            - key: ENC[AES256_GCM,data:M14eAjMj2w2+,iv:5vIosyFtwVgqs07CT2UcL1KL2hqg+rvKS5PWRvWRlPQ=,tag:OJ4RwlLAhGDN4JGPX8gypw==,type:str]
+              operator: Equal
+              value: ENC[AES256_GCM,data:KKZ1Kd5yRa8=,iv:adSUcfmjnDAc9QZ14NDfFz4oATpcBhtn8iHTF2xXpcw=,tag:G//QEbSKXiGP7nD5oAfNSA==,type:str]
+              effect: NoSchedule
+            - key: ENC[AES256_GCM,data:KtJS0g+/3y6VZb9VNKv1IWMwIi8FBURPGT6W6qvBAV1CHW4UMg==,iv:1I8o3SNljujN4FqByFC9uHpImJhHjj4Us0QEpHpDHZ4=,tag:FIpRW9I4OSVFVx4ejmau5w==,type:str]
+              operator: Exists
+              effect: NoSchedule
+        affinity:
+            nodeAffinity:
+                preferredDuringSchedulingIgnoredDuringExecution:
+                    - weight: 100
+                      preference:
+                        matchExpressions:
+                            - key: ENC[AES256_GCM,data:IkWdKkS/XlGw0u7I+Q2wYhXL+CU=,iv:2k9d0Jo4Vw5gA4tGeIq1Dw0IiZKQykfV8YAt2lYk7GE=,tag:Xnz58VP61Ew/J8hvIClOVg==,type:str]
+                              operator: NotIn
+                              values:
+                                - "true"
         livenessProbe:
             enabled: true
             initialDelaySeconds: 30
@@ -44,28 +62,49 @@ server:
         enabled: true
         hosts:
             - authentik.vhaudiquet.fr
+# Failover tolerations + affinity — allow scheduling on p330 during r740 outage.
+# The authentik chart reads global.tolerations (applied to server + worker)
+# and global.affinity.nodeAffinity (preset system with type + matchExpressions).
+global:
+    tolerations:
+        - key: ENC[AES256_GCM,data:yy3LVikckQ4y,iv:PlWRKZVoR6g/RtQdGBoIKbnwjX887xZY7ZhciTa0nps=,tag:yRMuJW3yjBV2C3LDWVdEHg==,type:str]
+          operator: Equal
+          value: ENC[AES256_GCM,data:IqKHhgroaK0=,iv:bvg7kLaVQEsH31xHH1rCW29dWlWDlDsAMAmR1cX2nS4=,tag:B77eeEjVnRwk/feuPyi2Kw==,type:str]
+          effect: NoSchedule
+        - key: ENC[AES256_GCM,data:QYGmJBgxhqC2L3UvGYyJKKWDfzSFq91w1ifwNn9rofPnBqL0Ug==,iv:+tI3xw51LqinZnEOi7Lgao9liOD8KT8rOGNtn10qkFg=,tag:fJhk0PifHEyFlX4j75Sklg==,type:str]
+          operator: Exists
+          effect: NoSchedule
+    affinity:
+        nodeAffinity:
+            type: soft
+            matchExpressions:
+                - key: ENC[AES256_GCM,data:qRInCqtomABFKcUFHzDQe2qmzJY=,iv:9cMrKqqpIMtNCUqkIWaCr5YGNoK6zsID3xAKp34q1kM=,tag:Zf1ZjGwCgnaOqTq8Hr++Fg==,type:str]
+                  operator: NotIn
+                  values:
+                    - "true"
+# (postgresql tolerations are set in the postgresql block above)
 sops:
-    lastmodified: "2026-06-05T20:13:55Z"
+    lastmodified: "2026-07-25T12:48:03Z"
-    mac: ENC[AES256_GCM,data:rtrnY4XXaovFr/oXuZWJQa17Ihlgb9W7WMKYeN10qqx8REf6a3zqlUUGGSndYhjAZwpUjo3qixqAjW6kItBDgeF3DLpQ2T6acyOgXmbYNN+dWK4zb7jE0yZz69XIkUirabBWTkLqDACvw3iW5x4EAjg6/VYRVt4g5eugET5F/IU=,iv:eM+Z4SEJ6/y6gHzlYjLxcbqRn4dP4gbkaQqN2EGL4BY=,tag:EXklnS8lNeQErKcdP0XD6Q==,type:str]
+    mac: ENC[AES256_GCM,data:noPba/BIxyoUo3sFfwXuMaZwwSMNADgWcwGaIoQchdzu3KodtWRHk5xJ9wy4VGelOuy3gS79j9ErDnsn8I1v05Bqqlrig99Sch5Jlu1UQpeUQtUjq9S7d5BQ0f3QdeU4iP2v28JovI3gBNHEIlZiueV6xYUrpeoKzS9ukrBtJsM=,iv:XRmbI5hX0OCh4I+7MmnI9LyBLJhf2CsfHgvsNaEUWgc=,tag:Ym5JeVRqTWHWl8qSQAFh0w==,type:str]
     pgp:
-        - created_at: "2026-06-05T20:13:55Z"
+        - created_at: "2026-07-25T12:48:03Z"
           enc: |-
             -----BEGIN PGP MESSAGE-----
 
-            hQIMA7uy4qQr71wiARAAsreeLsc0Qhc/RDFP3GmusT0PCFq0XTnin9THWE+7YC/5
+            hQIMA7uy4qQr71wiAQ//Sa5nJk6vXXFdf4C5XwsCeRmkSYKSdxJZ2auRzRi+2znv
-            +kPcfKfyQGQGImzkl6niBxzl4jjxiyOyhdDWzwFw34Ets+1wXLlur9mo0v8GLfZG
+            IqicZh0YENwEX++2NQNtt/jaPFCqWUMEVfncvYceutHzZxDZihkbIL/jKs0h/Hjr
-            DEO+1VEzFheZzjM6pVNOWCJdGKR8WEuadjmMMCsux463JiEPtatRJgcAim3ux6uV
+            gL/JNRPQbVD4HXLTQ5dOuAfixJXcN9DfYQGyBg1vOtvpKxR4cD4JirA3KyYjtSWf
-            TbtDNNF9yc9yp7GN8p0/kF3tnD40zaszM8+UznrWbDIFJOrixyVTZ3O9tKcjl55S
+            cYw5fMSalPJ0D3B5pObFZoVJhu8emO3FegSu/tqnZseJuT4mqZqdhpuej17FtkQB
-            m5upCrrbc4rcp+uRiYrsiMZoEg3KvlqtxUw9mHoeVce83l7bziobixzBhnGRtAar
+            il0tNwBdy9pDFXMxiZoGFtVb5Bii7ZU19MPleBBTFO9nNLXE8rCZ/1tJBtR+XUXE
-            zYUKsuu5PPpljV+ghDSIkcTrwLGQnInhlbOxKwDNVb0LiL/1c6dP4zPfZNh+Nvne
+            acglYoZHt3aHAqe1+8iLWAOmE1KkElUMU6/G5yZCpLNNwwgx1jXqGsvITMQQVptJ
-            XhUDx90WnGCFOI2t5tXelAymCBDgak2qCX4IxkxddBhL9lONwYIM8zmszjPQf7r1
+            RMMKfdNGZSC4mcVXT+Ux/uwvw09DO9Mrbf0tjiTTjLProNxUmY4pPMShwCmGsfTa
-            Gx4vFrUTqceEXI+JLiPog4/+gGHX1qzifdHBZR0ZtTi/XJfnhdH3fkJndLA65Pkg
+            si/bZbCui/Jryy02dYWuzA4bpReY+ySMwUGZ9C/0Gn/SBEAdeGQKDhJwgmE6BM1n
-            K4vY6KfomQLEmHurMVzEkoeXH95qSWmfmai4SJmKWqm33VFPZRFoVonnx4dP27SM
+            hMZOh9YPcGSc0n4a9q/CzFqcQ7T8m1uHgqciFG/zW8A+l8bC1H5uxsUwcYPALLxY
-            vZGmfgjOl5eeRk/jdrApJNTkCi/dkGUfNRDffwcYQgjCJaWOs1agchBSC/RLZVWN
+            adAZ6XMPZAZ1L/zp3NO/SpdHp7hQVqx+TZOWvmlZ4XiDrICp4AVK8oZAXKw/nrAu
-            3FPmohgmGa/MBOX3qS1r/5G3XBdKuUxd/NJ2Bw8L0wodteGSEClEiMEXGGZLkOzS
+            k4ZYVsrBKrkLBIMNmM1mxuhi248ELB7V+kN9zDxQZh87J+F2WXk8D4dMwBfRfYzS
-            XAFSk/qmO3tJGViwOVtnb0ZXyTRritW9uR13N+LcaKJhC0PB/m/lXR6IlrfLJXnu
+            XAGiurXdFcCH3i8O5mmE8QY+ZgdKzyAj7hVw8BvXlOIne8pPe3YuZsp68zwdz8MQ
-            AMD7w42wKMeBgGuQQ/Xm/iOVlS2H0kvbm1b1WEZMQUJlEleKSyQKbmQPujoi
+            Gr5oDPaZWhEwFEg+4H+ub5xSIZewY7Ib7FM8bzzfd23vODtZ+0sBoRH+Tq4R
-            =KjtC
+            =m933
             -----END PGP MESSAGE-----
           fp: DC6910268E657FF70BA7EC289974494E76938DDC
     encrypted_regex: ^(password|value|ssh-key|api-key|user|username|privateKey|clientSecret|clientId|apiKey|extraArgs.*|.*Secret.*|extraEnvVars|.*SECRET.*|.*secret.*|key|.*Password|.*\.ya?ml)$

kubernetes/infrastructure/observability/alloy-config.yaml

@@ -0,0 +1,68 @@
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: alloy-config
+  namespace: observability
+  labels:
+    app.kubernetes.io/name: alloy
+    app.kubernetes.io/component: log-collector
+data:
+  config.river: |
+    // -----------------------------------------------------------------------------
+    // Grafana Alloy — log collection only (metrics cluster+host collection is
+    // handled by vmagent/node-exporter/cAdvisor elsewhere).
+    // Forwards pod logs to the in-cluster Loki single-binary.
+    // River config produced by `alloy convert -f promtail` from an equivalent
+    // Promtail YAML (so component names and argument shapes are guaranteed
+    // correct for Alloy v1.18).
+    // -----------------------------------------------------------------------------
+
+    discovery.kubernetes "kubernetes_pods" {
+      role = "pod"
+
+      selectors {
+        role  = "pod"
+        field = "spec.nodeName=" + coalesce(sys.env("HOSTNAME"), constants.hostname)
+      }
+    }
+
+    loki.process "kubernetes_pods" {
+      forward_to = [loki.write.default.receiver]
+
+      // CRI-style log lines on the host: {"log":"...","stream":"stdout","time":"..."}
+      stage.cri { }
+    }
+
+    discovery.relabel "kubernetes_pods" {
+      targets = discovery.kubernetes.kubernetes_pods.targets
+
+      rule {
+        source_labels = ["__meta_kubernetes_namespace"]
+        target_label  = "namespace"
+      }
+      rule {
+        source_labels = ["__meta_kubernetes_pod_name"]
+        target_label  = "pod"
+      }
+      rule {
+        source_labels = ["__meta_kubernetes_pod_container_name"]
+        target_label  = "container"
+      }
+    }
+
+    loki.source.file "kubernetes_pods" {
+      targets    = discovery.relabel.kubernetes_pods.output
+      forward_to = [loki.process.kubernetes_pods.receiver]
+
+      file_match {
+        enabled = true
+      }
+      legacy_positions_file = "/tmp/positions.yaml"
+    }
+
+    loki.write "default" {
+      endpoint {
+        url = "http://loki.observability.svc.cluster.local:3100/loki/api/v1/push"
+      }
+      external_labels = {}
+    }

kubernetes/infrastructure/observability/alloy-release.yaml

@@ -0,0 +1,19 @@
+apiVersion: helm.toolkit.fluxcd.io/v2
+kind: HelmRelease
+metadata:
+  name: alloy
+  namespace: observability
+spec:
+  interval: 1m
+  chart:
+    spec:
+      sourceRef:
+        kind: HelmRepository
+        name: grafana
+        namespace: observability
+      chart: alloy
+      version: "1.11.0"
+      interval: 1m
+  valuesFrom:
+    - kind: Secret
+      name: alloy-values

kubernetes/infrastructure/observability/alloy-values.yaml

@@ -0,0 +1,57 @@
+# Grafana Alloy
+# Runs as a DaemonSet on every node (tolerates the failover taint).
+# Tails pod logs from /var/log/pods, attaches K8s labels (pod, namespace,
+# service, container, image), and pushes them to Loki at
+# http://loki.observability.svc.cluster.local:3100 (in-cluster, no auth).
+# Use the externally provisioned ConfigMap (alloy-config) carrying config.river.
+alloy:
+    configMap:
+        create: false
+        name: alloy-config
+        key: ENC[AES256_GCM,data:aevkZsfe4c4H34BJ,iv:4vrNZWwYWaxJUKGbNlnZfk0NM182nBQZuwvqUPslKNA=,tag:lh1iILM3uQAt0An6kN6EHg==,type:str]
+    # Mount host log paths so the DaemonSet can tail pod/container logs.
+    mounts:
+        varlog: true
+        dockercontainers: false
+    stabilityLevel: generally-available
+    tolerations:
+        - key: ENC[AES256_GCM,data:zOMWiA9dG+8o,iv:yHp9aBczV8nVRGlJmsBaKs9WdmE3+ii8gWOaj5Xok6M=,tag:8c5ZqBT+vvC0WCW8mi0c9g==,type:str]
+          value: ENC[AES256_GCM,data:2Q04VR9xLbQ=,iv:6+tQXx3IGUIcrse+poG4X/4pS5dN2dKvhb79sCy+nfA=,tag:jb/TZmGpvdsWnKoC5NdCow==,type:str]
+          operator: Equal
+          effect: NoSchedule
+        - key: ENC[AES256_GCM,data:aVq8GU1ZDg5m9XDEI3PoVbbykU36SnKpEgrohvZ0IBmH47qDKg==,iv:eMcHRO5mGp9hhY4KnuBmdPTjd/9lqtesjA/RRSu4QMw=,tag:2jaGuogkLcIuye2meJQMDQ==,type:str]
+          operator: Exists
+          effect: NoSchedule
+    resources:
+        requests:
+            cpu: 50m
+            memory: 64Mi
+        limits:
+            cpu: 200m
+            memory: 256Mi
+sops:
+    lastmodified: "2026-07-28T13:46:53Z"
+    mac: ENC[AES256_GCM,data:u6J/EIP5aFtOi8ZtT8BPn978wJAEL1E3r0WcmG5WIsWKJBZMMQHUYRmHgkS6n0/UMtczWoCdXlxObS73ZQMSSkdtagQHE/HPh69dDdGkjxyM4mlP+YMwcnIQkxpJBAQTERXA+SVoWCZb7ztKbiJS+zPjOCfA95ixQINmBCdUgEc=,iv:WOXngWu3uNWYp18/xWghGJyONBrq3SSnPDwbjHBDozU=,tag:1fasTfuLBGXBCo9JPytsJw==,type:str]
+    pgp:
+        - created_at: "2026-07-28T13:46:53Z"
+          enc: |-
+            -----BEGIN PGP MESSAGE-----
+
+            hQIMA7uy4qQr71wiAQ/8DSD0jEn3YE7XHTm4DWdfLHPs1IVg9hyiWSHe5/xH0siD
+            zWonfjcuytw8EHDzv6pQMNYmIpyAmJ92Al4ROJmBV8gZ59+U9goHmw5HgaQAupBJ
+            /gy8iFBteGOHJT+XJPoVb/bu449f7Vle4L3DrOAl6mwxE0WT/7ERoQBeSxL+2OTZ
+            iDPdq24Zy+crDHpoM0GcSnFRLX5UpmL2QXGB8PHyctwSQoDW+sbAY2ME6buOw8HI
+            AremKTRtFm6Pu1Ksdh6ZicaHpLSvLyByPPt2zYhlBQsG3s/SjJDlnakUZHhm7T4o
+            oHzSpKqV7NQbE09ZMyGJrEkl+rfCPeu6jHplW2EX3xjqLWNUF8GHcsDWljogZxXF
+            irBPkyKASEDdkHZgiZobU6BmvQlxILj104xNtUYTTChcKu3s0d5BrkIOGe+Y7rJd
+            fNvMfkmO8/URBgtsXy2xPKo0Jzu/0pZsPEp2lxNeyElwxQqo1CFwhKZPkeQ2eLpu
+            4X7/CMN8KJr6jLX8HhX9AivlCj+2CoT8SB5GAPw4p00eD0GIMWSNyOXjyW/qQZi3
+            82EyX9KIpHjALoOI4/3YgFpUO12AEsjirYKWkUM+IoDp5l+YX59QdM7zUVVyzsTG
+            2EXEdQqYVA7DIcMJb0tHqxDGMGxNayaUsikV7e/kcRVUyDSH+xOfaSUrDTqqv+XS
+            XAFro6848HF5uMIP4pfM+dxzkc9rFF4il0CuGWlscbz5q+Aw9LImGbpNzrWk3Lpi
+            beuwWKFTZ8zab5TH9UF/xqsQeg/nyzQnb3fFPk+owvY7fH6XiYbN5WjKjXGW
+            =kBpp
+            -----END PGP MESSAGE-----
+          fp: DC6910268E657FF70BA7EC289974494E76938DDC
+    encrypted_regex: ^(password|value|ssh-key|api-key|user|username|privateKey|clientSecret|clientId|apiKey|extraArgs.*|.*Secret.*|extraEnvVars|.*SECRET.*|.*secret.*|key|.*Password|.*\.ya?ml)$
+    version: 3.10.2

kubernetes/infrastructure/observability/grafana-repository.yaml

@@ -0,0 +1,8 @@
+apiVersion: source.toolkit.fluxcd.io/v1
+kind: HelmRepository
+metadata:
+  name: grafana
+  namespace: observability
+spec:
+  interval: 1m
+  url: https://grafana.github.io/helm-charts

kubernetes/infrastructure/observability/kustomization.yaml

@@ -0,0 +1,26 @@
+apiVersion: kustomize.config.k8s.io/v1beta1
+kind: Kustomization
+namespace: observability
+resources:
+  - namespace.yaml
+  - vmstack-repository.yaml
+  - grafana-repository.yaml
+  - loki-repository.yaml
+  - vmstack-release.yaml
+  - loki-release.yaml
+  - alloy-release.yaml
+  - alloy-config.yaml
+  - vm-internal-ingress.yaml
+  - loki-internal-ingress.yaml
+secretGenerator:
+  - name: vmstack-values
+    files:
+      - values.yaml=vmstack-values.yaml
+  - name: loki-values
+    files:
+      - values.yaml=loki-values.yaml
+  - name: alloy-values
+    files:
+      - values.yaml=alloy-values.yaml
+configurations:
+  - kustomizeconfig.yaml

kubernetes/infrastructure/observability/kustomizeconfig.yaml

@@ -0,0 +1,6 @@
+nameReference:
+  - kind: Secret
+    version: v1
+    fieldSpecs:
+      - path: spec/valuesFrom/name
+        kind: HelmRelease

kubernetes/infrastructure/observability/loki-internal-ingress.yaml

@@ -0,0 +1,23 @@
+# Ingress exposing the Loki push API (loki:3100) through Traefik as
+# loki-internal.lan — used by the docker-host promtail to push container logs
+# into the central Loki.
+apiVersion: networking.k8s.io/v1
+kind: Ingress
+metadata:
+  name: loki-internal
+  namespace: observability
+  annotations:
+    external-dns.alpha.kubernetes.io/enabled: "true"
+spec:
+  ingressClassName: traefik
+  rules:
+    - host: loki-internal.lan
+      http:
+        paths:
+          - path: /
+            pathType: Prefix
+            backend:
+              service:
+                name: loki
+                port:
+                  number: 3100

kubernetes/infrastructure/observability/loki-release.yaml

@@ -0,0 +1,19 @@
+apiVersion: helm.toolkit.fluxcd.io/v2
+kind: HelmRelease
+metadata:
+  name: loki
+  namespace: observability
+spec:
+  interval: 1m
+  chart:
+    spec:
+      sourceRef:
+        kind: HelmRepository
+        name: grafana-community
+        namespace: observability
+      chart: loki
+      version: "18.5.4"
+      interval: 1m
+  valuesFrom:
+    - kind: Secret
+      name: loki-values

kubernetes/infrastructure/observability/loki-repository.yaml

@@ -0,0 +1,9 @@
+apiVersion: source.toolkit.fluxcd.io/v1
+kind: HelmRepository
+metadata:
+  name: grafana-community
+  namespace: observability
+spec:
+  interval: 1m
+  type: oci
+  url: oci://ghcr.io/grafana-community/helm-charts

kubernetes/infrastructure/observability/loki-values.yaml

@@ -0,0 +1,83 @@
+# Loki Helm values
+# Uses Monolithic deployment mode with filesystem storage on a Longhorn PVC.
+loki:
+    deploymentMode: Monolithic
+    # Filesystem storage — chunks + rules on the PVC.
+    storage:
+        type: filesystem
+        filesystem:
+            chunks_directory: /var/loki/chunks
+            rules_directory: /var/loki/rules
+    # Required schema config. tsdb store + filesystem object store, schema v13.
+    # See https://grafana.com/docs/loki/latest/operations/storage/schema/.
+    schemaConfig:
+        configs:
+            - from: "2024-04-01"
+              store: tsdb
+              object_store: filesystem
+              schema: v13
+              index:
+                prefix: index_
+                period: 24h
+    # For monolithic mode, point tsdb_shipper at the local store (no index gateway).
+    storage_config:
+        tsdb_shipper:
+            active_index_directory: /var/loki/index
+            cache_location: /var/loki/index-cache
+    monolithic:
+        persistence:
+            storageClassName: longhorn
+            accessModes:
+                - ReadWriteOnce
+            size: 10Gi
+        resources:
+            requests:
+                cpu: 100m
+                memory: 256Mi
+            limits:
+                cpu: "1"
+                memory: 1Gi
+# 7d log retention (user can tune later).
+limitsConfig:
+    retention_period: 7d
+# Disable distributed components + gateway + minio (monolithic only).
+test:
+    enabled: false
+backend:
+    replicas: 0
+read:
+    replicas: 0
+write:
+    replicas: 0
+singleBinary:
+    replicas: 0
+gateway:
+    enabled: false
+minio:
+    enabled: false
+sops:
+    lastmodified: "2026-08-01T20:19:42Z"
+    mac: ENC[AES256_GCM,data:KYZQGIimeeOlt2lltCrKttLXzk2b7DOT8KNOhpeVqF/VNNcVxP7Srz6nCR5J7dI6wFSePspof5gNJEZlzP3EGJFQaGBf34hOGhHBUtlcscX8vj2JYmmJ85B+pylHjQNPHaXqLP4tAwIFfhcJi368K+GEAu8tztkxJKk7dtOOyrg=,iv:V24R3eSzW0edwPnPSSwEQqqSfiNrK6+ZyQ368xpJKhg=,tag:LmaMhhGwRIIG+SxTUJf5gQ==,type:str]
+    pgp:
+        - created_at: "2026-08-01T20:19:42Z"
+          enc: |-
+            -----BEGIN PGP MESSAGE-----
+
+            hQIMA7uy4qQr71wiAQ/+PDeQtUHMysWOkTjqU8vNAl1IQzDrwHskeBmLXLI99kdc
+            hU0afTjzUbomFdMRGvf0QZrcFbnD5OHskw5KRleDbYxOp4aXjzX0RsqSe9elTFAj
+            ml4XG0UrEV4Ygi4mTNOGzmINaIjobcn0tQd6nZFPUUA//udBcHsczs9x9Rrf+YTv
+            XrUdNAhveRlowvbvfNrDIh1OsOgyWyHJEJsKiLbt2b7e47PKhhRQzET+bNwhXzTb
+            GMY/w3rVYPddqx1wi5HPtYlhL3emYlmMYLLi1ITr1fKvAsgGKSJpaGS6yu+YkyiB
+            AJHou+2aCUy1TVk8U/Ne9TSrhsV1AJSIbtr8+iyKAtSsSuodam0Pyzn59uqL4nNd
+            1S4Nkvm1UDYv97hmSQdAzTGs+Aql6HyB/Wyi+rkMZQ0fl7+hdX5/eEDYZzijD0Nu
+            Mqk1iFbTn/rXir71VRWNttycIM1PfGAOUmLoyOR+TrL5Ecy0Sx9yx2DS9DecnqtS
+            39YI6UNqiEYdvW8Y+XELSFI/B5cjw7C1A5S9tecUxoOIBGZZSeRQa2WDSWb1dYky
+            Rzvncoeemgjkp8CHQhSqrylB+I4s0Y1HMUWR/q0wxew9juNEIP8ZUkfKcueuCYlm
+            5ui/RaeYatMFo8hg+DrxEtQIkbBZ/ILss/QzyHZZG3RMq0i9s/PIn5yipVhOgorS
+            XAFEeOrVaKpN06LfNP7/HTI+D/5HJL++Twuyz9fwC2Eb5saIIB6gpywiICH8HcD5
+            MRmmhl18x2YU+JdmkAcWNfH5+JHVnx8o+D/y9JqeCf6sle/9LyToazJv58Tc
+            =2u7h
+            -----END PGP MESSAGE-----
+          fp: DC6910268E657FF70BA7EC289974494E76938DDC
+    encrypted_regex: ^(password|value|ssh-key|api-key|user|username|privateKey|clientSecret|clientId|apiKey|extraArgs.*|.*Secret.*|extraEnvVars|.*SECRET.*|.*secret.*|key|.*Password|.*\.ya?ml)$
+    version: 3.10.2

kubernetes/infrastructure/observability/namespace.yaml

@@ -0,0 +1,6 @@
+apiVersion: v1
+kind: Namespace
+metadata:
+  name: observability
+  labels:
+    pod-security.kubernetes.io/enforce: privileged

kubernetes/infrastructure/observability/vm-internal-ingress.yaml

@@ -0,0 +1,26 @@
+# Ingress exposing the VictoriaMetrics vminsert endpoint (vmsingle:8428)
+# through Traefik as vm-internal.lan — used by the docker-host vmagent to
+# remote_write metrics into the central VictoriaMetrics.
+# external-dns auto-creates the vm-internal.lan A record pointing at the
+# Traefik LoadBalancer (see kubernetes/system/external-dns).
+apiVersion: networking.k8s.io/v1
+kind: Ingress
+metadata:
+  name: vm-internal
+  namespace: observability
+  annotations:
+    external-dns.alpha.kubernetes.io/enabled: "true"
+    # No chart ingress is available for vmsingle, so declare it here.
+spec:
+  ingressClassName: traefik
+  rules:
+    - host: vm-internal.lan
+      http:
+        paths:
+          - path: /
+            pathType: Prefix
+            backend:
+              service:
+                name: vmsingle-vm-victoria-metrics-k8s-stack
+                port:
+                  number: 8428

kubernetes/infrastructure/observability/vmstack-release.yaml

@@ -0,0 +1,19 @@
+apiVersion: helm.toolkit.fluxcd.io/v2
+kind: HelmRelease
+metadata:
+  name: vm-victoria-metrics-k8s-stack
+  namespace: observability
+spec:
+  interval: 1m
+  chart:
+    spec:
+      sourceRef:
+        kind: HelmRepository
+        name: vm
+        namespace: observability
+      chart: victoria-metrics-k8s-stack
+      version: "0.87.0"
+      interval: 1m
+  valuesFrom:
+    - kind: Secret
+      name: vmstack-values

kubernetes/infrastructure/observability/vmstack-repository.yaml

@@ -0,0 +1,8 @@
+apiVersion: source.toolkit.fluxcd.io/v1
+kind: HelmRepository
+metadata:
+  name: vm
+  namespace: observability
+spec:
+  interval: 1m
+  url: https://victoriametrics.github.io/helm-charts/

kubernetes/infrastructure/observability/vmstack-values.yaml

@@ -0,0 +1,199 @@
+# victoria-metrics-k8s-stack Helm values 
+# The chart installs the VictoriaMetrics operator + CRDs and creates VMSingle,
+# VMAgent, VMAlert, VMAlertmanager CRs.
+# ---------------------------------------------------------------------------
+# VictoriaMetrics single-binary (the metrics database)
+# ---------------------------------------------------------------------------
+vmsingle:
+    enabled: true
+    spec:
+        retentionPeriod: 7d
+        storage:
+            storageClassName: longhorn
+            accessModes:
+                - ReadWriteOnce
+            resources:
+                requests:
+                    storage: 20Gi
+        resources:
+            requests:
+                cpu: 100m
+                memory: 256Mi
+            limits:
+                cpu: "1"
+                memory: 1Gi
+# ---------------------------------------------------------------------------
+# vmagent — scrapes node-exporter, kube-state-metrics, kubelet, etc.
+# The docker host runs its own vmagent that remote_writes through
+# vm-internal.lan (the Traefik Ingress → this vmsingle).
+# ---------------------------------------------------------------------------
+vmagent:
+    enabled: true
+    spec:
+        selectAllByDefault: true
+        scrapeInterval: 20s
+        tolerations:
+            - key: ENC[AES256_GCM,data:SyutB0y9QtqQ,iv:EKRXNGtUovQYsESHFQTGygP09O2UWtLuBEQQcYZm3Oc=,tag:zh7CYOZleSgbbQhYZvcT+w==,type:str]
+              value: ENC[AES256_GCM,data:So+8xmHYSxs=,iv:7sfUBAaMZd7imjVx14lusVn58cJQ4HpTkaPy5Eal2dY=,tag:hRBjT6wuEfG3Z+ZiPJhwQQ==,type:str]
+              operator: Equal
+              effect: NoSchedule
+            - key: ENC[AES256_GCM,data:lj9OurynU8pRt4/MhgyevjmV0b7Zq+WG/9+zRRvPeA+RYZsV+A==,iv:LjuhErDT0bDlLcpo/AlNykz8Xu89B+nx1ja+HkPNXwo=,tag:5xM6XHrCdEejE7ituQ8W4Q==,type:str]
+              operator: Exists
+              effect: NoSchedule
+# ---------------------------------------------------------------------------
+# kube-state-metrics + node-exporter (metrics sources)
+# ---------------------------------------------------------------------------
+kube-state-metrics:
+    enabled: true
+prometheus-node-exporter:
+    enabled: true
+# ---------------------------------------------------------------------------
+# Grafana (UI) — served behind ingress at grafana.lan.
+# ---------------------------------------------------------------------------
+grafana:
+    enabled: true
+    persistence:
+        enabled: true
+        storageClassName: longhorn
+        accessModes:
+            - ReadWriteOnce
+        size: 5Gi
+    adminUser: admin
+    # SOPS encrypts this when the file is processed.
+    adminPassword: ENC[AES256_GCM,data:vpPvJA/PktMnsg/IovYfsD4lDBndUw==,iv:b4eLMEcUMJj4BwyeYw1kw2sZuieyr2xL9et+R7ss0BU=,tag:u0RcTp6mkQNAGkblG4/xww==,type:str]
+    # Provision a Loki datasource alongside the chart's default VictoriaMetrics
+    # datasource so metric/log correlation works in one UI.
+    datasources:
+        datasources.yaml:
+            apiVersion: ENC[AES256_GCM,data:Uw==,iv:KYTMBV53yMbojMaqNinYpm/Uj5ylV86jcdJTdxuy+G0=,tag:IbIpiSBQFw7YSTQg0WRGqg==,type:int]
+            datasources:
+                - name: ENC[AES256_GCM,data:BPfR8Q==,iv:MkMhH3IJfBZzbIcqQw7y6bfJB/4Ew4feZNugjovk5O0=,tag:jaa43tzIXkE8kWZp2R/kcg==,type:str]
+                  type: ENC[AES256_GCM,data:WWncww==,iv:hPqJE0KCidSBnf11wxFH+Je+YKztWmeToBqi39Qf4ZQ=,tag:c9enJ2VgQqDJVbasvgG+Dw==,type:str]
+                  url: ENC[AES256_GCM,data:YkS66tpWXpW62Dtful0KNCwoZ3rG3nQtDBUsNQUwdQ+eqtG38JTrAudQqnsRD51q,iv:Mu3Sc6T2xypdot9dA1DHc8nD7UI8372nqlveLITR82E=,tag:1VmmIg+xR9qjWcvPKVYsJQ==,type:str]
+                  access: ENC[AES256_GCM,data:af3GpjE=,iv:MAYKfklF+wCC/XN+4snebjPAnW4KpYzaMMGYds7rG18=,tag:X+oFiDAiCb1f7cfREqFT7A==,type:str]
+                  isDefault: ENC[AES256_GCM,data:Zgu4Hn8=,iv:nnVB/in75nPAQos0Dfl6abBrHwT8tCt5X7lcCYBZbLc=,tag:u1UvkSj6SjcMm06NzKMGDA==,type:bool]
+                  jsonData:
+                    maxLines: ENC[AES256_GCM,data:88JJqA==,iv:lKsZa024I6223UEu0SHuZ3N9LZrzw0WBTfC+L2/3ljI=,tag:YvqieXK36xMmC/sSJ11NIg==,type:int]
+    ingress:
+        enabled: true
+        ingressClassName: traefik
+        annotations:
+            external-dns.alpha.kubernetes.io/enabled: "true"
+        hosts:
+            - grafana.lan
+        path: /
+        pathType: Prefix
+        tls: []
+# ---------------------------------------------------------------------------
+# vmalert — evaluates VMRule CRs against VictoriaMetrics, forwards firing
+# alerts to Alertmanager. selectAllByDefault picks up all VMRules in the
+# namespace (including our vmalert-rules.yaml VMRule CR).
+# ---------------------------------------------------------------------------
+vmalert:
+    enabled: true
+    spec:
+        selectAllByDefault: true
+        evaluationInterval: 20s
+        notifiers:
+            - url: http://vm-victoria-metrics-k8s-stack-alertmanager.observability.svc.cluster.local:9093
+        resources:
+            requests:
+                cpu: 50m
+                memory: 64Mi
+            limits:
+                cpu: 200m
+                memory: 128Mi
+# ---------------------------------------------------------------------------
+# Alertmanager — 1 replica; inline config with a single n8n webhook receiver.
+# n8n runs on the docker host and fans out to email/Telegram/whatever.
+# ---------------------------------------------------------------------------
+alertmanager:
+    enabled: true
+    spec:
+        replicaCount: 1
+        port: "9093"
+        selectAllByDefault: true
+        storage:
+            volumeClaimTemplate:
+                spec:
+                    storageClassName: longhorn
+                    accessModes:
+                        - ReadWriteOnce
+                    resources:
+                        requests:
+                            storage: 2Gi
+        resources:
+            requests:
+                cpu: 50m
+                memory: 64Mi
+            limits:
+                cpu: 200m
+                memory: 128Mi
+    config:
+        route:
+            receiver: n8n-webhook
+            group_wait: 30s
+            group_interval: 5m
+            repeat_interval: 4h
+        receivers:
+            - name: n8n-webhook
+              webhook_configs:
+                - url: http://n8n.lan/webhook/observability-alert
+                  send_resolved: true
+extraRules:
+    node-alerts:
+        groups:
+            - name: node
+              rules:
+                - alert: HighNodeCPU
+                  expr: 100 - (avg by (instance) (rate(node_cpu_seconds_total{mode="idle"}[5m])) * 100) > 80
+                  for: 10m
+                  labels:
+                    severity: warning
+                  annotations:
+                    summary: High CPU on {{ $labels.instance }}
+                    description: CPU usage above 80% for 10 minutes.
+                - alert: HighNodeRAM
+                  expr: 100 - ((node_memory_MemAvailable_bytes / node_memory_MemTotal_bytes) * 100) > 85
+                  for: 10m
+                  labels:
+                    severity: warning
+                  annotations:
+                    summary: High RAM on {{ $labels.instance }}
+                    description: RAM usage above 85% for 10 minutes.
+                - alert: LowDiskSpace
+                  expr: |
+                    100 - ((node_filesystem_avail_bytes{mountpoint!~"/run.*|/var/lib/docker.*"} /
+                            node_filesystem_size_bytes{mountpoint!~"/run.*|/var/lib/docker.*"}) * 100) > 85
+                  for: 10m
+                  labels:
+                    severity: warning
+                  annotations:
+                    summary: Low disk space on {{ $labels.instance }} {{ $labels.mountpoint }}
+                    description: Disk usage above 85% for 10 minutes.
+sops:
+    lastmodified: "2026-07-28T16:39:09Z"
+    mac: ENC[AES256_GCM,data:Ymxmmkui0Nuv35Yq6v4BXM4xGhvjAufexW7fIm9fUIdTTXZ3e10PitNCZXKiaDwwCJfdEkNZri7jPAYzhTzKMIdcTKnOMImsMjdiJZwRwwbKcYVay0A6KSLJ8eFjDRRYhlCXbsAIRlQHpFh4UkssKdu40mIiudU0qIVm04ppbno=,iv:R8kpITPYSbQ1yZw5IK0fArssxUtIjNmvzknOrLJQcuk=,tag:/CRcr1VZAIKIPbcag05uTw==,type:str]
+    pgp:
+        - created_at: "2026-07-28T16:39:08Z"
+          enc: |-
+            -----BEGIN PGP MESSAGE-----
+
+            hQIMA7uy4qQr71wiARAAmOxyx5eH5Jykp/GDL79REgf61iGujtPiUVy8CT3o7O5k
+            tTGma6rRXK5wZV/iGN3wOQehF19Oy2iYwMfxilRwpB69RFwyKPKdHDDWMWNXv5Th
+            wJnNjVjWZ0xiRautWPn/TjDOJZ6j6p7lQ3IcQ3vzuap1RVRIXv081DxJGR9JbeYl
+            khmrl1+zecwxS87JK+fLko9NIBb5JgFEtEf1RvLyFak8nZu5cNCu+PCykALvUaO7
+            V5NAaF1aV9cdz8G9TZWmHZsIlDDE9/oro2aLR7cr8uxnKQhWerttqvyDYNwYzNKW
+            qgsUplTxTmyuc9DdnLDfJ55Ncuv9qS5/4/p5GpsHrteD6oiPNjXaxR8cBVFLvGwp
+            kQb/VZgOGAnkYlMCUG8APiwRhbpAlyy51hKBqVh+a9iRlzZEWHKlxcvIrcFAQaHt
+            sDXyg36F0GCoHiMAqHKG0kRcbTV7Sq0JZVoLU/qxvaDfqvyCZO7X/U/YsSWNbWL0
+            emyBfeSvBX9h62NJx2aw8xKyiL+aeVfzSKAaXHMd/r4gKxBw/kVHAljKwRzbL85j
+            UHeLUvqVkXs1mtIqGgUnpCcAVuV83omgcOwF44FitG/f9GCx+UWTvScwaw5ywuG3
+            J/4fOaA3K0Z4e/mOejARd4jLSPktSRpA01k9kjYY8UBn7TzQ0LrMLj5dxh8BpWrS
+            XgECZapi+q6IjIses7+JqkWiP25hcglhqK3LAiUapA0c5UmAa5Vf7jfnkFm8eUdZ
+            bA7ZnhX8nxK9n7JdSxrZhrbPWMO8hF/XKi7KTZMvgXFlb4/qP96y7oeqzKYcnvE=
+            =pmSq
+            -----END PGP MESSAGE-----
+          fp: DC6910268E657FF70BA7EC289974494E76938DDC
+    encrypted_regex: ^(password|value|ssh-key|api-key|user|username|privateKey|clientSecret|clientId|apiKey|extraArgs.*|.*Secret.*|extraEnvVars|.*SECRET.*|.*secret.*|key|.*Password|.*\.ya?ml)$
+    version: 3.10.2

kubernetes/kustomization.yaml

@@ -19,11 +19,12 @@ resources:
   - code/gitea
   - code/harbor
   # Home
-  - home/home-assisant
+  - home/home-assistant
   - home/zigbee2mqtt
   # Infrastructure
   - infrastructure/authentik
   - infrastructure/mail/stalwart
+  - infrastructure/observability
   # Personal
   - personal/linkwarden
   - personal/notesnook

kubernetes/personal/notesnook/values.yaml

@@ -1,7 +1,7 @@
 instance:
     name: vhaudiquet-notesnook
 api:
-    secret: ENC[AES256_GCM,data:C3mpoEG6y6IShpX1+o9eNn8NACaKy8s1xw5tY1/ncBzqaKrK3YiE7K0rl4d6Bq6q,iv:rGWxSmV98ef8Qx1jkVbQEKPkFmGEaCOXXFFZ4I1US7s=,tag:VBDOhPRTRRhFu5cU024Sqg==,type:str]
+    secret: ENC[AES256_GCM,data:R3pAcgHfCeBrgNJ8Fc1lvNU3sUcWicIglZDtInaOzw11s43GaPPiR6t3lGiiN5oA,iv:xUqXD7tMrjxzBwIk051d5DeVOGh57O19ogt5RPHQppY=,tag:sZ+HgFbWv5jpia+D/tKqnw==,type:str]
     knownProxies: 10.0.0.0/8
     disableSignups: true
 publicUrls:
@@ -10,8 +10,8 @@ publicUrls:
     monograph: https://n.vhaudiquet.fr
     attachments: http://localhost:9000
 smtp:
-    username: ENC[AES256_GCM,data:C4dTnVaJCwxqTdevLJ+a9eJOWPk=,iv:9iHoQzZjHjmOuaoOWdedPHuv06MqtXZXJhWGiTdzhwE=,tag:xDL+WInm/Ms/LuZi53JuHA==,type:str]
+    username: ENC[AES256_GCM,data:FXfsyRX1PD7rG8TKGtQD5yj33BY=,iv:rb++wIa9mr0i9OJ5JMv27aETxuyUfCDOKGwp2u0wUUA=,tag:ZYg/1vHTGypdw597lU4Itw==,type:str]
-    password: ENC[AES256_GCM,data:tIkKqwVBy94oqFJH0V8=,iv:cOKiwDhngz6mnZlD+XSfWFg1KZa+UCkhXKBgjK7IdnE=,tag:91d/aUlfeHbJqtRWPmTskQ==,type:str]
+    password: ENC[AES256_GCM,data:3lYaORGHK988XBnwY64=,iv:Tc307c0DS6kTXGySSgJMhaceDqps+pyNlKqcab9UHho=,tag:h7e4pMLi5sVyqIgNVg8tZg==,type:str]
     host: mail.vhaudiquet.fr
     port: 465
 ingress:
@@ -37,28 +37,68 @@ ingress:
               paths:
                 - path: /
                   pathType: ImplementationSpecific
+# Failover tolerations + affinity — allow scheduling on p330 during r740 outage
+# Applied to all notesnook components (identity, notesnook, sse, monograph, minio)
+tolerations:
+    - key: ENC[AES256_GCM,data:VxDGa/LUCm1t,iv:6MU3sqjnkm0Gxdqt8tD01Fqxcg9oK42f8K9aHDQMDY4=,tag:9liOuvSqHNdfVcRfephFpA==,type:str]
+      operator: Equal
+      value: ENC[AES256_GCM,data:3HkYc/Zk+50=,iv:P5+YjonnfjVkMmO+6/HqxEO/AyLxNlIIOyqi3gyYXRs=,tag:3XtUzwXJ3vQiMEe+3R7lAw==,type:str]
+      effect: NoSchedule
+    - key: ENC[AES256_GCM,data:JqPxmvAU1+nMu4YBnp1deCH/y6byaGc4hrfKbVFVvqGdr1XgQA==,iv:/YKU8LhYfzp3gUOojTTfPCKqLgL5qSncDYTKlLmyrqk=,tag:wFPEWson8XZPNPV3kXnGPw==,type:str]
+      operator: Exists
+      effect: NoSchedule
+affinity:
+    nodeAffinity:
+        preferredDuringSchedulingIgnoredDuringExecution:
+            - weight: 100
+              preference:
+                matchExpressions:
+                    - key: ENC[AES256_GCM,data:dk8VoO7VNzvfdQhZyP+WnhO46ak=,iv:tkH2ElerO8BIALsvUNJIl/qnlGVdOS76DPXRi2J+ovg=,tag:XYMz3w8GthdOV3x6+Ybkxw==,type:str]
+                      operator: NotIn
+                      values:
+                        - "true"
+# Subchart tolerations — mongodb must also tolerate p330
+mongodb:
+    tolerations:
+        - key: ENC[AES256_GCM,data:1TnuKgSBqYCU,iv:nSK4pFqVnVxioaZSp9rwKFyEyIbyxeEs/2luzKjnwLw=,tag:jgA6oDj0foG2YjAieJLE8Q==,type:str]
+          operator: Equal
+          value: ENC[AES256_GCM,data:9sKq4+CgI5k=,iv:3jA01Zaohix3HBe8FnI9VzlLXli1ZThWw6JU5hhysYQ=,tag:8qdivZP7PwGC4gnVNuUupw==,type:str]
+          effect: NoSchedule
+        - key: ENC[AES256_GCM,data:8PbTBL+v3LluefFLYxXUmgQw/ROlIL4nLJVDqrVLvFvrtUqWbw==,iv:khQcqyoBLgsDk3hndnqKGmSdO9e++5o0M0PTNq6jf9g=,tag:8A4f/zp7DimjAKGjdPbvuw==,type:str]
+          operator: Exists
+          effect: NoSchedule
+    affinity:
+        nodeAffinity:
+            preferredDuringSchedulingIgnoredDuringExecution:
+                - weight: 100
+                  preference:
+                    matchExpressions:
+                        - key: ENC[AES256_GCM,data:Q4SFghGuAo8RjD6OYqUxt9sYlfU=,iv:TbR9TmGAmzSZg9+/ZDzsGmt4yA+KQBIDPnUQbQOogTw=,tag:FbEPi3gxlpJuxuerAruTmw==,type:str]
+                          operator: NotIn
+                          values:
+                            - "true"
 sops:
-    lastmodified: "2025-12-26T18:18:01Z"
+    lastmodified: "2026-07-25T12:44:32Z"
-    mac: ENC[AES256_GCM,data:Xy9P+Ifuz18apN7GoYdehc2bzTjUKMJAT7f8HZNTnvV/wkZEt4EUGJL2WGex12nYQyj6Ut+I9pwFwwX5m0oLO82s1zS2DK3BiaxFa6LFJ2VDUthKt8h9ZTNeT+2P5S5cOvEMvS6tljX8y8/HCUwVMCXGMNCIl8RtWo1Q9CgLjrw=,iv:6MdJwJh3xVrXX6sKCQMAEIpdOD8E0V6+305xcaQnnMI=,tag:5JE9M09u6ukPoQBzT9g0sA==,type:str]
+    mac: ENC[AES256_GCM,data:+vnDAlFBB7zP4YnxE9fqNa9dawgeuCB2vTet4yh5Cgqp36p1uPxg68uaBeZPrBFBj2fjNs3g/C22ny0/ui2wb5+pGorzXKxVpjWX40hMXyiCQwJXejlZCiXKAFLmOQMiv1oJQQ0PqhAeOckhI5dEc0D2PAUynXHAXh49xNrzcVQ=,iv:St70rkrQpCA81MpPpxyUtXQyvGccy8pgEJUOrdF7ihs=,tag:l0YZdLDkfu1miVMpDJD4FQ==,type:str]
     pgp:
-        - created_at: "2025-12-26T18:18:01Z"
+        - created_at: "2026-07-25T12:44:32Z"
           enc: |-
             -----BEGIN PGP MESSAGE-----
 
-            hQIMA7uy4qQr71wiAQ/+KZ6WDFaDTZVt6n+pu1hwxhNC9ShCSSLkiHlZjtX9DW8O
+            hQIMA7uy4qQr71wiARAAkDKwwK82i30IyWe3/35JGxQJ1NoFc6bbKnFBW30JFR2o
-            eizwH6GAG8LQeKYr61aXXVwxsqDdM9gx/i0ynDO4fflqEQniEjlFzBeVhsqOf3CX
+            7vVl1e7eBfWwIZ/yD+fHkyiskYe+UWT/ZDph526JKXJxuiZ1XPzCBQ2iD8hhABNi
-            rL5WImv0LSe9KjFNAZvfBhtUVENfN/Tu6RGagNb98ChOoH10MvmrzQ5t754AKPVM
+            /8YrlLd/R3kjGebXTAZ6/+/DijuAeiVxyaiMDkmo2YXz8x2+nWXV7I5mxMqDdTwh
-            hB2f5kmOn0evFQQftYyXkQqzLvhlqdGGfBnCutXDoLz67RQwB37npzXx/c+GtlOX
+            fIldLmX9IJKDJrAp5RpFn+g5PIwspQJmA3s86FbMAolLAHAUJ3zNCy1D17LFK+kw
-            Msmp4wiYqlX9v97vjLHUX/ZeBtzm38DzUw+OTcnDPbsdZpeUMqYnpkIl8GIgSd2D
+            X34yHZPwPcbP7udIhGMTZvpUawPkxXBBJfTj6ElBWhR+8OHP6zgEYTZyY/evMQJO
-            2jZUJU4wYyOCYCnfw0+zXu3O5/1bzZ9fdc5FH6A+OlJUFbMLPXx7n9uhVOztxQ1g
+            fr7n0rWv0gtlNymfX/xyRo8JuVAW8Qh2bOkFh7+fmlMAtdrg6MXVLyGIiQBh1NIN
-            5ajZKkszSWk58IvaTueDL2QjxhJLop1y4JITgzOhiSPserXrizGn765bVEAtY8kR
+            8hQz++ZrOMLx1hoxqehZ/lKUq6ZJTzqXcfMZHipFUxlFVVA9zD+qM0TUxiliTdiy
-            zgULAG5f41zeoJzzdMyuCFUlIJec4DFigbe7fRul4sJEjeeHtDdyINWuUXCnS4ZX
+            GJlq61xwmvRafbUrpmNcon202jJCYJ5g/K/PFmSx8yTa7cIp0FOpNbZIrTmnFffI
-            e8/ZZ0+z98IMm7Lstb75rUezTUdlIxfJ+EtAX3L1Bdb4yJZ3i3E1VTQKWxaZWrcc
+            bJv0gYk/3f0EiTEjKAnl915IUj7KSRGWS5nH5RRuVLw575+glnDwhB9gax/uZN39
-            zF5nOIZsHgD22Um8fA4mDlhX5/ygiKMHxZe2pdwps4F7H0rRzVLlM5n4g+71WeMe
+            gUuRiMm3MeQJ0FX6lYOpu1+3dnEZlaCMhJw4Lif571bOEi4KBcMAuHbLBfMmoL5N
-            0eT4atgKPICzMNkyRJ/KWrtk4c6Kq7sSqpQKlcFsRLMga1ZBt0LQwFW+G7PU30vS
+            dfuwNq3s+kWfu4lAYwxrX3ThvoRa7UABdmvJHqo+raAlqShjKjNxLs9IAL41+m7S
-            XAFljxtoFrbdc8Yl37cQh9XCBqXMqUXSNd9t1hsgv+Vsrmv6ntPVStep8Fod/n16
+            XgF83/ecTRqQ2f4gio7drRr6K5QIPrG/UVuuzghqfDTdzgNXRbTObjDxiifBjcGd
-            nCACAKTOe//z0OmwUXxdUCIqT/N+XEA8tOrHS+HsbB5MCfnexLDul/yUfviZ
+            MBxTQvtvxjLObgO5s7RWZe78/1/4T8hOKYW4qGGG7iO/lggqPljHMVKyLHUzadE=
-            =yK0X
+            =UGsX
             -----END PGP MESSAGE-----
           fp: DC6910268E657FF70BA7EC289974494E76938DDC
     encrypted_regex: ^(password|value|ssh-key|api-key|user|username|privateKey|clientSecret|clientId|apiKey|extraArgs.*|.*Secret.*|extraEnvVars|.*SECRET.*|.*secret.*|key|.*Password|.*\.ya?ml)$

kubernetes/production/vhaudiquet-fr/values.yaml

@@ -41,28 +41,48 @@ nfs:
     storageClassName: ""
     # Allow the pod to start even when NFS is unreachable (uses emptyDir + initContainer)
     softFail: true
+# Tolerate failover nodes
+tolerations:
+    - key: ENC[AES256_GCM,data:whCOkQ/xDUyG,iv:MU1iyLWTp21dUA0Ukn7L4Kdmp4DcD2946/1lHrgNJJg=,tag:VdVgjP831+lV9IdabEYvOQ==,type:str]
+      operator: Equal
+      value: ENC[AES256_GCM,data:IXlCv4VONI8=,iv:/7uiSYGVY85/8nGgnQjeO+hClQPqWEJWOMwEZs1VuzI=,tag:gZI49sfALmsfaxjF4mP4WQ==,type:str]
+      effect: NoSchedule
+    - key: ENC[AES256_GCM,data:7MK/rNoqAhDUJh3O4r0b47PVfT3ItjnnCOKJWZCThzHNIrw0bw==,iv:hUCE6+NwucVw5lwktzOpXRqjG51w4g/AQceyBiduu+E=,tag:DWGdD++4ZptIsSmLpwZAsg==,type:str]
+      operator: Exists
+      effect: NoSchedule
+# Prefer (non-failover nodes) in normal operation
+affinity:
+    nodeAffinity:
+        preferredDuringSchedulingIgnoredDuringExecution:
+            - weight: 100
+              preference:
+                matchExpressions:
+                    - key: ENC[AES256_GCM,data:EDr52ZzUZjtJe+oxleIJZZCYCl8=,iv:ZafjjEMMaWYP7JKAhnRkMB9kFghd5LB593mUELzDLwQ=,tag:4EChdzHuvLSxY+fEum1Gzw==,type:str]
+                      operator: NotIn
+                      values:
+                        - "true"
 sops:
-    lastmodified: "2026-07-09T14:38:39Z"
+    lastmodified: "2026-07-25T08:12:52Z"
-    mac: ENC[AES256_GCM,data:CTnqJl6YPYCt3E0A4QAyt5guqgqiNfFk1xJPAvNM1/kv4HbEVeQxG349AaRSc8bQXcMBIVMf022L4+r0DnTEnyy71vMJwWAJTpnEKBQNaDZNCBraGcznSpOP3t7xibjsQ+pqx6d8xXUm9mDfIJU8Nmqvq+iGEUM1PYsKYfDwhQc=,iv:yBZlb8SujF8nv796elcBMQWbyBg+CQ388UzjVqUUyt4=,tag:P9rW9vl5vkuVFv/RM3gGcA==,type:str]
+    mac: ENC[AES256_GCM,data:i7rQRSLbo97jjEksr0NlQzaW27RmYpHwupe+KSAk5Y9yCSN389EKDWGiOlnCbIDAFuvAeIwZOtZah39g1F0f6EW4UWhxP2URqhFtcc9UeDExEi15lbI0/6cgWsv3mHIZb6WPp8dHrfF0JszEPinb154TB3PpUmhyLkY6gghTYBY=,iv:d/pCDDw8itNu9NCkwTxPbr8w+unmSFrjbvaw6KYlrEA=,tag:x+0Us26vvNcvDyn45ilEQQ==,type:str]
     pgp:
-        - created_at: "2026-07-09T14:38:39Z"
+        - created_at: "2026-07-25T08:12:52Z"
           enc: |-
             -----BEGIN PGP MESSAGE-----
 
-            hQIMA7uy4qQr71wiAQ/+PV8hvPHvkmAg1BLaMWbpkMD0Y5/+uQbnnEfIaJdiE4VG
+            hQIMA7uy4qQr71wiARAAr6IcISfUd787wl46NW/mDOEv+sQCBCvFUSgVUETIVB2W
-            bxtqfBeB/f5kfLDoeVZOCO64tuLtljiZhGjFMA7qESLw0qAvRaDJhXZ9qD1cWSgG
+            lryFntO+jxYjHYTDoevCimyJpFWai6YzkBvSVb50xXObpyHgjLl3IgQx6B3LZxk7
-            DgVhp/4lb+yo37lLUYaqfeRDD5GTXOpKSZOf5WFPr5tsiL9YcfWVVWaf6uSjUl7I
+            mwIgpa/mEyvlP06GzU4NZy3ac0X9bKpQo5IcU0nwfs5ZedCZ0i8FN76PUp/nIfFJ
-            OZkDwi3sDfypuen12I9BM580QRamQzsp6gCynQdE/4PjzND9vAvhEBtftZPLmd2/
+            6AzjxVzcJTDDEPu1ckn1g3sRvLsGsQVzagrLtyEFbYxKxugSE2sswYCMjbnxwMpW
-            5O4XgjaiJXo2PstAD3VD7c7vGbtEUfvn0vmd79yHz1YBtHRYtDfmnWwHJjeOpxP5
+            NTWGal/6eW7W+oR5iJseYnkFz643ZbkI8BLyEuJtCql14kw5W245FZwbffJQpGpU
-            ooFW3uJuwL3N9/IaFTmgGHynjhQAqVdC79PL4XSVkSMoqdW8YDHBgErH9WI4Lsbz
+            NbEuRjbVCF0tDwH9zEPNlBOsrmw0FM2HB6sBCMHA9xexuhP3OdIjWuyNHf6oMwS5
-            vl2Wp9eIeopiyIzGXHsUvv/nP3ayp9pS16fV9xJGVczSi5c5YqFRYGtX7QIyXtqA
+            d2FCOT8xrYBWNOwcoe+xgLam94L2jPcL3SizP7AUEQU1CjXO5q49RvYjPrrgH5zb
-            syQlnzoVr4MeT2A2MNdJEWKtfBRQoQR1Qc7ZX/XwfR4b5/58ci9HwBL2eLu0GzRg
+            PvE8yiB1zDPcTqnBf4O+lIUkdEQ0IxkVWuYKljsaJN35yNgFkCoROaGsZD4IcdD6
-            e8LDnZsGQCozI7h7x332+Wu/9Mbs0wJ1TslQbXt2vRhbve0IvPjNSi54RKI4r8uV
+            EgtYAllK+NXdzd2/iLHf63J4mJsvRHzLsWcdHZwvJxi3RN5Gi18EB1boCEhFM3Bg
-            /iFnRqPp2ow030nvW+LYt5jcqyKK9poC9nVoytUU57Pc3VmyppOJxc8/AfIezo95
+            KkaNDNwZtWtLl5QLkfO6zTEBx1E49hGRlE1nuDpSbqer7cL5n5Ca5TRAm0BqAC5Z
-            LFFWj2CjWr47UspHcqgk84cBP13LmUhK7IeHNBzoWl6LPztuVCczoy7wxn8fGCfS
+            VIjt5CqNcojTVsZLUsYMwq/yI2Lij8/OdM1datuNM8RD33mE+tDB3n8WdNFgN8TS
-            XgFm8RoiId9S5XLhCZ7qV70HhWOZemd0PksBP054xuWvlDMe/HqxvxUmqDbH/vGv
+            UQG0emw3jDSkHQHVWGc2iPZ033NobqICdv3xvjyjdjngDWEtPj4OoLmriqxi0y7Z
-            49o6a/FJyF2ncU01j72KMyn/ZbYevniX+QG+U7vWuhVRH9tRYYQUrgC0QnPqFKc=
+            LhlZa4RWyBc8TWe9Q8paBUh/NzFGC4v1zt8mfUMSjYfyFA==
-            =duXn
+            =ePQy
             -----END PGP MESSAGE-----
           fp: DC6910268E657FF70BA7EC289974494E76938DDC
     encrypted_regex: ^(password|value|ssh-key|api-key|user|username|privateKey|clientSecret|clientId|apiKey|extraArgs.*|.*Secret.*|extraEnvVars|.*SECRET.*|.*secret.*|key|.*Password|.*\.ya?ml)$

kubernetes/system/blocky/values.yaml

@@ -21,6 +21,9 @@ service:
         enabled: true
         type: LoadBalancer
         loadBalancerIP: 10.1.2.148
+        # Cluster (not Local) so any node can forward to pods on any other node.
+        # Required for Cilium L2 announcement failover
+        externalTrafficPolicy: Cluster
         ports:
             dns:
                 port: 53

kubernetes/system/caddy/caddyfile.yaml

@@ -17,7 +17,7 @@ data:
   Caddyfile: |
     vhaudiquet.fr {
       tls /etc/caddy/certs/vhaudiquet-fr/tls.crt /etc/caddy/certs/vhaudiquet-fr/tls.key
-      reverse_proxy 10.1.2.171:80
+      reverse_proxy traefik.traefik.svc.cluster.local.:80
     }
 
     *.vhaudiquet.fr {

kubernetes/system/caddy/values.yaml

@@ -13,7 +13,8 @@ image:
     tag: 2.11.2
 service:
     type: LoadBalancer
-    externalTrafficPolicy: Local
+    # Cluster (not Local) so any node can forward to pods on any other node.
+    externalTrafficPolicy: Cluster
 # Disable ingress - Caddy IS the edge proxy
 ingress:
     enabled: false
@@ -35,25 +36,25 @@ health:
 volumes:
     - name: vhaudiquet-fr-tls
       secret:
-        secretName: ENC[AES256_GCM,data:mdTS+RBfbQiY0c5aezbTURE=,iv:gQRWoRmGujhptSXm16JhX6VSHbB4yMdC3T7tPH1vfJs=,tag:DViHZOwOIPmHJQWTPUO8cw==,type:str]
+        secretName: ENC[AES256_GCM,data:91g6xGYke1FlteHJX17r6w4=,iv:uquY66VZIAPnFse/maIQUcdRVFXUfb/M9tFvNqdgSfE=,tag:P4/D7jpUgEhcOnoxk+QO9A==,type:str]
-        optional: ENC[AES256_GCM,data:apLSVg==,iv:4ZZa85zmTuYb7D/p4mLZO++esFrPQyQ5xHCtT+aghS4=,tag:wWAEf3XWYKvsjwpfpyCb7w==,type:bool]
+        optional: ENC[AES256_GCM,data:KgTxuQ==,iv:KIxXxFGKVA07qMVYZIcBE2BnYTtLl9tFrOZE57xIHXk=,tag:lCyA/TXsxTUZSZDveQSDdg==,type:bool]
     - name: wildcard-vhaudiquet-fr-tls
       secret:
-        secretName: ENC[AES256_GCM,data:Ww5/viWqr5mI51FqHizTRerAZnEyQONqdEg=,iv:VAMLvZ9yJxUCh7Kjf43J+M+ellkm5Xi3ZD5AlO95Rm4=,tag:F6H8LBi9kL9rrXJgVSbAmA==,type:str]
+        secretName: ENC[AES256_GCM,data:cZlwLn0jjz4GGX2z/w4ooeWBbyOAMZ9UHgc=,iv:Yh4wuMcyDekHzGUxCLldnAOaqLPS45YvybwhdZd5kGo=,tag:Z/jti3YF0/wdsrC9PxHVgw==,type:str]
-        optional: ENC[AES256_GCM,data:LQEFAA==,iv:JkudOsHgn8qLEwqWhsuWnXjpCLJ3XrEexXtzjmOD+/Y=,tag:amUBIkXQIZnGQtxAAQ5mcw==,type:bool]
+        optional: ENC[AES256_GCM,data:ZULOMg==,iv:qjb4nvvineVsTEp35JxPcMRDS5Ij4zIsc7PI+d6/M60=,tag:LcOjEqypcQLrBpXBrpRvOw==,type:bool]
     # semery.fr certs from cert-manager (OVH DNS-01)
     - name: semery-fr-tls
       secret:
-        secretName: ENC[AES256_GCM,data:dmLfoVbH9QKPMXPZZw==,iv:l9BKz66sdl9SssNzaz8yWoQ84ewDQprMv3rcHJQZhRA=,tag:NudQDM8eKC5aF6vX2ZQVVg==,type:str]
+        secretName: ENC[AES256_GCM,data:rxGGzrGRZPfzNjFIIQ==,iv:wvgWZIbAoBeJ/7/F4WeBuCHi1I4xHxwLtJZD7/SwV4Y=,tag:n8Zr+hwkD/Vm0LI/Se2TvQ==,type:str]
-        optional: ENC[AES256_GCM,data:Xc2tmw==,iv:xSa/xz/WibDqExKXjKSwLm5w8DOw9aZY44O1TZalfO8=,tag:Mpqi2ipn82+eWv8T2YvSng==,type:bool]
+        optional: ENC[AES256_GCM,data:5M+93g==,iv:KEMghZxEKvH+3vM/mtuVCIvzb2X19quX8Kd+oVJp+y4=,tag:mRqR+Fj86AWfGnvlTzRh2w==,type:bool]
     - name: wildcard-semery-fr-tls
       secret:
-        secretName: ENC[AES256_GCM,data:jwwNceE6ZAD0bdUFceBBG+HQPjmLSQ==,iv:15rvS8NjtKszT8LFS9Z3DLCVMBwQ0GnY+o3DSr+G3jA=,tag:GMZ68i+1XmcvIJrOZxBo1w==,type:str]
+        secretName: ENC[AES256_GCM,data:L6rjVVVtpOne6oog+Ru07ryaS2aVzg==,iv:Qg8wv1LUeaKKGzHpe2HHGR5GzJXYYog9HPJrcPhrMq4=,tag:8Z2yVeFv0mLZlQ9Yn3oRVg==,type:str]
-        optional: ENC[AES256_GCM,data:8p7hYQ==,iv:a3Al2ph6xuJCXkLaFW6K7WSJss51/d9kVrX3Xuh/5e4=,tag:JYUYRsmEN7brNrENFPe/DA==,type:bool]
+        optional: ENC[AES256_GCM,data:Na+sDg==,iv:3r1+mX+APihx5w5sU3/Rnyl4/ZjO8lO3jAcaBuG4B/8=,tag:WKYxuFraq8+659StfI79FA==,type:bool]
     - name: buildpath-win-tls
       secret:
-        secretName: ENC[AES256_GCM,data:kTI0dBzdTAYXuvSZwZC9REg=,iv:xf4pQABQGiuv3IJtiZhNMsLeDPVef2awVW1DmDThOHI=,tag:74IqPca8TuCFuIpJ8RpQ2Q==,type:str]
+        secretName: ENC[AES256_GCM,data:cHgBMV3iWihTRS2h4Gu9CZY=,iv:IttOh8Ra68dwxlkT8hZ4vhg8uEwDvByjFv/pT5ae9jM=,tag:+7xH89jLUMfJB/mJzyuE3Q==,type:str]
-        optional: ENC[AES256_GCM,data:mxxqOA==,iv:p1omSzhLh4yYs303OZCE5DfCNrzjAxEo+TOfezw/oCo=,tag:4ye1QalWmtW98B37DtkcuQ==,type:bool]
+        optional: ENC[AES256_GCM,data:QVSy/Q==,iv:25ldKY94OaF5IMSHtCzNacsBKA3MHTOHQtGVe0fY9g4=,tag:3fl1RCkfEjF86fMsYv8W4g==,type:bool]
     - name: routes
       configMap:
         name: caddy-routes
@@ -92,6 +93,16 @@ config:
 
         import /etc/caddy/routes/Caddyfile
 affinity:
+    # Prefer non-failover nodes in normal operation
+    nodeAffinity:
+        preferredDuringSchedulingIgnoredDuringExecution:
+            - weight: 100
+              preference:
+                matchExpressions:
+                    - key: ENC[AES256_GCM,data:U5edeFnG3Eu+7hhCmkUZXTf1wrU=,iv:hczbYIbQAq1XSCughZdVa9ChQ7nv62elsMtrUzCVV6Y=,tag:/Uh3rsO43oSfyX4fFCEujQ==,type:str]
+                      operator: NotIn
+                      values:
+                        - "true"
     podAntiAffinity:
         preferredDuringSchedulingIgnoredDuringExecution:
             - weight: 100
@@ -100,28 +111,37 @@ affinity:
                     matchLabels:
                         app.kubernetes.io/name: caddy
                 topologyKey: kubernetes.io/hostname
+tolerations:
+    # Tolerate the failover taint so caddy CAN schedule on p330 (small-node-1).
+    - key: ENC[AES256_GCM,data:wGARNk8BSuSn,iv:K1uKmlmQYv0/Y2lXP6JvIi1kbx1VwIb2m3XK1o+OHSI=,tag:PETfq6sO58vkFZmJBWJvDA==,type:str]
+      operator: Equal
+      value: ENC[AES256_GCM,data:2Z/x7/cf6CQ=,iv:yuFeoo3LvfTL2yDpCbOEb994lGao7B/mQnFKrdxEtlk=,tag:xN7t4jD6wZV8e92959BFoQ==,type:str]
+      effect: NoSchedule
+    - key: ENC[AES256_GCM,data:4mk1b0YQAmUOZ+0k7PsilvYd/oXlCjS4HT8QzfyZpUNSjla5vA==,iv:GhJGcDRX25UsKH3B8ZKPLUJOoo4+EqnzlNuBfy8ggeU=,tag:pzY8xT1m++02YD8YRPFpog==,type:str]
+      operator: Exists
+      effect: NoSchedule
 sops:
-    lastmodified: "2026-07-09T21:17:10Z"
+    lastmodified: "2026-07-25T12:06:27Z"
-    mac: ENC[AES256_GCM,data:5APgoOMa+m5xjn+wurqyOtBoohpV271AnW1Ic4Thcg88bXryhr4Xn86Yo7fMbE+N3YqJMCaeaH6hm4iY8IIhCHgO2ybgMq50Bo8AqWxr6MKEpTJDQS5vSSWPq/Y5qr7B3f5/gOyxwIGslF4isVRaQzlcjoKzfvHsZX8SPdFhXT8=,iv:ZuareabZmxIyhh8kZs+eHGeUI7NZhzxER6KJE1/EiWE=,tag:zQuTYSrU76YEf1u2312ruw==,type:str]
+    mac: ENC[AES256_GCM,data:uIOoHpLdiuIOF82w84PsziKyAglkY02OrGU0sigjnhyt+6S9ryy74oiJivzxtniTevsv2PKecN0uXABKzabp+ysY7/uDix/lE0KzoUs6sYA5kVfUj1JPoDQ6bQxoi0y8zKeOqMoqHwIAs+2N2VEriqtRxn+4L3kbTrbe9CFtxIg=,iv:5ZIt9yXKJo5pF9ghi+uVs7zxsdLtQ+pTlc37FKVz2iU=,tag:sTOq1pJh0KQ06soBcmcWRA==,type:str]
     pgp:
-        - created_at: "2026-07-09T21:17:10Z"
+        - created_at: "2026-07-25T12:06:27Z"
           enc: |-
             -----BEGIN PGP MESSAGE-----
 
-            hQIMA7uy4qQr71wiAQ//eR5eIclcLkl8wtERUoVHXL5KGKzEaxyVv9D5gHz51C31
+            hQIMA7uy4qQr71wiARAAvv+DByIWPPLBdqAB39jdZPkrWzVN/7TJ3DU8q380NNMI
-            kmAfrAnjT9G5UeeoJlW8XLkh1IWNhArNBMtoiCNUPAtfVDxszs5IwPWul+s3wJhc
+            2w2I28Ee3iXl4e/Qj66fx8U8NGwdYo/j+mUjIddBjiJB5xzu97k+nT4toDocutGT
-            X7sDq7EnIgaJr2IszIOTJ4PYxZ/uhyMdqmvRTZXMaz43F96cuDLctV4CFaiig4aJ
+            bOI4KZP/J7vh+spb6TMJvH39IGLOYH6N2NagRD6r15kd37ddoRbV8OFTZXvXf524
-            IcNlPIyYyS4tuaGB+ScLtbcDDhNc/8VRWn09SRtpTop+PV4KRX6AqLmhGOc9Jf9l
+            d3+PbqQz9FoulLy+Yt8tWWXD1MpqRbIHL65dAhLF0vZRxUVcnRmg77ZX2Q1yD7Wq
-            Dg3Gqdkaxl3OuqIvcrTmEPBe3sjnjIIdtJeG4SAbMUzj1xoKupXeDBZDnfmS5nku
+            i8c0Pe1dhqg6ig+njshiC7HkWOQoINYzk012hz+Cid5JZAIyxsqxkdoFUWHeWI0h
-            42WR5H4RInzusj5NXP50C7AeLD4qavB9hl5CaxzA6TVG1X7LcpmcqBsVVuFcSjZj
+            u3XUpjjxj+eg0YlnAqrnK16B6+pscZWjEgHJEmrGQo8AGFMdALoubv+8nGiywNZP
-            T80iKklV6Ct0MLlmbxlKFTJm8fSKevhS/hdfrkUNcqc4N6eBb8tx93x+FcDn0A6P
+            tqM9oT1ePa0pIJNzhNgFWGkwwuyH4jPXrRji9fpLASAihsM1WQIh3smd373C0Z/W
-            c1zPSKj1vs326nJSHw87wvooQWn/f9j+Om1KLOR/MiUWmoEAbgkIHi3+k2wpzFb/
+            sQYFTS9B6FUCCdDPA/St0Rd/RHFW+eJPC6Jh4728Pqa67FjnuIVB5VJ33Dgpl8OE
-            faKpshye0/XA/CZHFePv/U7sw8nKhfPhJCuO/IRrHKYzx0u/aqgOtPCURh5hS/fU
+            kEyY3WvtoYOaw+6VqWetg79sCruibmlenDWSrT8VQ8ksk/YIj+iicPnI/BY5S2Wz
-            8dPQkbywgN67IRRKVz70cxn367cP1vn7L1AmzJfBd56NnViXXlS/DuitpnyTQpnP
+            j2s1Lpvvhr/gZvF7z+vfhX1Fn28xIg5/gkcoYSKgFhqxG69IlPbZsr/D2oWyeygd
-            tLG6qelq3Ul2A1J3BB1pfHR27IpKzd1UPD7HMhvvgSA7ErLq/VhF/kT1dx6s1fXS
+            tIp/D1yHGFJ1Cb56qr400t9xF5BD605rf/Aczg8mdvI41jGvmECQrSKHsxI25xLS
-            XgFZ5XcyJQRc4Uxh2eFeEPOdWMAFRps6scfBsQuWR8m3cQvLatzcku1S/YDp56Kn
+            XgEVhCAm6HDAKEhjnm1wAZG8DJRhcrSLqLq7SPKJaTPJpjeQwXOb6OIs9ugQx9ON
-            oUZmfvdQGpUuxsumVYCqk1+mQn02lkA1p2Y4ztrpLDCbhZXZ/fIT6Y1uYX+c0ew=
+            w96InR1VMsEUlh1EWBXUqBgQgyDIIEAa+wHBnY7U06fkazN6lcLkAp17yTitTEk=
-            =0kw6
+            =gULt
             -----END PGP MESSAGE-----
           fp: DC6910268E657FF70BA7EC289974494E76938DDC
     encrypted_regex: ^(password|value|ssh-key|api-key|user|username|privateKey|clientSecret|clientId|apiKey|extraArgs.*|.*Secret.*|extraEnvVars|.*SECRET.*|.*secret.*|key|.*Password|.*\.ya?ml)$

kubernetes/system/cert-manager/values.yaml

@@ -5,7 +5,7 @@ crds:
     keep: true
 # Enable DNS01 challenge providers
 extraArgs:
-    - ENC[AES256_GCM,data:yzuTi9Hu7Dx95MQN+H/6gul381m64KYv5ZsHwg92BE/aUZyJzYArniIC+Nio+SygUXXb,iv:eBizFSW1T2/VvN3k8VRsEIpllHs5MA6Nr+jh9fCzZCs=,tag:sz9WBU687eTgnDXWoE02Nw==,type:str]
+    - ENC[AES256_GCM,data:RcKzGzDSFXAw32vBKyvMTzMx5u0AmXKpXEY9/YRxUI0SNVy68/uUSpFhdmcB0RvmzG7y,iv:AraIcgoVYO1IVAmN3dd6pXr3v2BXaizbK8HefwsDRqE=,tag:QlOBuyxbddyk++nYLhMl6Q==,type:str]
 # Resource settings
 resources:
     requests:
@@ -14,6 +14,15 @@ resources:
     limits:
         cpu: 100m
         memory: 128Mi
+# Tolerate failover nodes
+tolerations:
+    - key: ENC[AES256_GCM,data:WXIPFO1nBe+6,iv:t8RP3KL6niOVe84Qqu9+JKp2qNR87qnMzJOh7pUsudg=,tag:uT7gWSdOz86PPi2NZIxZeA==,type:str]
+      operator: Equal
+      value: ENC[AES256_GCM,data:N1IHrGvrobA=,iv:QiKeIiSBFThw05b5gkU41koci4CDZ0ZxdLX1r1DD7Dc=,tag:QGWcNIYqUdEy0f0bEYC91Q==,type:str]
+      effect: NoSchedule
+    - key: ENC[AES256_GCM,data:QMtUDj9EDXWbsUBLlQJ5yFEsQ3fksIRUqIk7tKyts1aHqNQ56A==,iv:GVEl1AcWNLAqdXqyJTMoYBPRqqvhSHGBN7fuMLWtjoY=,tag:bmCK8yCiVCXci2JJFvw/GA==,type:str]
+      operator: Exists
+      effect: NoSchedule
 webhook:
     resources:
         requests:
@@ -22,6 +31,14 @@ webhook:
         limits:
             cpu: 100m
             memory: 128Mi
+    tolerations:
+        - key: ENC[AES256_GCM,data:QZjbo0u9rqht,iv:H3mZnC4a2abmOiW0RhoFAiAvVfaRFpGU2+XKYy8uBUQ=,tag:xwp+rkiYrSo1q+CsLdup7Q==,type:str]
+          operator: Equal
+          value: ENC[AES256_GCM,data:UsVMCFtyl0c=,iv:ImpKIQtC1jX2NLNTSDhq4a9nq3I2HlpSr7QLUOw5gd0=,tag:xM/vf2MhasBNWbLqxkUrQg==,type:str]
+          effect: NoSchedule
+        - key: ENC[AES256_GCM,data:31pC3NOA0AH2d0V/N10xX1U8E1OD4YL/27+xFLxElfYb1xWSDA==,iv:aaaci0Eu3Mb9og3Za0beB9js3wSjcNE7BOS5iAko3aU=,tag:h40mrm+gPTf9jCP7ax1ihQ==,type:str]
+          operator: Exists
+          effect: NoSchedule
 cainjector:
     resources:
         requests:
@@ -30,6 +47,14 @@ cainjector:
         limits:
             cpu: 100m
             memory: 128Mi
+    tolerations:
+        - key: ENC[AES256_GCM,data:rqZTE8j9xDAy,iv:EvztWg01VbvJhdhkrVz7M0mA0EUhVW2Ty/de0D3pgmw=,tag:rGUZky1bUhewhKDp3Vo8Eg==,type:str]
+          operator: Equal
+          value: ENC[AES256_GCM,data:fPt+n+ftbGI=,iv:xHy9IUIBgkHDQcVZ4MyA3tXPvMAy14gHhkxe2AYthR0=,tag:szam+dyODZSjfX6oxsFdSg==,type:str]
+          effect: NoSchedule
+        - key: ENC[AES256_GCM,data:As98cp+6SvVwxRtdxx7TT+HGzejgWiacf0TYm2JU9kPqEA7ljA==,iv:dd+TS9T57NrpgyZjYgPi/aCiInfnVxh3VbO7gddkw8k=,tag:KQMA0Kn0YgE/HpAJk83pTw==,type:str]
+          operator: Exists
+          effect: NoSchedule
 startupapicheck:
     resources:
         requests:
@@ -39,27 +64,27 @@ startupapicheck:
             cpu: 100m
             memory: 128Mi
 sops:
-    lastmodified: "2026-06-16T08:15:29Z"
+    lastmodified: "2026-07-25T08:12:52Z"
-    mac: ENC[AES256_GCM,data:ZB7igt7ciH6X0DmvDi2gzg1eA8EYXqq/VRBSbaLT9x2SUi+9ax9w0V+fcTwOTWlwCvHOtSAZ2RYgX/wKFmbnKgrwoSpskGFsRlY947oLkRTk4HbTRP5HbciAQsRw3AVB4pgkR7maVZ9n15gzNHTTkd9x1akAeGHJg9lzg9+N6rE=,iv:jqyi/ZGLjKFBe9XSj2WBBOUbn9xvV9Wf4wrYxSs4t/M=,tag:2u84qHa69IsYFtSwJ6yFqQ==,type:str]
+    mac: ENC[AES256_GCM,data:yFXZxbhJi5PPVXl4j73FfOO5hurvD7ZXgWr5eZS8poE7IEg6IgZKBjwq1ITMUP5DyS6K2onR/mgZVzgj5X5I7hEwj13DAPW7wWthZql+De7oGxe5Nv74fDFEwCoqh2sTbdrQS2KyPY+rrb0ZMZD5qpdZzyw1AVGzVKeJJj9DdBY=,iv:SW7sb6mmJulekZn1Eh2pfYml1PwJ5v6z/PBdV3TL29A=,tag:5HzpRnH2G8J//J2fZhrxrQ==,type:str]
     pgp:
-        - created_at: "2026-06-16T08:15:28Z"
+        - created_at: "2026-07-25T08:12:52Z"
           enc: |-
             -----BEGIN PGP MESSAGE-----
 
-            hQIMA7uy4qQr71wiARAAiNHNEJjBOd6AK/moP1Xw7nvOidkRtIqw+QJRmfhPDG7B
+            hQIMA7uy4qQr71wiARAArV5QxST5xu2jG+X9Fo6F4o3uxMS/Kw56ey1MMTfgkTlM
-            CQy3y2armtFYzaWhDcMfiy1SkcD91kadbfxAifb5/Oit48VTv03bXZDvyrEdRwcI
+            HdV1n6lityMGXB278GiYhmZYRdZl0kAT+Dj9Y0RdaiI8BszocD7IbBKmE1tVseCr
-            n81rt3GQuyNDCj9aSUVi1nT5GHETrvdnyJp3u1dzfc/0i7GK+FiCbGAGzSnGdN77
+            Xa4y7nWE58xz2Ca26KVf/I64QPHsKKus4wECGOgZrh7Hzb+KdeTJBpuMMIgnJe3l
-            +/caT5Jib+hIVskiOjhE85K3rTrh+aJMFKC6NG7Vw8GXh7N4r3EcazcX9KuJXxsn
+            1C6VdfTgNlYPW+zAmX3i39ZwYO7wcz5UZ1aXozAtoOjWyjdcArEdoCH9iD2GXTOe
-            NRT7XyYA8y15PlZQFkhRs1QCVvqH31WDwCEt0cE4r6Qn2aGy+v8ygJkum2sYPH7g
+            0EsvgdUD1U2n4ffy+Z8mga6V8URkJQ86XTuFD6PhFF1bhBwa36MCUXAdlD9ONB6d
-            WZfC548mZrMrI9UNo10UfPCqGew/XpEla1/bgVByMHMdM9n961KRzCPtR349xXI4
+            jaKSAPAeia4C+0vCXTv9OYEHizSe11iwsUyPq+bmMrYCUslaiOCC+QfGtJEcCJj/
-            1APX4WClQjbL/cXR2zMs19Y+GDjL7XF6rdSRDpEgTzQXb4f1ctswutcYxftvKI3c
+            ZeAjXMoyi5Fnmwj+y5ED+O1DKp/sDyUdvix83L1UWVfTrgNVfXFCGzHJUkcZhfUk
-            EdfCGiRpVJx/wIuh82mL3SWqdi05Lekvt1zdLcG8Tx2+nF+52dWoyxNy1YMVyk0m
+            Ye9KcbH+7+mj3p/lAgbQjBjVemnW6GFzGJYcL/rWwRBxw2ElLxKPnR9NQK9VY5w/
-            RXZFMZLHKumyDooairf1P11DOXcsa8FjZrqjvn6QI2LfmEGm3PQQQUXT8Wp/10e2
+            wCmWzG4Cjr2gfLlR05A72/i68fRM9eAy4CFL5y4xTN1eYe+MCz2ZedZDd3uYTKUC
-            78Hr1rTGqQEzvSJu0FfDuundPws97ftPiuGcUD5hUj3rS6iGxdEp4trAPi0DCCd3
+            TMPGXfjhy64W/cV8MWa/aYCyNguVlKbmUl7kZwb+8DAxrSyPqQal11DJtcsv3zFO
-            g7m6fNX41O8n39EU0Kp2G9kfICMghGEVL5czzRA4EsLH35K8XqP0ig6ay5kVXWrS
+            BUf8GQFn4jXXYqd1LojLM7bUjAkXWZUIUbg1+jWfPGsgsTGn4R0tshtY9JOfQpfS
-            XgFic5jk5GdxzUL7TFjtr9AFFaWhIZkyyXwvEod3Ur3gB64Pi/ktet6OZSWU/7Wn
+            XAGL6+0+DQf8wY7yzbds4DSDMNTlOH7iWoX77okayY3L5Jqc3jB0glMJbRwdmgnj
-            eRKn9yEm+W5Xzn9eiN6TYDsCWQBY5nP80YVuj53AMSu3KbR7UGy+AbJF3T+V6lo=
+            d5JzAoqDhi1F+MQdTYGu7RQpawKr6ndWqwkT5ef5BQ86TQq3/+jZ5uuC3qnE
-            =GscV
+            =uS2O
             -----END PGP MESSAGE-----
           fp: DC6910268E657FF70BA7EC289974494E76938DDC
     encrypted_regex: ^(password|value|ssh-key|api-key|user|username|privateKey|clientSecret|clientId|apiKey|extraArgs.*|.*Secret.*|extraEnvVars|.*SECRET.*|.*secret.*|key|.*Password|.*\.ya?ml)$

kubernetes/system/cilium/kustomization.yaml

@@ -3,3 +3,4 @@ kind: Kustomization
 
 resources:
   - pool.yaml
+  - l2-announcement.yaml

kubernetes/system/cilium/l2-announcement.yaml

@@ -0,0 +1,22 @@
+# Cilium L2 Announcement Policy
+#
+# This makes LoadBalancer VIPs (from the local-pool in pool.yaml) truly virtual:
+# Cilium answers ARP/NDP for the VIP from whichever node currently has healthy
+# service endpoints. Without this policy, L2 announcements are disabled even
+# though enable-l2-announcements=true in the Cilium config — the CRD policy is
+# the gate that controls which services get announced and on which interfaces.
+apiVersion: cilium.io/v2alpha1
+kind: CiliumL2AnnouncementPolicy
+metadata:
+  name: default-l2-announcement
+  namespace: kube-system
+spec:
+  # Announce IPs for LoadBalancer services.
+  loadBalancerIPs: true
+  # Announce externalIPs assigned to services.
+  externalIPs: true
+  # Which interfaces to announce on. empty = all interfaces.
+  interfaces: []
+  # Select which services to announce. empty selector = all services with
+  # type=LoadBalancer.
+  serviceSelector: {}

kubernetes/system/coredns/values.yaml

@@ -80,28 +80,37 @@ extraVolumes:
     - name: zone-config
       configMap:
         name: coredns-lan-zone
+# Tolerate failover nodes
+tolerations:
+    - key: ENC[AES256_GCM,data:jLsxKqX8HEo8NbwtjeGf+8TdVzTgpba8myb09z1FOIoBsplZ9g==,iv:c+rAeZRAL8ZmQwpHDF5xv/l0ZqbWsnsdDlpFDSDfZrA=,tag:u9TfeXzHJ6W0ifaAEoPYFw==,type:str]
+      operator: Exists
+      effect: NoSchedule
+    - key: ENC[AES256_GCM,data:hPb/xXAmQrH6,iv:xo935JWcl3a3LyQ8YNIgNrjQkdqlP5aN4ZnjTnp4RM4=,tag:/rkpqOI/pHqXP47OGULnOg==,type:str]
+      operator: Equal
+      value: ENC[AES256_GCM,data:nblSYUAjuMk=,iv:yu/IToMo3DC9kfI5DcBPvreJE3dkxdFA5x2fy6dUpjE=,tag:NGVTMBXa/STsOCD+g00yaA==,type:str]
+      effect: NoSchedule
 sops:
-    lastmodified: "2026-05-02T16:59:44Z"
+    lastmodified: "2026-07-25T08:12:52Z"
-    mac: ENC[AES256_GCM,data:H4uRid1Fqx4JzsF43TSGa7QcGjpXLAHiM0N3Kf4z7ab4eMlTy1+RXMV7xVT9BinjZzH6P+ENxo0yVOsdt0Yu467KJhGznNWlb2MC2TElPxZ9/yItJ+hdVGHGWbVGFWUL5NOUQ9fY2NPGw0CGr8qyftLr5Qkx0LO/VUgKWkq6RWM=,iv:9+V/sCBhfWAsIvr4DsWQgkeqQZQyT4Ti3Y+qCEZqU5c=,tag:JCRONb54BpXQzYhhPs7VGA==,type:str]
+    mac: ENC[AES256_GCM,data:fyuCFq0hIVP0P8e8vqF/OKjBway2EzDo8HJhLpTQqhIOhOP6PsMszovi7BzG4FuqT0Hbcqo7YrjcVKY2WPGIlxoMVBuuXcUSvNB8Tdu6jv1jgq9KOoB0HNbnqX/l3TzN9ckC6Hr+QrwzLIpTYNefDlNovMCz+zBP6ALfMwADCYM=,iv:EiXUFcWr9iL1qdfzcTiL6LYLLrsShzwNNhE0eyLVYrU=,tag:JyN9WrQWSE/3wFksooRRFA==,type:str]
     pgp:
-        - created_at: "2026-05-02T16:59:43Z"
+        - created_at: "2026-07-25T08:12:52Z"
           enc: |-
             -----BEGIN PGP MESSAGE-----
 
-            hQIMA7uy4qQr71wiAQ/9HzeTVqelbvPtluYa5xGvoYNeEEXg43CwrwZ1/z5yFWvx
+            hQIMA7uy4qQr71wiAQ//efLh++Dcra+LFtcgz/BRXdLyE+RhWXpQjABE2PzgOJpQ
-            DoOCeyro5wFsNC6td7n2HVhtK0ULkfrMHH8OC+7L3bXbnlEnQzITmDggAUvfegCv
+            xubJJDmIV7BtUFyJS2lEjuZ2Be8uYxs2WtWILDktz2IwNQ882NVCWNCQTPuzEpoK
-            b/7ohPkOdLvi6qXbr8bgqCZYFnPq+gUs3UOPh5Tl6wgzRSFXw2Hsb4YmQkvZJUNb
+            DWqwi7/L6iQldx1aeE/VUocyIxYsqS8ZUHqOM8Z95xlQ0/fmhB7reTCXvuEOiRy6
-            PhPpLIUe/ECE4hmEjO5v9o3X0o7qZ3bahf9mZZlnJnvXT7R/DM8eeWTis/q0WSHE
+            Aapvb2yW9nm7PwBVhJiWQQLGVGUkgPE7HIK1ar3ODl+b4T0FCYqjUzeB78mG2D5i
-            XnclhOX4GlMwXxa65sRrShuPcsV3qqX3VWOSWJFBhGx/FDtZTkhlHGQ9YhF2TzbB
+            RI0SRt56w+g9XVWk1tijKL9wx11U9KOBbmbdBuE9fkHuenns5omAzKIK0iwdHQkV
-            xxCrn87mH2W13NH6jQOQYPh1JTTJbgZZMZXgyPNmPDSYZE1kxTdrz4l4mcmCDND0
+            cl3rr6m79OAqJglDYLFf8LOgHxuVcnlVxOYgmiHXjG4hlJlt7rGE96+UlRHCemhD
-            hY3T8iR8ap2b3HhSNCqC1C0QN/bK217hTs8cJHWRRfa6jfh12imwk2XhJkB3zZxV
+            bgyc7kDR3h4XMagI0OcLzjyvwsymCtZFiE5PLRVLkSCQuyrp5B3QL7Z5hbRZFiKF
-            O1oSb6eiP0ba0CgXu31shmfXuTAeVbTm6E50heYorjQKR5djjnOVwQUdmis1Awae
+            UlOziOw8sPOcvjOOJpPYF2IqWAGLSYwo5S0nI0O0bwOiM+T+ehZ5q8p47NzFp04q
-            AQTiWtBBbOgfX5WA5b6wInFr0WEsshG+YuqfB7FhJpo2SHyeFhgk47ssHWSeBpPv
+            l6JpHtWdjsxVS38nDb9AWqvoHyf8EtFbJYmKeowmbN1/toPQ1q80at/NtUZIzn/3
-            wa4OAGaMkdGoePQhApZFrBCZHslEhPE+XQlDdyOtXCmxBOcLwe59ikWLV75j0DzS
+            +VhxDP9u38u+lAAuhaBdIolitOi9CWwHqrnRXFa3V4fpQfJqXUc9You5bw/+EABV
-            NRUNOBYQ8Q1Y6Su/sJWW7TykQkmDirU+oIYxAngZyIyJSWvARPd6fJJvkqqg013S
+            jTjMpcN7cPWE7SC9R1ZZjvxrMKK0qxdKO5cvXtP/hBd4zOnBqhsQBzr/QB3DycPS
-            XgH1+LQJWNEJzIaLKCWbkZXnMstsOYrs4ynV4f/QZKU+Md5CgVbjy9KIC/trfNhj
+            XAExosGRaKVumsUVYqEnh7ACVtMHGDLvupj1mqSa8oxRJLVDXNFLpEllFqLpoxjD
-            1t9kkyVVOEO7UmRhMyl8pK2gQDiOBrkhUJ5tSNFEfxM1llZ4GZRV+SUuMC3UzVA=
+            qbCMib+JeL9/Aq83Apk3HQ48WX+4hvtkKIJ7gzeIPC/IvIVJHk0shd49kxBN
-            =l7Wo
+            =Xsas
             -----END PGP MESSAGE-----
           fp: DC6910268E657FF70BA7EC289974494E76938DDC
     encrypted_regex: ^(password|value|ssh-key|api-key|user|username|privateKey|clientSecret|clientId|apiKey|extraArgs.*|.*Secret.*|extraEnvVars|.*SECRET.*|.*secret.*|key|.*Password|.*\.ya?ml)$

kubernetes/system/coredns/zone-configmap.yaml

@@ -30,6 +30,9 @@ data:
     docker-r740     IN  A   10.1.2.212
     truenas         IN  A   10.1.2.139
 
+    ; P330
+    p330            IN  A   10.1.2.132
+
     ; PVE
     pve             IN  A   10.1.2.10
     docker-homeprod IN  A   10.1.2.12

kubernetes/system/longhorn/kustomization.yaml

@@ -5,6 +5,7 @@ resources:
   - namespace.yaml
   - repository.yaml
   - release.yaml
+  - recurring-jobs.yaml
 secretGenerator:
   - name: longhorn-values
     files:

kubernetes/system/longhorn/recurring-jobs.yaml

@@ -0,0 +1,34 @@
+# Recurring backup and snapshot schedules for Longhorn.
+apiVersion: longhorn.io/v1beta2
+kind: RecurringJob
+metadata:
+  name: daily-snapshot
+  namespace: longhorn-system
+spec:
+  # Fast local recovery from accidental deletes.
+  name: daily-snapshot
+  task: snapshot
+  cron: "0 2 * * *"
+  retain: 7
+  concurrency: 2
+  labels:
+    type: snapshot
+  groups:
+    - default
+---
+apiVersion: longhorn.io/v1beta2
+kind: RecurringJob
+metadata:
+  name: daily-backup
+  namespace: longhorn-system
+spec:
+  # Daily NFS backup for DR if a node is lost.
+  name: daily-backup
+  task: backup
+  cron: "0 3 * * *"
+  retain: 7
+  concurrency: 2
+  labels:
+    type: backup
+  groups:
+    - default

kubernetes/system/longhorn/values.yaml

@@ -1,10 +1,7 @@
 # Longhorn Helm Values
-# Single-node R740 cluster, backup target on TrueNAS NFS
 defaultSettings:
-    # Single node — only 1 replica needed
+    # 2 replicas so volumes survive single-node loss
-    defaultReplicaCount: 1
+    defaultReplicaCount: 2
-    # NFS backup target on TrueNAS
-    backupTarget: nfs://truenas.lan:/mnt/fast_app_data/longhorn-backup
     # Allow full disk usage on single-node
     storageOverProvisioningPercentage: 100
     storageMinimalAvailablePercentage: 10
@@ -12,70 +9,95 @@ defaultSettings:
     autoSalvage: true
     # Disable strict flushing for better single-node performance
     strictSyncReplicas: false
+    # Toleration for system-managed Longhorn pods (instance-manager, engine,
+    # replica, CSI plugin). These pods are created dynamically by
+    # longhorn-manager — NOT from Helm templates — so their tolerations must be
+    # set here via the `taintToleration` setting (semicolon-separated `kubectl
+    # taint` syntax), not via per-component `tolerations` blocks.
+    #
+    # The P330 node is tainted `dedicated=failover:NoSchedule` (see
+    # infra/p330/kube/variables.tf). Tolerating it lets Longhorn schedule
+    # engine/replica pods there so a volume can be failed over to the P330.
+    # The control-plane taint is also tolerated defensively, although with
+    # allowSchedulingOnControlPlanes=true on the R740 it is not actually applied.
+    taintToleration: dedicated=failover:NoSchedule;node-role.kubernetes.io/control-plane:NoSchedule
 persistence:
     defaultClass: true
-    defaultClassReplicaCount: 1
+    defaultClassReplicaCount: 2
     # Retain volumes on PVC delete — safer during migration
     reclaimPolicy: Retain
     # Mount options for ext4
     fstype: ext4
-longhorn-ui:
+# longhorn-manager is a DaemonSet — it must run on every storage node,
-    replicas: 1
+# including the P330 failover node. The control-plane toleration is defensive
-longhorn-manager:
+# (allowSchedulingOnControlPlanes=true means it is not actually applied on the
+# R740, but keeps things correct if that ever changes).
+longhornManager:
     tolerations:
-        - key: ENC[AES256_GCM,data:HXcRGwnZkabc9hqnCNMYN05JYriiMQx7A7qTWXFvSvrHrBNP/A==,iv:obaOkaDCXPf1krWzs2XX3K2tKNQmDIAjCI7axto4q4o=,tag:2PQeZIy6E4UyEyDGjqqmpg==,type:str]
+        - key: ENC[AES256_GCM,data:DDNTXJiwP2LQiOmqgf/L6tLIZiFgqh42dLDXj8xomGv7rlUP9A==,iv:eu6vnl3EiLu1DWWDWlxkQNcN6/uEhSkpufw7L9Ikr14=,tag:DJOMo8LDGmzkqvIEkfCTHg==,type:str]
           operator: Exists
           effect: NoSchedule
-longhorn-driver-deployer:
+        # Tolerate the failover taint so longhorn-manager runs on the P330
+        - key: ENC[AES256_GCM,data:rbWq0sruC6zC,iv:Eq5izQ4LXZ90Bhf1BJ7V9/Xo7HVMULeDHZ6NgItZfGQ=,tag:ixJt7YD4cBy+SrG2WVn0ng==,type:str]
+          value: ENC[AES256_GCM,data:wxGNkS97rg8=,iv:bBGGqcwibhXE0zkH+ZJEVgr20ogu7JwQxxDXqJcMdsQ=,tag:VLBQh8vRl1SiYB9S/vwLeg==,type:str]
+          operator: Equal
+          effect: NoSchedule
+# longhorn-driver (CSI provisioner/attacher deployer) — tolerate the same
+# taints so CSI operations work from the failover node.
+longhornDriver:
     tolerations:
-        - key: ENC[AES256_GCM,data:NOHqzVl7VyrhnoS7UzdYuFDRReXnfb7gZKyEHwJkYJtVUbaZaA==,iv:P9v9AV2mKt6uPiKUU/2RW8Ncpefn9fbz1bmcVgUg2zQ=,tag:Qfy2EpKVyKWGlK/OgkY4Hw==,type:str]
+        - key: ENC[AES256_GCM,data:1EBSG4hL4heCnLFJYzTNaQWtFU3v6xwTlG+Y9/H4p5d0TZdFog==,iv:9ILcNXXzxzG5CYA4Ny7bKHBI581B1qJC3OBUObScxbQ=,tag:4B4JaFO0tfyVA6w7ypCY6g==,type:str]
           operator: Exists
           effect: NoSchedule
+        - key: ENC[AES256_GCM,data:ZSC5qg+ChOkI,iv:rIq8j7L7F641kWDzcSV9lrYxhHggT1FTZ8WF53vxvV0=,tag:Hj4RlF3usdIQi0G6Ecmdbg==,type:str]
+          value: ENC[AES256_GCM,data:YfiuCv2EKec=,iv:+6i5CT1ToCKy1gFtVVIFv5FLQaJaKHx4vQk8b0IBy3U=,tag:zHgzNuijbwI1QbvuLnZNvw==,type:str]
+          operator: Equal
+          effect: NoSchedule
+# longhorn-ui — tolerate the same taints so the UI is reachable even if it
+# lands on the failover node.
+longhornUI:
+    replicas: 1
+    tolerations:
+        - key: ENC[AES256_GCM,data:n4UdwssGYhHQgDTpCpu+WgIPsdNc8uAaZtnURZt0/C3VfeLR4w==,iv:mqy0GqHN0KxtkETS7psu61Mr1oGmtPDhg2vQhVDas6c=,tag:uUd+NY90XK3i0aigc5ndCg==,type:str]
+          operator: Exists
+          effect: NoSchedule
+        - key: ENC[AES256_GCM,data:isSkUNqax+AU,iv:jWgZyZL9VMmzYqoLCUkIiUuy1iBeOm2RE5CjZGCcXxQ=,tag:714V0YGb7jqebC4IRX/phw==,type:str]
+          value: ENC[AES256_GCM,data:Qbtu1wgV1oI=,iv:RihD43IA1+C3m9OkXACGgd5lY9lDWCyzIcw/MCoSMvg=,tag:WzQtC4rT+uovS6s36DJWUw==,type:str]
+          operator: Equal
+          effect: NoSchedule
 # Disable ingress for now
 ingress:
     enabled: false
-# Recurring backup and snapshot schedules
+# Default backup store (NFS target on TrueNAS).
-# - Daily snapshots: fast local recovery from accidental deletes
+defaultBackupStore:
-# - Daily NFS backups: DR if node is lost
+    # NFS endpoint exposing the TrueNAS dataset as a backup target.
-recurringJobs:
+    backupTarget: nfs://truenas.lan:/mnt/fast_app_data/longhorn-backup
-    enable: true
+    # No credentials required for an anonymously-readable NFS export.
-    jobs:
+    backupTargetCredentialSecret: ""
-        - name: daily-snapshot
+    # Poll the backup target every 5 minutes for new/removed backups.
-          task: snapshot
+    pollInterval: 300
-          cron: 0 2 * * *
-          retain: 7
-          concurrency: 2
-          labels:
-            type: snapshot
-        - name: daily-backup
-          task: backup
-          cron: 0 3 * * *
-          retain: 7
-          concurrency: 2
-          labels:
-            type: backup
 sops:
-    lastmodified: "2026-07-18T14:22:17Z"
+    lastmodified: "2026-07-26T08:44:08Z"
-    mac: ENC[AES256_GCM,data:cFl45hiUDL47h0V61Sk9L1ZOb3NsXwePA2IROgOCYa7xnxe4iPM+wGS+rDQWhrmMY8HkDTEyKK1s4ylXXqhMSeyvMTPzib58OC+N10UbcmztPDSQEYzfGrDQ2o6MCcSWTapW1pHersPjMEZjc+j629TYriSIjZZM6k9ufUnMMos=,iv:fU/F4ZOjuT0TFIagLUejrxXV6x9Q9P5mrIxy8cQBR7o=,tag:VfSH7cDZD754ma6SaT9/sA==,type:str]
+    mac: ENC[AES256_GCM,data:4shHlVWcVol/Z8b+5+tz6shT8x6gCZz2xhYv1VT7dHlgcYWWgq2V5mbGUtsVmWFzkGPC5rYvU3dUEiEkl0zb6Q1rP/fD6xQmlyIgH/ioHrfCvjorY5Rv8t4OFH/TyBK9hMwC+iFdAUJOdE9vYQoDmUMTGz6XwWiDY3PxRv1X9D4=,iv:DDDiStImg5qH2QP9lcSyW5+b+1P1mk6F0TWkt9p2syo=,tag:zJnZ455gsXPkE4Lfup2Rcg==,type:str]
     pgp:
-        - created_at: "2026-07-18T14:22:17Z"
+        - created_at: "2026-07-26T08:44:08Z"
           enc: |-
             -----BEGIN PGP MESSAGE-----
 
-            hQIMA7uy4qQr71wiARAAmfVmgR9JYZ7qkBty+kYxe/4c0pHq2kab8ORYKfCrADeb
+            hQIMA7uy4qQr71wiAQ/9FsBtkggrKwHfh5FFRUFkyqCU596d/R7ZEvsMP7zzCH9H
-            oVJvIt/WO6PbE4RKJLOQzQnlRvdqAKukKYhOP6yiBgZpvxY6W6knJWNv+zad0uzH
+            VAAYJxl8D6/KY5EavpeZH4s2ZnlzZjlb/qS2CKpSzXuN7ADPnK10cBxxAOQLp6KF
-            wppTMDHFo3RW/DFUmeqGjBLGi1V4FTlnURGdYzMMuVWDaBp6B4F6ZmtV05Td2Ije
+            pqdkv6x5MyWbfv87IzM8cI0D/jlG2/iZPfBZy/YU4Rg8cefIl82vNgNADW2O68bK
-            JHrBw+LuR01wooaL65rNkiAZbaIu+fFWMlSKME3kie3U3/mWRvJyPUmB07JjO+6Z
+            w1VS75AYKmUfXZ2vqz+wX46YcjdXkrvRwzM9lzfIm2Q2dKMqBKSYzxMICZ/hWGRj
-            Ew2JTB94fhgnVfhP0gK3DYZLu4ZlqKC3bndRNoODFr+eut5OEzABh6b5eJRWjaGw
+            t33KbZLKC1Kg0NmYOY4GCt8hi6Ck4TjKvzxktsEpXYFDhWzXe9nr+axtrfypLrAn
-            8uJ0ymml1swba3DbhpxJeWPRUlnyVIJK2pWzKXm4R4jFLxk+7gpA/2EwhbGvdA8Z
+            fcLCQV1FSovMK+P4JxB/dd0u+F3RR500F3tBJPQ1xs1Rc1f8Cd2Ynr/3LWtTstkq
-            gNEH35ng5Ej6t2XWXTOBZ0Fn7TrX2ED0WjL7M86pTargAjVZiEfTK4zJwUh+nmo1
+            JyOfLFAj4/LG8o7ub/FSvLZZI052iRLZqlIIe0cFTqCz31yVEJOnLJWmME36hDEm
-            HQU6EXBEtF4ZfAlwb1xuXLCBiJlLciyjjD6WcvWP3ql6JW6JzULIMPBLNPo8J6fp
+            rfd3X9IvixfwWVTa3K0TxZta7u0Q0DsqHIt3eLdeTJC3wY4J+A/vUvPFv137YEqp
-            YvqKNzjCG1kf/DmsF7tfC0LpNlbn+2/QlZ/PYtWXK5fUtD1JIdjBjxQ1qeNNPplA
+            O+/ALTbwK7ViSQIupjr4kEcklBtiu9myNi3u8JNg67/XHwahCh0tfGk4jCevve5S
-            tfOEkCMhzj6zah/xDvfa8dearZPJAfJexpn8UFVg1BsLyfkoFWRfssj7oDCGCr9x
+            dDk4Ud7469LuDwYSZw0RdwuU89jQivvsp8/mZpQIoSVzl//iEQ2EOAff1gMfiPoj
-            yj2jIn3ELq09vvGjyu/n1+XGDwPTSK9SXGvVrXEaj1N/e9pjDVzeXgvh55mkC7nS
+            1MrwGtrrMAo+2Uoxacd7LfFn/l1zRNiGw1rX21NzswmzIU84US4Grv9LGealQ8TS
-            XgF/f+adllB8hz3k0BrQMCSGBHyZo4QDH8QTO3E3jN0t0qzEif9bQ82mOCvh4XJj
+            XgH+datMRHIAPKcYY0u/B/PfyoKArlPUzQsIQOm02+s1zuvK/dArKxgObKTAr/u4
-            w1cvZEUL3+b9/0tVaBpHNCebRagRAaIay0Y1Fs3GqsdJ1GkI93pumLbL10OYs6M=
+            uNuVH3ObUVMb6VX4iwlxfS+VKpjotTbgY2TJfV46kGC3JJ0w1LdjEyJbpYlIz/k=
-            =5Sp8
+            =HWEp
             -----END PGP MESSAGE-----
           fp: DC6910268E657FF70BA7EC289974494E76938DDC
     encrypted_regex: ^(password|value|ssh-key|api-key|user|username|privateKey|clientSecret|clientId|apiKey|extraArgs.*|.*Secret.*|extraEnvVars|.*SECRET.*|.*secret.*|key|.*Password|.*\.ya?ml)$

kubernetes/system/traefik/values.yaml

@@ -11,28 +11,48 @@ ports:
                 # caddy
                 - 10.1.2.152/32
                 - 10.0.0.0/8
+# Tolerate failover node
+tolerations:
+    - key: ENC[AES256_GCM,data:730ZJScF+2rP,iv:n1QgsKA3w6rv1d2tQddG8FkmADp6tI1gj+p7FnD1uA8=,tag:wtlI+476AdFHSp8I68Qs4Q==,type:str]
+      operator: Equal
+      value: ENC[AES256_GCM,data:uSBDyP1NASA=,iv:wFwuuUWxQKqmb8HMVvbBmMEwuWfxrCVE+soDvM0co6s=,tag:1OZ1KaTKMfMhSZhA1+/zXg==,type:str]
+      effect: NoSchedule
+    - key: ENC[AES256_GCM,data:eWiYi/AytLLefqvh2WeoC557WVtWH7c5wYFZYHH0oaIPHd/UyQ==,iv:zreUmSRjGCYywVO1w7+axPmt1E6WON9k63EWhzN6L5c=,tag:zS/rA1dTwp8QMzdi8g8uLg==,type:str]
+      operator: Exists
+      effect: NoSchedule
+# Prefer non-failover nodes in normal operation
+affinity:
+    nodeAffinity:
+        preferredDuringSchedulingIgnoredDuringExecution:
+            - weight: 100
+              preference:
+                matchExpressions:
+                    - key: ENC[AES256_GCM,data:1eMq19wPgLa5uYzYlrrN783Ps5U=,iv:WU2PcuCN9YsT3k86TN9347iErqH7s5JiOKlPfiBK2oU=,tag:BWEdA34tVe8FYUjaztPTqw==,type:str]
+                      operator: NotIn
+                      values:
+                        - "true"
 sops:
-    lastmodified: "2026-05-26T11:18:08Z"
+    lastmodified: "2026-07-25T08:22:16Z"
-    mac: ENC[AES256_GCM,data:mA5hLNB0rwSiGhnyi24AhZIPJsLpZ6PpbXDyoxZ0q6YjitrClxBEnn2dHtEl2MD6dSLmNMVxnnGyGtl7j4ahfqhuct+oPSepeWT1QX8Xj/mJ2Yrt8UZfGQ1R0Ye+rKGFybluMguCRufioGQpU3TLs2TxB6RxUAiGMI1GyT3JBDY=,iv:Pf617ZQBgYbGEsF7AOtyZBCPUycQ7U/D+Sdl+MCF4y0=,tag:tleTblRukBO0V+zfL05fQw==,type:str]
+    mac: ENC[AES256_GCM,data:3L2sGcpCY1DbFib+13cTJ382EB3DgDJyT5nrE6Pz2gJEqDoF867vDYFvl7qxmT0VXaX6a9g/Q0j97EBbq+oiko0t+sYU0dCRRvcnmgbZaY6TA6WhwMzB6STqZW0XhUc6o5mgvIh7YCgCpPl1yFI4Oeffkrd7EQnx66FPxZYtNzg=,iv:PBZSXeYx8ywDR3wsahldquUfUDTNgC+znYdVcV9SDuw=,tag:d48/XYEaIC5I6cZXCRW/kg==,type:str]
     pgp:
-        - created_at: "2026-05-26T11:18:08Z"
+        - created_at: "2026-07-25T08:22:16Z"
           enc: |-
             -----BEGIN PGP MESSAGE-----
 
-            hQIMA7uy4qQr71wiAQ/+PJv3xyn1nqVN4ENsKv1GgMaCrmD+F1bpyCEM6quoOe1V
+            hQIMA7uy4qQr71wiARAAxFUgi7EmKbvClLsuCBT3H40QeXdnEoTYmpVJHhoUp9ex
-            Q3jnfTrHNKQySQFS+56KdWkpvf54jUtl2N+v+mAT+02FkdH+fkFQc0rSfqfpwE7H
+            vw8K49rg1UueJ7E3jenOhsFw7XiuH3JfCu/CjRW4kMdhBm9G7vGQLdoUub0gvW6Q
-            8peu9wY5+nVRp3sUEacJjCgu44dzzIu0MzO9aHZw/JMr1Z+OF3mMeZ1vVEp43i6E
+            OgJLJcpoYaIkTowZLHxyQOVYv23FFHCfoEWHUEidHnDswy6ngtljS9U3aQFq67oR
-            ZwCd3HYBQVJ8DaSdkrT/a3r69dJZKYq3mL9XLYzc/6C4JFAF3oOO5Xzkx4BAQKrM
+            REbqMcaotkAUg921ZA60BrctJFvm5DKt33LbgrI6NQ7ku4cDHnRsJLxW0X5F0srl
-            CDyVCJ6RihOyysiexRFGu72wIbiE1gSVqLAYzl8XDJyCcVOX8ZRuMh2ImGyuCXtI
+            vuHfez7wTk/az7H07aPGlNcBc3MiXmpC9EpRj3ZCgi8RsWm58MFO4ErNAF1DGd1A
-            8dANf39gf57Z3qM9ljNHgtUkFqcpeE66SdNCSeLK7pDl/02k6dPJr98+e0VtZPfv
+            flfpDRyoDxichpKGnNs1vrWOI3ceXj5241HsQne7dqBPO2sn0QhY0tN9imjKteab
-            xt/sFUnAppi5dI4I4AZ1upqFWH3zpud+3G+5CYnfn0yqSgysuJyqzSYv7rjJVStY
+            bn3C3QSh/jePSJYoBE1tIr0b3+7NIUKeEHBZuS250KL3nZGAGHmYRGZubvrLag9u
-            oMdKprkLGcdLc9FJm3yaPluCnTGpPCaBpvx1bUlQSCIHcrbNj98kHjnuSDcUjgBc
+            pnEWZw1yGI2glR/aXFgOk4eHqGfNkBSCQTN/7vErYe1FmUbsaiiJBLpzmNDPrWGy
-            xpcWmknbsZ0SFnU8fhe/p80Ud7h2Ya5Hp+GY+FPV2YDeQf03YKLGXujsVhm0scpo
+            d77Zv/GVCQ5Tod0sD714IyEj5j2ZH9nBGXW60VY/indmMoFpWlUzvii5x3g8Ig6G
-            KtAALuJo8uP3vniAaY6E1eQVZ09psqR3lFHgaRIyLVbOGkvOoE/sMVcS/odcKhgw
+            R0cG20g/x/Q2IgEQuisVfZBq28hJ9SDQNequ+XjARWeO4Bg6VbQvYSiPMamEd1U4
-            Gr6bb2iQ9b0seITxk09HV84uRirzgR+R2A2bxD38ASRzbYwDyA0UnJMSJK+ZttfS
+            IBBWy+Q/MWMaMXN/igfEptvC89ttqK4ckPjgiN8dQyG97jy+xs0kll7JGvydyK/S
-            XgERXIWVWlTL77IVifpdu+7EU+ElyCzlTLHIb0uqywVS2o/LzE+OgR+1tSj0SVl0
+            XAGFitVH3lGXSmsDfijXKVP9/dUixM4Z38lxOwlb6cUyyREH6rRjWyqJcM9tTUVo
-            +NvGYouZlrecytaBj8MG9thv9kK/vNShrl0QeAWNch3Qza7Xb8eLOhumqZVFI9U=
+            ArLybrpWYkZ/eVWi2/EXQ7I9nWLoivzrilf80Gh4VPBesuisZdATZUwUY0EG
-            =JbZl
+            =L8ob
             -----END PGP MESSAGE-----
           fp: DC6910268E657FF70BA7EC289974494E76938DDC
     encrypted_regex: ^(password|value|ssh-key|api-key|user|username|privateKey|clientSecret|clientId|apiKey|extraArgs.*|.*Secret.*|extraEnvVars|.*SECRET.*|.*secret.*|key|.*Password|.*\.ya?ml)$