fix: [alert] KubeletTooManyPods (cluster) (#937 )

Merge pull request 'fix(alertmanager): route InfoInhibitor to null and add the info inhibit rules' (#936 ) from fix/alertmanager-info-inhibitor into master
Reviewed-on: k.renner/homelab#936
2026-09-23 19:44:05 +00:00 · 2026-09-23 19:24:22 +00:00 · 2026-09-23 21:23:29 +02:00 · 2026-09-23 19:15:33 +00:00 · 2026-09-23 21:12:43 +02:00 · 2026-09-23 19:04:44 +00:00
31 changed files with 428 additions and 264 deletions
@@ -1,12 +1,13 @@
 # Nightly wiki maintenance for the Notes vault (k.renner/notes), replacing the
-# workstation's wiki-maintenance.timer: distill the raw/ backlog into wiki/,
-# then audit wiki/. Each agent is its own `claude` run, not a subagent of an
-# orchestrator, so nothing depends on background tasks in print mode (the
-# workstation job died on their 600 s ceiling). The pipeline commits and pushes
-# as Claude Bot; the push webhook then refreshes the wiki-index service.
+# workstation's wiki-maintenance and wiki-consolidate timers: distill the raw/
+# backlog into wiki/ (wiki-distill agent), audit wiki/ with the vault's own
+# script (structural checks, no model), then consolidate (wiki-consolidate
+# agent: verifies a cohort of notes against sources, repos and the live
+# cluster and edits them). The pipeline commits and pushes as Claude Bot; the
+# push webhook then refreshes the wiki-index service.
 #
-# The agents and their retrieval scripts address the vault as ~/Notes, so each
-# agent step links $HOME/Notes to the checkout instead of rewriting them.
+# The agents address the vault as ~/Notes and the repos as ~/source/<repo>, so
+# each agent step links those into the workspace instead of rewriting them.
 when:
  - event: cron
    cron: wiki-maintenance
@@ -30,6 +31,27 @@ steps:
        git -c http.extraheader="AUTHORIZATION: token $$TOKEN" \
          clone --depth 50 --branch main https://gitea.renner.uno/k.renner/notes.git notes
        echo "raw backlog: $$(ls notes/raw/*.md | grep -v _index.md | wc -l) file(s)"
+        # Full history for the consolidate agent's `git log` checks; the
+        # workspace clone is shallow and tree-filtered.
+        mkdir -p sources
+        git clone --quiet https://gitea.renner.uno/k.renner/homelab.git sources/homelab
+        git -c http.extraheader="AUTHORIZATION: token $$TOKEN" \
+          clone --quiet https://gitea.renner.uno/k.renner/homeassistant.git sources/homeassistant
+
+  - name: fetch-diag
+    # Read-only live-cluster access for the consolidate agent, same mechanism
+    # as issue-fix.yml: ./.ci-diag/kubectl plus a 1h ci-diagnostics token.
+    image: gitea.renner.uno/gitea_admin/ci-k8s:2026-09-08
+    environment:
+      SOPS_AGE_KEY:
+        from_secret: sops_age_key
+      S3_ACCESS_KEY_ID:
+        from_secret: hetzner_s3_access_key
+      S3_SECRET_ACCESS_KEY:
+        from_secret: hetzner_s3_secret_key
+    commands:
+      - bash scripts/ci/fetch-kubeconfig.sh
+      - bash scripts/ci/make-diag-kubeconfig.sh

  - name: distill
    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.280
@@ -44,24 +66,36 @@ steps:
    commands:
      - |
        set -eu
+        # An empty queue must not start the agent: its startup context alone
+        # costs more than the whole audit.
+        BACKLOG=$$(ls notes/raw/*.md | grep -v _index.md | wc -l)
+        if [ "$$BACKLOG" -eq 0 ]; then
+          echo "distill: raw/ is empty, skipping the agent"
+          exit 0
+        fi
        ln -sfn "$$PWD/notes" "$$HOME/Notes"
        mkdir -p "$$HOME/.config/wiki-index"
        printf '%s' "$$WIKI_INDEX_BASIC_AUTH" | tr -d '\r\n' > "$$HOME/.config/wiki-index/basic-auth"
        chmod 600 "$$HOME/.config/wiki-index/basic-auth"
        "$$HOME/Notes/bin/wiki_rank.py" --probe-index
-      - |
-        set -eu
        MODEL=$$(sh scripts/ci/pick-model.sh "distill raw source notes into atomic wiki notes: read sources, check the wiki for existing coverage, write or merge notes; careful writing, moderate reasoning")
        if [ -n "$$MODEL" ]; then echo "distill: selected model $$MODEL"; else echo "distill: using the CLI default model"; fi
        cd notes
        cat ../ci/agents/wiki-maintenance/distill.md | claude $${MODEL:+--model $$MODEL} \
          --agent wiki-distill \
          --permission-mode acceptEdits \
-          --allowedTools "Read,Edit,Write,Glob,Grep,Bash(/root/Notes/bin/wiki_rank.py:*),Bash(/root/Notes/bin/wiki_ask.py:*),Bash(mv /root/Notes/raw/:*),Bash(mkdir -p /root/Notes/:*)" \
+          --allowedTools "Read,Edit,Write,Glob,Grep,Bash(/root/Notes/bin/wiki_rank.py:*),Bash(/root/Notes/bin/wiki_ask.py:*),Bash(mv /root/Notes/raw/:*),Bash(mkdir -p /root/Notes/:*),Bash(~/Notes/bin/wiki_rank.py:*),Bash(~/Notes/bin/wiki_ask.py:*),Bash(mv ~/Notes/raw/:*),Bash(mkdir -p ~/Notes/:*)" \
          --settings ../ci/agents/wiki-maintenance/settings.json \
          -p

  - name: audit
+    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.280
+    pull: true
+    commands:
+      # Findings are report lines in output/, not a pipeline failure.
+      - python3 -B notes/bin/wiki_audit.py --vault "$$PWD/notes"
+
+  - name: consolidate
    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.280
    pull: true
    environment:
@@ -75,20 +109,40 @@ steps:
      - |
        set -eu
        ln -sfn "$$PWD/notes" "$$HOME/Notes"
+        ln -sfn "$$PWD/sources" "$$HOME/source"
        mkdir -p "$$HOME/.config/wiki-index"
        printf '%s' "$$WIKI_INDEX_BASIC_AUTH" | tr -d '\r\n' > "$$HOME/.config/wiki-index/basic-auth"
        chmod 600 "$$HOME/.config/wiki-index/basic-auth"
-      - |
-        set -eu
-        MODEL=$$(sh scripts/ci/pick-model.sh "structural audit of ~1700 markdown notes: frontmatter completeness, broken links, orphans; mechanical checks, little reasoning")
-        if [ -n "$$MODEL" ]; then echo "audit: selected model $$MODEL"; else echo "audit: using the CLI default model"; fi
+        if [ -f .ci-diag/kubeconfig ]; then
+          ln -sf "$$PWD/.ci-diag/kubectl" /usr/local/bin/kubectl
+          export KUBECONFIG="$$PWD/.ci-diag/kubeconfig"
+        else
+          echo "consolidate: no ci-diag kubeconfig, live cluster checks unavailable"
+        fi
        cd notes
-        cat ../ci/agents/wiki-maintenance/audit.md | claude $${MODEL:+--model $$MODEL} \
-          --agent wiki-audit \
+        # JSON output: the agent's report and the run's cost come out of one
+        # record at the end instead of a 45-minute silent log.
+        # One source for the Bash allowlist: the settings file (run 10608: the
+        # agent writes ~/Notes, the rules carried only /root/Notes).
+        BASH_RULES=$$(jq -r '[.permissions.allow[] | select(startswith("Bash("))] | join(",")' ../ci/agents/wiki-maintenance/consolidate-settings.json)
+        rc=0
+        cat ../ci/agents/wiki-maintenance/consolidate.md | claude \
+          --agent wiki-consolidate \
          --permission-mode acceptEdits \
-          --allowedTools "Read,Write,Glob,Grep,Bash(/root/Notes/bin/wiki_rank.py:*),Bash(/root/Notes/bin/wiki_ask.py:*)" \
-          --settings ../ci/agents/wiki-maintenance/settings.json \
-          -p
+          --allowedTools "Read,Edit,Write,Glob,Grep,Agent,$$BASH_RULES" \
+          --settings ../ci/agents/wiki-maintenance/consolidate-settings.json \
+          --output-format json \
+          -p > /tmp/consolidate.json || rc=$$?
+        # A killed or crashed claude leaves no valid JSON; report rc, not a jq error.
+        if [ -s /tmp/consolidate.json ] && jq -e . /tmp/consolidate.json > /dev/null; then
+          jq -r '.result // empty' /tmp/consolidate.json
+          jq -c '{duration_ms, num_turns, total_cost_usd, is_error}' /tmp/consolidate.json
+        else
+          echo "consolidate: no result record (exit $$rc)"; head -c 2000 /tmp/consolidate.json || true
+        fi
+        echo "consolidate: git log --oneline of the run:"
+        git log --oneline origin/main..HEAD
+        exit $$rc

  - name: push-notes
    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.280
@@ -102,7 +156,13 @@ steps:
        cd notes
        git config user.email "claude-bot@renner.uno"
        git config user.name "Claude Bot"
-        if git diff --quiet && git diff --cached --quiet && [ -z "$$(git ls-files --others --exclude-standard)" ]; then
+        # Only the vault content is committed; anything an agent leaves elsewhere
+        # in the checkout (run 10362: scratch scripts in the root) stays behind.
+        # The consolidate agent commits its own snapshots, so HEAD may already be
+        # ahead of origin/main with a clean tree.
+        git add -A -- wiki raw output .wiki-consolidation-state.json
+        git diff --cached --quiet || git commit -q -m "wiki maintenance $$(date -u +%F): distill + audit + consolidate (CI run $$CI_PIPELINE_NUMBER)"
+        if [ "$$(git rev-list --count origin/main..HEAD)" -eq 0 ]; then
          echo "wiki-maintenance: no changes"
          exit 0
        fi
@@ -112,13 +172,11 @@ steps:
        # leaves the PR open for a human and fails the run.
        BRANCH="wiki-maintenance/$$(date -u +%F)-$$CI_PIPELINE_NUMBER"
        git checkout -q -b "$$BRANCH"
-        git add -A
-        git commit -q -m "wiki maintenance $$(date -u +%F): distill + audit (CI run $$CI_PIPELINE_NUMBER)"
        TOKEN=$$(printf '%s' "$$WIKI_BOT_TOKEN" | tr -d '\r\n')
        git -c http.extraheader="AUTHORIZATION: token $$TOKEN" push origin "HEAD:$$BRANCH"
        API="https://gitea.renner.uno/api/v1/repos/k.renner/notes"
        jq -n --arg title "wiki maintenance $$(date -u +%F) (CI run $$CI_PIPELINE_NUMBER)" \
-              --arg body "Nightly distill + audit. Merged by the pipeline; open only if the merge failed." \
+              --arg body "Nightly distill + audit + consolidate. Merged by the pipeline; open only if the merge failed." \
              --arg head "$$BRANCH" '{title: $$title, body: $$body, head: $$head, base: "main"}' > /tmp/pr.json
        code=$$(curl -s -o /tmp/pr-resp.json -w '%{http_code}' -X POST \
          -H "Authorization: token $$TOKEN" -H "Content-Type: application/json" --data @/tmp/pr.json "$$API/pulls")
@@ -1,5 +0,0 @@
-You are running as the scheduled wiki-maintenance job in CI. The Notes vault is checked out at ~/Notes (the repository k.renner/notes).
-
-Run the standard audit pass over ~/Notes/wiki/ and write the dated report to ~/Notes/output/. Do not modify any wiki note. Do not commit or push; the pipeline does that.
-
-End with two lines: notes scanned, and issues found by category.
@@ -0,0 +1,120 @@
+{
+  "//": "Permission scope for the wiki-consolidate agent in CI: read the vault and the",
+  "//2": "source repos cloned under /root/source, read the live cluster through the",
+  "//3": "ci-diagnostics kubeconfig, edit wiki/, output/ and the rotation state only.",
+  "//4": "Deny rules win; git and kubectl are read-only subcommands, no network tools.",
+  "permissions": {
+    "allow": [
+      "Read(//root/Notes/**)",
+      "Read(//root/source/**)",
+      "Edit(//root/Notes/wiki/**)",
+      "Edit(//root/Notes/output/**)",
+      "Edit(//root/Notes/.wiki-consolidation-state.json)",
+      "Bash(/root/Notes/bin/wiki-snapshot.sh:*)",
+      "Bash(/root/Notes/bin/wiki_rank.py:*)",
+      "Bash(/root/Notes/bin/wiki_ask.py:*)",
+      "Bash(~/Notes/bin/wiki-snapshot.sh:*)",
+      "Bash(~/Notes/bin/wiki_rank.py:*)",
+      "Bash(~/Notes/bin/wiki_ask.py:*)",
+      "Bash(~/Notes/bin/wiki-delete.sh:*)",
+      "Bash(git -C /root/Notes log:*)",
+      "Bash(git -C /root/Notes show:*)",
+      "Bash(git -C /root/Notes diff:*)",
+      "Bash(git -C /root/Notes status:*)",
+      "Bash(git -C /root/Notes rev-parse:*)",
+      "Bash(git -C /root/Notes ls-files:*)",
+      "Bash(git -C ~/Notes log:*)",
+      "Bash(git -C ~/Notes show:*)",
+      "Bash(git -C ~/Notes diff:*)",
+      "Bash(git -C ~/Notes status:*)",
+      "Bash(git -C ~/Notes rev-parse:*)",
+      "Bash(git -C ~/Notes ls-files:*)",
+      "Bash(kubectl get:*)",
+      "Bash(kubectl describe:*)",
+      "Bash(kubectl version:*)",
+      "Bash(kubectl api-resources:*)",
+      "Bash(kubectl explain:*)",
+      "Bash(git log:*)",
+      "Bash(git show:*)",
+      "Bash(git status:*)",
+      "Bash(git diff:*)",
+      "Bash(git rev-parse:*)",
+      "Bash(git ls-files:*)",
+      "Bash(git -C /root/source/homelab log:*)",
+      "Bash(git -C /root/source/homelab show:*)",
+      "Bash(git -C /root/source/homelab diff:*)",
+      "Bash(git -C /root/source/homelab status:*)",
+      "Bash(git -C /root/source/homelab rev-parse:*)",
+      "Bash(git -C /root/source/homelab ls-files:*)",
+      "Bash(git -C /root/source/homeassistant log:*)",
+      "Bash(git -C /root/source/homeassistant show:*)",
+      "Bash(git -C /root/source/homeassistant diff:*)",
+      "Bash(git -C /root/source/homeassistant status:*)",
+      "Bash(git -C /root/source/homeassistant rev-parse:*)",
+      "Bash(git -C /root/source/homeassistant ls-files:*)",
+      "Bash(git -C ~/source/homelab log:*)",
+      "Bash(git -C ~/source/homelab show:*)",
+      "Bash(git -C ~/source/homelab diff:*)",
+      "Bash(git -C ~/source/homelab status:*)",
+      "Bash(git -C ~/source/homelab rev-parse:*)",
+      "Bash(git -C ~/source/homelab ls-files:*)",
+      "Bash(git -C ~/source/homeassistant log:*)",
+      "Bash(git -C ~/source/homeassistant show:*)",
+      "Bash(git -C ~/source/homeassistant diff:*)",
+      "Bash(git -C ~/source/homeassistant status:*)",
+      "Bash(git -C ~/source/homeassistant rev-parse:*)",
+      "Bash(git -C ~/source/homeassistant ls-files:*)",
+      "Bash(/root/Notes/bin/wiki-delete.sh:*)",
+      "Bash(ls:*)",
+      "Bash(date:*)",
+      "Bash(wc:*)",
+      "Bash(stat:*)",
+      "Bash(head:*)",
+      "Bash(tail:*)"
+    ],
+    "deny": [
+      "Read(//proc/**)",
+      "Read(//etc/**)",
+      "Read(//root/.config/**)",
+      "Read(//root/.kube/**)",
+      "Read(//woodpecker/**/.ci-diag/**)",
+      "Read(//root/Notes/.git/**)",
+      "Read(//root/source/*/.git/**)",
+      "Read(//root/Notes/_archive/**)",
+      "Edit(//root/Notes/raw/**)",
+      "Edit(//root/Notes/_archive/**)",
+      "Edit(//root/source/**)",
+      "Bash(kubectl apply:*)",
+      "Bash(kubectl delete:*)",
+      "Bash(kubectl patch:*)",
+      "Bash(kubectl edit:*)",
+      "Bash(kubectl scale:*)",
+      "Bash(kubectl rollout:*)",
+      "Bash(kubectl exec:*)",
+      "Bash(kubectl cp:*)",
+      "Bash(kubectl drain:*)",
+      "Bash(kubectl cordon:*)",
+      "Bash(kubectl label:*)",
+      "Bash(kubectl annotate:*)",
+      "Bash(git push:*)",
+      "Bash(git reset:*)",
+      "Bash(git checkout:*)",
+      "Bash(git clean:*)",
+      "Bash(git rebase:*)",
+      "Bash(git config:*)",
+      "Bash(git -C /root/Notes config:*)",
+      "Bash(git -C ~/Notes config:*)",
+      "Bash(git -C /root/source/homelab config:*)",
+      "Bash(git -C /root/source/homeassistant config:*)",
+      "Bash(git -C ~/source/homelab config:*)",
+      "Bash(git -C ~/source/homeassistant config:*)",
+      "Bash(rm:*)",
+      "Bash(rm -rf:*)",
+      "Bash(curl:*)",
+      "Bash(wget:*)",
+      "Bash(ssh:*)",
+      "Bash(tea:*)",
+      "Bash(woodpecker:*)"
+    ]
+  }
+}
@@ -0,0 +1,5 @@
+You are running as the scheduled wiki-maintenance job in CI. The Notes vault is checked out at ~/Notes (the repository k.renner/notes); the homelab and homeassistant repositories are checked out under ~/source/. `kubectl` reads the live cluster with a read-only identity.
+
+Run the wiki-consolidate agent now for today's scheduled consolidation pass. Follow its instructions exactly: snapshot the vault to git first, abort if that fails, select today's rotating cohort of 20 from the state file, verify claims against both their sources and the live systems, act on the verdicts, write the dated report to ~/Notes/output/, update the state file, and snapshot again at the end. Do not push; the pipeline pushes the run's commits.
+
+Report the counts and the coverage progress when you are done.
@@ -1,5 +1,5 @@
 {
-  "//": "Permission scope for the wiki-maintenance agents (wiki-distill, wiki-audit) in CI.",
+  "//": "Permission scope for the wiki-maintenance agent (wiki-distill) in CI.",
  "//2": "The step's environment holds a Gitea push token, a TypeSafe key and the wiki-index",
  "//3": "client credential, so the Read tool is kept away from /proc (env dumps), /root outside",
  "//4": "the vault symlink target, /etc and the checkout's .git/. The vault itself is the work.",
@@ -15,6 +15,10 @@
      "Bash(/root/Notes/bin/wiki_ask.py:*)",
      "Bash(mv /root/Notes/raw/:*)",
      "Bash(mkdir -p /root/Notes/:*)",
+      "Bash(~/Notes/bin/wiki_rank.py:*)",
+      "Bash(~/Notes/bin/wiki_ask.py:*)",
+      "Bash(mv ~/Notes/raw/:*)",
+      "Bash(mkdir -p ~/Notes/:*)",
      "Bash(ls:*)",
      "Bash(date:*)",
      "Bash(wc:*)",
@@ -1,6 +1,6 @@
 FROM alpine:3.21@sha256:48b0309ca019d89d40f670aa1bc06e426dc0931948452e8491e3d65087abc07d

-ARG TALOS_VERSION=v1.13.9
+ARG TALOS_VERSION=v1.13.10
 ARG VELERO_VERSION=v1.16.0
 ARG KUBECTL_VERSION=v1.36.3
 ARG FLUX_VERSION=2.9.5
@@ -18,7 +18,7 @@ RUN apk add --no-cache \

 RUN curl -sfL -o /usr/local/bin/talosctl \
      "https://github.com/siderolabs/talos/releases/download/${TALOS_VERSION}/talosctl-linux-amd64" \
- && echo "7e1d4b7d5846964bdcf63a794e3c8161bb6ef2983d5ace58ea5322f3bf32a27e  /usr/local/bin/talosctl" | sha256sum -c - \
+ && echo "28e9ccd9aebd690fa03d8cf2582221d87ea38b2a1704f411d75a03589c58f2aa  /usr/local/bin/talosctl" | sha256sum -c - \
 && chmod +x /usr/local/bin/talosctl

 RUN curl -sfL -o /usr/local/bin/kubectl \
@@ -14,7 +14,7 @@ FROM alpine:3.21@sha256:48b0309ca019d89d40f670aa1bc06e426dc0931948452e8491e3d650
 # kubernetes/talos/patches/cluster.yaml. The phase reads the version from that
 # file at run time and passes it to the provisioner; this pin only decides which
 # client talks to it.
-ARG TALOS_VERSION=v1.13.9
+ARG TALOS_VERSION=v1.13.10
 ARG KUBECTL_VERSION=v1.36.2
 ARG FLUX_VERSION=2.9.2
 ARG KOPIA_VERSION=0.23.1
@@ -49,7 +49,7 @@ RUN apk add --no-cache \

 RUN curl -sfL -o /usr/local/bin/talosctl \
      "https://github.com/siderolabs/talos/releases/download/${TALOS_VERSION}/talosctl-linux-amd64" \
- && echo "7e1d4b7d5846964bdcf63a794e3c8161bb6ef2983d5ace58ea5322f3bf32a27e  /usr/local/bin/talosctl" | sha256sum -c - \
+ && echo "28e9ccd9aebd690fa03d8cf2582221d87ea38b2a1704f411d75a03589c58f2aa  /usr/local/bin/talosctl" | sha256sum -c - \
 && chmod +x /usr/local/bin/talosctl

 RUN curl -sfL -o /usr/local/bin/kubectl \
@@ -1,117 +1,99 @@
 # Traffic classes on the k8s nodes

-Issue #576. Three of the four classes are separated; management is not.
+Issue #576. Five classes, each on its own VLAN.

 ## Where each class runs

 | class | link | mechanism |
 |---|---|---|
 | Ceph replication | `ceph-san`, VLAN 49 | the whole link is handed to the OSD pod (Multus `host-device`) |
-| world→pod | `ingress0`, VLAN 47 | BGP session sourced there, so the advertised next hop is that address |
-| pod→world | `ingress0`, VLAN 47 | routing rule on the pod source prefix → table 47 |
-| pod↔pod | `pod0`, VLAN 48 | routing rule on pod source **and** destination → table 48 |
-| management | `mgmt0`, VLAN 40 | not started |
+| pod↔pod | `pod0`, VLAN 48 | pod-sourced traffic is routed from table 47; bird puts the peers' pod CIDRs there via `pod0` |
+| world↔pod | `ingress0`, VLAN 47 | BGP session sourced there (next hop of every LoadBalancer VIP); pod egress takes table 47's default |
+| node↔node | `node0`, VLAN 46 | the node InternalIP and etcd live there; so do apiserver↔kubelet, cilium-health and host→remote-pod traffic |
+| world↔node | `mgmt0`, VLAN 40 | API VIP for kubectl, talosctl from outside, node egress (DNS, NTP, image pulls) |

 Hard constraint, set by the owner: inter-node and pod traffic never traverses
-the UDM. Everything above stays on the 10G switch.
+the UDM. Everything but world↔node stays on the 10G switch. VLAN 46 and 48 are
+statically addressed, without a gateway; VLAN 46 is isolated in UniFi.

-## Why policy routing rather than moving the node IP
+## Pod traffic: one table

-Cilium reaches a peer's pods at that peer's Kubernetes InternalIP, in native
-and tunnel mode alike. `--ipv4-node` is overwritten from the Node object, and
-`nodePort.directRoutingDevice` selects the BPF NodePort device, not a next hop.
-Both were established by test (#837, #838), not by reading.
+Two rules, one per family, send every packet with a pod source to table 47
+(`talos/patches/site-homelab.yaml`, priorities 1000/1001, below main). The
+table decides by longest prefix; no rule carves out exceptions.

-So the InternalIP decides where pod traffic goes — unless the lookup happens in
-a different table. The rules do that, and nothing about the node's identity
-moves. The routing decision runs before Cilium's masquerade hook, so a rule
-matching the pod source sees the pod address, and the masquerade then rewrites
-it to the egress link's.
+bird (`kubernetes/infrastructure/bird/`) owns table 47 and fills it with:

-## Rule order
+- the node's attached networks — `mgmt0`, `ingress0`, `pod0`, `node0` —
+  taken from the interfaces;
+- the peers' pod CIDRs, learned over an iBGP mesh on `pod0`, next hop the
+  peer's `pod0` address;
+- the node's own pod CIDR via `cilium_host`, originated rather than read back
+  from the kernel (reading it back made BIRD depend on the shape of Cilium's
+  route, and KRT rejects the IPv4 one);
+- the egress default via the UDM on `ingress0`, from the node annotations
+  `net.renner.uno/egress-gw-v4/v6`.

-Lower priority runs first.
+`node0` has to be in the table: a pod reaching a node address — the
+`kubernetes` ClusterIP, which socket LB rewrites to the apiserver's advertise
+address, or a hostNetwork pod — would otherwise take the default to the UDM.

-| prio | match | table |
-|---|---|---|
-| 980 | src pod, dst pod | 48 |
-| 981 | src pod, dst pod (v4) | 48 |
-| 990 | dst `fd00::/8` | main |
-| 991 | dst `10.0.0.0/8` | main |
-| 1000 | src pod | 47 |
-| 1001 | src pod (v4) | 47 |
+The routing decision runs before Cilium's masquerade hook, so the rule sees the
+pod address and the masquerade then rewrites it to the egress link's.

-980/981 match source **and** destination. Source alone would also catch the
-node's own traffic to a pod, which would then leave with a `pod0` source — an
-address `CiliumNode.spec.addresses` does not list, so Cilium classifies it as
-`world` rather than `host`.
-
-990/991 are Cilium's `ipv4/ipv6NativeRoutingCIDR`, and they are what keeps
-cluster-internal traffic out of table 47. Without them a pod's packet to a
-service matches the source rule, finds only a default route, and leaves for the
-UDM — measured, in-cluster DNS stopped resolving.
-
-Traffic to a pod on the same node falls through on its own: table 48 holds only
-the peers' CIDRs, so the lookup misses and 990/991 send it to main.
-
-## Where each table comes from
-
- **47** — two default routes declared per node in `talos/patches/k8s-*.yaml`.
-  The IPv6 gateway is the UDM's configured address on the segment, not the
-  RA-learned link-local; `accept_ra_defrtr` stays `0` on `ingress0` so the RA
-  default does not land in main.
- **48** — filled by the `bird` DaemonSet: an iBGP mesh over `pod0`, one route
-  per peer pod CIDR with the peer's `pod0` address as next hop.
-
-## The bird mesh
-
-`kubernetes/infrastructure/bird/`. Each node advertises only its own pod CIDR
-and installs what it learns into table 48. Cilium's routes in main are
-untouched, which is why the two coexist.
+Node-originated traffic does not match the rules and uses main, where Cilium
+routes each peer's pod CIDR via the peer's InternalIP — on `node0`.

 No address is written into this repo twice. Each node publishes its `pod0`
-addresses as `nodeAnnotations`, beside the `VLANConfig` that assigns them; the
-pod CIDRs come from the Node objects. An init container renders both from the
-API with `kubectl -o go-template-file`, excludes the node itself with a field
-selector, and ends with `bird -p` so BIRD's own parser gates startup.
+addresses as `nodeAnnotations` beside the `VLANConfig` that assigns them; the
+pod CIDRs come from the Node objects. bird's init container renders both from
+the API and ends with `bird -p`, so BIRD's own parser gates startup.

-The node's own CIDR is **originated**, not read back from the kernel. Reading
-it back made BIRD depend on the shape Cilium installs, and the IPv4 route is
-rejected outright: its next hop sits inside the prefix it describes, which KRT
-refuses with `strange next-hop`. The IPv6 route is a device route and passed,
-so the failure was silent on one family only.
+## Node↔node
+
+`kubelet.nodeIP.validSubnets` and `cluster.etcd.advertisedSubnets` point at
+VLAN 46. Talos picks one InternalIP per family, IPv4 first — the service CIDR
+is IPv4, and the apiserver's advertise address must match it.
+
+The API VIP stays on `mgmt0` for kubectl. No in-cluster client uses it:
+kubelet, scheduler, controller-manager and Cilium go through KubePrism
+(`localhost:7445`), which balances over every address of every control-plane
+member; pods use the ClusterIP.
+
+Talos never updates an existing etcd member's peer URLs
+(siderolabs/talos#12720). Changing `advertisedSubnets` means, per member:
+stage the config, `talosctl etcd forfeit-leadership`, `talosctl etcd leave`,
+reboot. `leave` removes only `/var/lib/etcd`; the node rejoins as a learner
+and is promoted. EPHEMERAL does not need a wipe.
+
+## Moving node IPs: the mixed state
+
+While some nodes have their InternalIP on the new network and some on the old,
+host→remote-pod traffic leaves with a node address that is not an InternalIP:
+the kernel picks the source from the route to the peer's InternalIP. Cilium
+classifies that as `world`, and every ingress policy that admits only
+`host`/`remote-node` (rook-ceph, monitoring, …) drops it. Kernel Ceph clients
+lose the OSDs on the other side; RBD mounts hang.
+
+Bridge it for the duration with a temporary `CiliumClusterwideNetworkPolicy`
+admitting the node subnets of both networks, both families, with
+`enableDefaultDeny: false` so it restricts nothing else, and delete it once
+every node has moved. Keep all OSD nodes schedulable while one is drained:
+draining two takes Ceph below its data availability.

 ## Failure behaviour

-Benign at every layer. An empty or absent table 48 makes 980/981 fall through
-to main, and pod traffic returns to VLAN 40 — unseparated, not broken. The same
-holds if the DaemonSet cannot start, which is the case during a cold rebuild
-until Gitea serves its image.
-
-## Management, the remaining class
-
-Not started. The levers differ per service and none of them is the node IP:
-
-| service | can leave the node IP? | lever |
-|---|---|---|
-| etcd | yes | `cluster.etcd.advertisedSubnets`; needs a reboot per node |
-| kube-apiserver endpoint | yes | it is a VIP — `Layer2VIPConfig.link` |
-| Talos API | n/a | `apid` answers on every node address; a firewall question |
-| kubelet | no | it *is* the node IP |
-
-Two approaches were rejected by the owner and are not to be proposed again:
-moving the node IP and etcd to `pod0` (#838), and relabelling VLAN 40 as the
-cluster VLAN while moving etcd and the API VIP to a new management VLAN.
-Replacing the CNI (#839, Antrea) was shelved.
-
-Note that the reasoning which produced that list also concluded pod↔pod was
-blocked, which policy routing disproved. Treat it as a record of decisions, not
-as an exhaustive analysis.
+If bird stops, table 47 empties, the rules find nothing and fall through to
+main: pod traffic returns to the node networks, unseparated rather than
+misrouted. The same holds during a cold rebuild until Gitea serves bird's
+image. The default route lives in table 47, not in the Talos config, so a
+default cannot outlive the specific routes and send pod traffic to the UDM.

 ## Sources

-Cilium v1.20.1 `pkg/datapath/linux/node.go`, `pkg/node/types/node.go`,
-`pkg/node/sync/local_node_sync.go`; BIRD 2.17 user manual (kernel protocol,
-`learn all`, `strict bind`, `direct`); Talos v1.13 `RoutingRuleConfig` and
-`machine.nodeAnnotations`. PRs #831, #833, #835, #836, #837, #842, #852, #854,
-#855; closed unmerged: #830, #838, #839.
+Cilium v1.20 `pkg/datapath/linux/node.go`; BIRD 2.17 user manual (kernel
+protocol, `strict bind`, `direct`); Talos v1.13 `RoutingRuleConfig`,
+`machine.nodeAnnotations`, configuration patching (lists are appended,
+`$patch: delete`), etcd `LeaveCluster`; KubePrism docs;
+siderolabs/talos#12720. PRs #831, #833, #835, #836, #837, #842, #852, #854,
+#855, #857, #858, #918, #919, #922; closed unmerged: #830, #838, #839.
@@ -107,7 +107,7 @@ config, and the nodes are addressed by name.
 ## Step 1b — boot all three machines from install media

 ```
-https://factory.talos.dev/image/0f6fb8bd8bb53013e9743b7c7b099d96fb9c32fd2464ced32693c2dc0ea5feb9/v1.13.9/metal-amd64.iso
+https://factory.talos.dev/image/0f6fb8bd8bb53013e9743b7c7b099d96fb9c32fd2464ced32693c2dc0ea5feb9/v1.13.10/metal-amd64.iso
 ```

 Write it to a USB stick and boot each MS-01 from it. Each comes up in
@@ -6,7 +6,6 @@ kind: Kustomization
 # and verified. See MIGRATION-gitea-woodpecker.md.
 resources:
  - statefulset.yaml
-  - pdb.yaml
  - service.yaml
  - ingressroute-ssh.yaml
  - ingress.yaml
@@ -1,23 +0,0 @@
---
-# Eviction protection for the single-replica Gitea StatefulSet.
-#
-# Gitea runs on SQLite (single writer) and its repos + DB live on a single
-# ReadWriteOnce PVC, so it is deliberately replicas:1 and cannot fail over.
-# With no PDB, a node drain / maintenance evicts the pod abruptly, mid-write,
-# with zero coordination. minAvailable:1 on a 1-replica workload makes the
-# eviction API refuse *voluntary* evictions: `kubectl drain` blocks and
-# surfaces this pod, forcing a conscious decision (drain with
-# --disable-eviction, or scale down first) instead of yanking the live DB.
-# It does NOT block involuntary disruptions (node crash, OOM).
-apiVersion: policy/v1
-kind: PodDisruptionBudget
-metadata:
-  name: gitea
-  namespace: gitea
-  labels:
-    app.kubernetes.io/name: gitea
-spec:
-  minAvailable: 1
-  selector:
-    matchLabels:
-      app.kubernetes.io/name: gitea
@@ -4,7 +4,9 @@ metadata:
  name: gf-pg
  namespace: monitoring
 spec:
-  instances: 1
+  instances: 2
+  smartShutdownTimeout: 10
+  stopDelay: 60
  imageName: ghcr.io/cloudnative-pg/postgresql:16
  storage:
    storageClass: ceph-block
@@ -30,7 +30,9 @@ metadata:
  name: ha-pg
  namespace: home-assistant
 spec:
-  instances: 1
+  instances: 2
+  smartShutdownTimeout: 10
+  stopDelay: 60
  imageName: ghcr.io/cloudnative-pg/postgresql:16
  storage:
    storageClass: ceph-block
@@ -124,8 +126,7 @@ spec:
      # padding rather than data.
      #
      # The cost is RPO: up to 30min of WAL is unrecoverable if this cluster and
-      # its volume are lost outright. This is a single-instance cluster, so the
-      # archive is the only off-node copy.
+      # its volumes are lost outright.
      archive_timeout: "30min"
      # Fewer checkpoints => fewer full-page writes, since a page is only
      # fully-logged on its first modification after a checkpoint.
@@ -59,8 +59,7 @@ spec:
      # padding rather than data.
      #
      # The cost is RPO: up to 30min of WAL is unrecoverable if this cluster and
-      # its volume are lost outright. This is a single-instance cluster, so the
-      # archive is the only off-node copy.
+      # its volumes are lost outright.
      archive_timeout: "30min"
      # Write-reduction tuning. These databases are largely idle, and the
      # Postgres defaults favour throughput over storage that wears out.
@@ -61,6 +61,9 @@ spec:
      enabled: true
      storageClass: ceph-block
      size: 50Gi
+    # A minAvailable:1 PDB on a single replica blocks every node drain.
+    pdb:
+      create: false
    # Bootstrap admin/org/bucket on the fresh PVC. Admin password + token come
    # from the influxdb-bootstrap Secret (OpenBao via ESO), deterministic across
    # rebuilds so the bootstrap token is the live admin token. `influx setup`
@@ -57,10 +57,14 @@ spec:
          # the ClusterSecretStore goes InvalidProviderConfig and no
          # ExternalSecret can refresh.
          #
-          # No window: below quorum the secret store is down and every
-          # ExternalSecret stalls with it, so waiting only delays the page.
+          # Short window: readiness is only a proxy for quorum. A one-second
+          # node NotReady flap makes the node lifecycle controller mark that
+          # node's pods not-ready for ~30s while raft keeps its leader and
+          # serves (issue #920, during a rolling reboot with one replica
+          # already evicted). A really sealed or crashed majority persists,
+          # and existing Secrets keep their last values, so 2m costs little.
          expr: max by (namespace, statefulset) (kube_statefulset_status_replicas_ready{namespace="openbao", statefulset="openbao"}) < 2
-          for: 0m
+          for: 2m
          labels:
            severity: critical
          annotations:
@@ -304,6 +304,10 @@ spec:
            - matchers:
                - alertname = "KubeCPUOvercommit"
              receiver: 'null'
+            # Helper alert that only drives the info inhibit rule below.
+            - matchers:
+                - alertname = "InfoInhibitor"
+              receiver: 'null'
            # Flux notification-controller posts info-severity events
            # here (e.g. image-automation commits). Keep them visible
            # but don't repeat-spam every 12h on the same event.
@@ -313,6 +317,23 @@ spec:
              group_wait: 1m
              group_interval: 10m
              repeat_interval: 24h
+        # From kube-prometheus manifests/alertmanager-secret.yaml.
+        inhibit_rules:
+          - source_matchers:
+              - severity = "critical"
+            target_matchers:
+              - severity =~ "warning|info"
+            equal: ['namespace', 'alertname']
+          - source_matchers:
+              - severity = "warning"
+            target_matchers:
+              - severity = "info"
+            equal: ['namespace', 'alertname']
+          - source_matchers:
+              - alertname = "InfoInhibitor"
+            target_matchers:
+              - severity = "info"
+            equal: ['namespace']
        receivers:
          - name: email
            email_configs:
@@ -11,7 +11,6 @@ resources:
  - protonmail-bridge.yaml
  - ftp-scanner.yaml
  - syncthing.yaml
-  - pdb.yaml
  - network-policy.yaml
  # pprl-pg CNPG Postgres + data/media/consume/redis PVCs (moved here from the
  # retired seed-paperless KS; each carries prune: disabled so ownership
@@ -1,37 +0,0 @@
---
-# Eviction protection for the single-replica stateful paperless sidecars.
-#
-# protonmail-bridge and syncthing each hold state on a single RWO PVC and run
-# replicas:1. minAvailable:1 makes the eviction API refuse voluntary evictions,
-# so a node drain blocks and surfaces these pods instead of evicting them
-# abruptly. Involuntary disruptions (node crash) are not affected.
-#
-# NOTE: the other single-replica paperless pods (paperless, redis, gotenberg,
-# tika, ftp-scanner) are intentionally omitted — they are caches / stateless
-# converters where a blocking PDB would only wedge node drains with no
-# recoverability benefit.
-apiVersion: policy/v1
-kind: PodDisruptionBudget
-metadata:
-  name: protonmail-bridge
-  namespace: paperless
-  labels:
-    app.kubernetes.io/name: protonmail-bridge
-spec:
-  minAvailable: 1
-  selector:
-    matchLabels:
-      app.kubernetes.io/name: protonmail-bridge
---
-apiVersion: policy/v1
-kind: PodDisruptionBudget
-metadata:
-  name: syncthing
-  namespace: paperless
-  labels:
-    app.kubernetes.io/name: syncthing
-spec:
-  minAvailable: 1
-  selector:
-    matchLabels:
-      app.kubernetes.io/name: syncthing
@@ -4,7 +4,7 @@ metadata:
  name: tr-pg
  namespace: trmnl
 spec:
-  instances: 1
+  instances: 2
  smartShutdownTimeout: 10
  stopDelay: 60
  # PG 18, NOT the repo-standard 16. Terminus ships a committed
@@ -47,8 +47,7 @@ spec:
      # padding rather than data.
      #
      # The cost is RPO: up to 30min of WAL is unrecoverable if this cluster and
-      # its volume are lost outright. This is a single-instance cluster, so the
-      # archive is the only off-node copy.
+      # its volumes are lost outright.
      archive_timeout: "30min"
      # Write-reduction tuning. These databases are largely idle, and the
      # Postgres defaults favour throughput over storage that wears out.
@@ -22,7 +22,9 @@ metadata:
    # migration deliberately withheld this while a rollback still existed.
    kustomize.toolkit.fluxcd.io/prune: disabled
 spec:
-  instances: 1
+  instances: 2
+  smartShutdownTimeout: 10
+  stopDelay: 60
  imageName: ghcr.io/cloudnative-pg/postgresql:16
  storage:
    storageClass: ceph-block
@@ -91,8 +93,7 @@ spec:
      # padding rather than data.
      #
      # The cost is RPO: up to 30min of WAL is unrecoverable if this cluster and
-      # its volume are lost outright. This is a single-instance cluster, so the
-      # archive is the only off-node copy.
+      # its volumes are lost outright.
      archive_timeout: "30min"
  resources:
    requests:
@@ -3,7 +3,6 @@ kind: Kustomization
 resources:
  - rbac.yaml
  - server.yaml
-  - pdb.yaml
  - agent.yaml
  - service.yaml
  - ingress.yaml
@@ -22,7 +21,7 @@ resources:
  # OpenBao access, no NetworkPolicy needed. server.yaml/agent.yaml keep their
  # Reloader annotations; externalsecret.yaml keeps the 10m refreshInterval.
  # --- namespace migration, stages 1-2 (issue #433) ---
-  # Stage 2 has moved the workloads above (server, agent, service, ingress, pdb,
+  # Stage 2 has moved the workloads above (server, agent, service, ingress,
  # rbac, secrets, policies) into the `woodpecker` namespace and pointed the
  # server's DSN at wp-pg-rw.woodpecker.svc. Stage 3 retired the `apps`
  # originals; these four now describe the surviving cluster. History and the
@@ -1,27 +0,0 @@
---
-# Eviction protection for the single-replica Woodpecker server StatefulSet.
-#
-# The server is SQLite-backed (single writer) on a single RWO PVC, so it is
-# deliberately replicas:1. minAvailable:1 makes the eviction API refuse
-# voluntary evictions so a node drain blocks and surfaces this pod rather
-# than yanking the live CI DB mid-write. Involuntary disruptions are not
-# affected.
-#
-# The selector MUST include app.kubernetes.io/component: server: the agent
-# Deployment in this same namespace shares app.kubernetes.io/name: woodpecker
-# (component: agent), and a name-only selector would wrongly also cover the
-# stateless, disposable agents.
-apiVersion: policy/v1
-kind: PodDisruptionBudget
-metadata:
-  name: woodpecker-server
-  namespace: woodpecker
-  labels:
-    app.kubernetes.io/name: woodpecker
-    app.kubernetes.io/component: server
-spec:
-  minAvailable: 1
-  selector:
-    matchLabels:
-      app.kubernetes.io/name: woodpecker
-      app.kubernetes.io/component: server
@@ -48,8 +48,8 @@ spec:
        tier: node
        app: bird
    spec:
-      # Carries pod east-west once the routing rules land. Without it table 48
-      # stays empty and the rules fall through to main, i.e. back to VLAN 40.
+      # Owns table 47, where all pod traffic is routed. Without it the table
+      # stays empty and the rules fall through to main, unseparated.
      priorityClassName: system-node-critical
      # Writes host routes and peers on host addresses.
      hostNetwork: true
@@ -1,11 +1,16 @@
 apiVersion: source.toolkit.fluxcd.io/v1
-kind: HelmRepository
+kind: OCIRepository
 metadata:
  name: k8s-gateway
  namespace: kube-system
 spec:
  interval: 24h
-  url: https://k8s-gateway.github.io/k8s_gateway/
+  url: oci://codeberg.org/k8s-gateway/charts/k8s-gateway
+  ref:
+    semver: ">=2.4.0"
+  layerSelector:
+    mediaType: "application/vnd.cncf.helm.chart.content.v1.tar+gzip"
+    operation: copy
 ---
 # In-cluster authoritative resolver for renner.uno.
 #
@@ -28,14 +33,10 @@ metadata:
  namespace: kube-system
 spec:
  interval: 30m
-  chart:
-    spec:
-      chart: k8s-gateway
-      version: ">=2.4.0"
-      sourceRef:
-        kind: HelmRepository
-        name: k8s-gateway
-        namespace: kube-system
+  chartRef:
+    kind: OCIRepository
+    name: k8s-gateway
+    namespace: kube-system
  install:
    remediation:
      retries: 3
@@ -21,17 +21,29 @@ spec:
    modules:
      - name: pg_autoscaler
        enabled: true
-  # Rotates daemon keys to aes256k for CVE-2025-30156 (#527). Opt-in; disabled
-  # by default. Increment keyGeneration to rotate again.
-  #
-  # Daemon keys only — aes256k over the CSI mount path needs kernel 7.0+ and
-  # these nodes run 6.18, so CSI stays on aes and the client-key checks stay
-  # raised. keyType is unset so Rook picks it.
+  # CephX keys on aes256k for CVE-2025-30156 (#527). Increment keyGeneration
+  # to rotate again. CSI kernel mounts need aes256k in libceph: Talos
+  # >= v1.13.10 backports it to 6.18 (siderolabs/pkgs 2da2cdc).
  security:
    cephx:
+      # auth_allowed_ciphers. Every key must be aes256k first, or the cluster
+      # locks itself out; recovery is in Rook's cephx-key-rotation doc.
+      allowedCiphers:
+        - aes256k
      daemon:
        keyRotationPolicy: KeyGeneration
        keyGeneration: 2
+      csi:
+        keyRotationPolicy: KeyGeneration
+        keyGeneration: 2
+        keyType: aes256k
+        # 0 deletes prior keys. Set 1 before the next rotation so PVCs
+        # mounted with the old key keep working until they are remounted.
+        keepPriorKeyCountMax: 0
+      rbdMirrorPeer:
+        keyRotationPolicy: KeyGeneration
+        keyGeneration: 2
+        keyType: aes256k
  # CephFS cloner pool raised from the default 4. Velero clones a
  # VolumeSnapshot per PVC, and 4 threads stall a multi-PVC backup with
  # `rados: "all cloner threads are busy"`.
@@ -24,7 +24,7 @@ machine:
  # The Kubernetes version is a separate decision, moved by
  # `talosctl upgrade-k8s`.
  install:
-    image: factory.talos.dev/metal-installer/0f6fb8bd8bb53013e9743b7c7b099d96fb9c32fd2464ced32693c2dc0ea5feb9:v1.13.9
+    image: factory.talos.dev/metal-installer/0f6fb8bd8bb53013e9743b7c7b099d96fb9c32fd2464ced32693c2dc0ea5feb9:v1.13.10
  features:
    rbac: true
    # stableHostname is NOT here: it is the v1alpha1 spelling of what the
@@ -57,7 +57,19 @@ machine:
    extraConfig:
      imageGCHighThresholdPercent: 70
      imageGCLowThresholdPercent: 60
+      # Default 110 leaves no N-1 headroom: with ~207 pods on three nodes, a
+      # rolling drain puts the survivors at 105/110 and KubeletTooManyPods
+      # fires (#937). The ceiling is IP space, not the kubelet: Cilium runs
+      # ipam.mode=kubernetes on the node's /24 podCIDR (~253 usable after
+      # cilium_host), so 200 keeps a margin for pods still terminating.
+      maxPods: 200
 cluster:
+  # etcd peers on node0, VLAN 46 (#576). Pinned: Talos never updates an
+  # existing member's peer URLs, so changing this subnet means leave + rejoin
+  # per member (talosctl etcd leave, reboot).
+  etcd:
+    advertisedSubnets:
+      - 10.2.46.0/24
  # controlPlane.endpoint is NOT here: `talosctl gen config` takes it as its
  # second argument, and bootstrap.sh already passes https://k8s.renner.uno:6443
  # there. Setting it again in this patch restated the same value in a second
@@ -17,6 +17,8 @@ machine:
    # pod0 is a closed L2 fabric between the three nodes: statically
    # addressed, no gateway, nothing to learn from an RA.
    net.ipv6.conf.pod0.accept_ra: "0"
+    # node0 is closed the same way.
+    net.ipv6.conf.node0.accept_ra: "0"
    # ingress0 keeps RAs, at the default accept_ra of 2: an IPv6 default
    # router is a link-local address learned only from the RA, and with
    # net.ipv6.conf.all.forwarding=1 — which Cilium sets — Linux ignores RAs
@@ -122,8 +124,8 @@ kind: LinkConfig
 name: uplink
 up: true
 ---
-# Management, VLAN 40. Carries the node IP, etcd, the API and Ceph's public
-# network.
+# Management, VLAN 40: world↔node. The API VIP, admin access to the nodes, and
+# node egress (DNS, NTP, image pulls). Node↔node traffic is on node0 (#576).
 apiVersion: v1alpha1
 kind: VLANConfig
 name: mgmt0
@@ -177,6 +179,19 @@ addresses:
  - address: 10.2.48.11/24
  - address: fd02:48::11/64
 ---
+# Node network, VLAN 46: traffic among the three nodes themselves (#576) — the
+# node IP, etcd, apiserver↔kubelet. Closed like pod0: statically addressed, no
+# gateway, never routed via the UDM.
+apiVersion: v1alpha1
+kind: VLANConfig
+name: node0
+vlanID: 46
+parent: uplink
+up: true
+addresses:
+  - address: fd02:46::11/64
+  - address: 10.2.46.11/24
+---
 # Ingress network, VLAN 47. World-to-pod: the BGP session, the next hop for
 # every advertised LoadBalancer VIP, and the masquerade source for pod egress.
 # Publicly routed, which is what separates it from pod0.
@@ -16,6 +16,7 @@ machine:
    # ingress0 keeps RAs: it learns its IPv6 router and its address from them,
    # and only the default route is held back — see k8s-1.yaml.
    net.ipv6.conf.pod0.accept_ra: "0"
+    net.ipv6.conf.node0.accept_ra: "0"
    net.ipv6.conf.ingress0.accept_ra_defrtr: "0"
    # Answer ARP only on the link that owns the address — see k8s-1.yaml (#793).
    net.ipv4.conf.all.arp_ignore: "1"
@@ -82,7 +83,7 @@ kind: LinkConfig
 name: uplink
 up: true
 ---
-# Management, VLAN 40. Node IP, etcd, the API and Ceph's public network.
+# Management, VLAN 40: world↔node — see k8s-1.yaml.
 apiVersion: v1alpha1
 kind: VLANConfig
 name: mgmt0
@@ -122,6 +123,17 @@ addresses:
  - address: 10.2.48.12/24
  - address: fd02:48::12/64
 ---
+# Node network, VLAN 46 — see k8s-1.yaml.
+apiVersion: v1alpha1
+kind: VLANConfig
+name: node0
+vlanID: 46
+parent: uplink
+up: true
+addresses:
+  - address: fd02:46::12/64
+  - address: 10.2.46.12/24
+---
 # Ingress network, VLAN 47. World-to-pod, publicly routed, statically
 # addressed — see k8s-1.yaml for why it is not leased.
 apiVersion: v1alpha1
@@ -12,6 +12,7 @@ machine:
    # ingress0 keeps RAs: it learns its IPv6 router and its address from them,
    # and only the default route is held back — see k8s-1.yaml.
    net.ipv6.conf.pod0.accept_ra: "0"
+    net.ipv6.conf.node0.accept_ra: "0"
    net.ipv6.conf.ingress0.accept_ra_defrtr: "0"
    # Answer ARP only on the link that owns the address — see k8s-1.yaml (#793).
    net.ipv4.conf.all.arp_ignore: "1"
@@ -109,6 +110,17 @@ addresses:
  - address: 10.2.48.13/24
  - address: fd02:48::13/64
 ---
+# Node network, VLAN 46 — see k8s-1.yaml.
+apiVersion: v1alpha1
+kind: VLANConfig
+name: node0
+vlanID: 46
+parent: uplink
+up: true
+addresses:
+  - address: fd02:46::13/64
+  - address: 10.2.46.13/24
+---
 # Ingress network, VLAN 47. World-to-pod, publicly routed, statically
 # addressed — see k8s-1.yaml for why it is not leased.
 apiVersion: v1alpha1
@@ -7,19 +7,15 @@
 # part of the config that could not be true anywhere but here.
 machine:
  kubelet:
-    # Constrain kubelet's auto-picked node IP to the mgmt LAN. Without this,
-    # kubelet picks the first non-loopback IP, which could land on the v30iot
-    # NIC (ens19, 10.2.30.0/24). Control-plane traffic and hostNetwork pod IPs
-    # must stay on mgmt LAN.
-    # fd40::/64 is the mgmt LAN ULA, and it is listed so the Node object carries
-    # an IPv6 InternalIP at all. Without one, Cilium falls back to detecting the
-    # node's IPv6 from mgmt0 and picks the SLAAC address from the dynamic
-    # provider prefix, which it then probes until the agent restarts (#789).
-    # The GUA is not a control-plane address: BGP peers on fd40::/64 too.
+    # The node IP lives on node0, VLAN 46: the closed node↔node network (#576).
+    # Both families are listed so the Node object carries an IPv6 InternalIP at
+    # all; without one Cilium detects the node's IPv6 itself and picks a SLAAC
+    # address from the dynamic provider prefix (#789). Talos picks one address
+    # per family, IPv4 first, whatever the order here.
    nodeIP:
      validSubnets:
-        - 10.2.40.0/24
-        - fd40::/64
+        - fd02:46::/64
+        - 10.2.46.0/24
  # The UDM on the ingress segment. Published here rather than as a route
  # because bird owns table 47 and originates the default from it; this is the
  # one place the address is stated.