feat(talos): move the node IP and etcd to pod0 (#576 )

Pod east-west has to leave the management VLAN, and the node IP is the only thing that controls where it goes. Three tests established that, rather than argument: - --ipv4-node / --ipv6-node, applied via CiliumNodeConfig: the agent logged receiving both and still published its mgmt0 address. - directRoutingDevice=pod0: the agent logged "Direct routing device detected direct-routing-device=pod0" and still installed 10.244.2.0/24 via 10.2.40.138 dev mgmt0. - Both routing modes, both address families, always keyed to the peer's Kubernetes InternalIP. etcd is pinned to the same segment. Left unset it advertises "the first routable address of the node", which is not deterministic once the node is re-addressed, so it has to be stated either way. pod0 is a closed 10G L2 between the three nodes; the management VLAN is about to become a single 1G NIC shared with everything else management does, which is the wrong link for quorum traffic. listenSubnets defaults to this, so peer and client listeners move together. This does move the etcd peer URLs off 10.2.40.x, so unlike the node IP it is not a no-op: one node at a time, waiting for a healthy quorum between each. The API VIP stays on mgmt0; it is a Layer2VIPConfig and does not follow the node IP. Apply one node at a time. Each node re-IPs, so per the note in k8s-1.yaml (#635) kube-apiserver keeps advertising the old address until its static pod restarts: after each node, check that `kubectl get endpoints kubernetes -n default` agrees with `kubectl get nodes -o wide` and reboot the node if it does not. The cilium agent reads the node IP only at startup, so it needs recreating on each node as well. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
2026-09-08 16:54:57 +02:00
66 changed files with 282 additions and 2195 deletions
@@ -376,61 +376,3 @@ steps:
          --dockerfile Containerfile \
          --destination gitea.renner.uno/gitea_admin/netbird-proxy-services:$TAG \
          --destination gitea.renner.uno/gitea_admin/netbird-proxy-services:latest
-
-  # wiki-index: embedding index over the Notes repo's wiki/, fed from Gitea and
-  # queried by ~/Notes/bin/wiki_rank.py. Runtime artifact like netbird-proxy-services.
-  - name: build-wiki-index
-    image: registry.gitlab.com/gitlab-ci-utils/container-images/kaniko:v1.25.18-debug
-    directory: ci/images/wiki-index
-    depends_on: []
-    when:
-      - event: deployment
-      - event: push
-        path: ["ci/images/wiki-index/**", ".woodpecker/build-ci-images.yml"]
-    environment:
-      GITEA_USER: gitea_admin
-      GITEA_TOKEN:
-        from_secret: ci_images_token
-    commands:
-      - mkdir -p /kaniko/.docker
-      - AUTH=$(printf '%s:%s' "$GITEA_USER" "$GITEA_TOKEN" | base64 | tr -d '\n')
-      - printf '{"auths":{"gitea.renner.uno":{"auth":"%s"}}}' "$AUTH" > /kaniko/.docker/config.json
-      - TAG=$(date +%Y-%m-%d)
-      # `build.<unix-ts>` is the tag Flux image automation tracks
-      # (infrastructure/image-automation/wiki-index.yaml): unique per build, so
-      # every push rolls out without a hand-pinned digest. The dated tag is
-      # rewritten several times a day and is kept for humans only.
-      - BUILD=build.$(date +%s)
-      - |
-        /kaniko/executor \
-          --context . \
-          --dockerfile Containerfile \
-          --destination gitea.renner.uno/gitea_admin/wiki-index:$TAG \
-          --destination gitea.renner.uno/gitea_admin/wiki-index:$BUILD \
-          --destination gitea.renner.uno/gitea_admin/wiki-index:latest
-
-  # bird: BGP daemon for the pod-network mesh over pod0 (#576). Runtime
-  # artifact like heating-can; same registry so the audit tracks it.
-  - name: build-bird
-    image: registry.gitlab.com/gitlab-ci-utils/container-images/kaniko:v1.25.18-debug
-    directory: ci/images/bird
-    depends_on: []
-    when:
-      - event: deployment
-      - event: push
-        path: ["ci/images/bird/**", ".woodpecker/build-ci-images.yml"]
-    environment:
-      GITEA_USER: gitea_admin
-      GITEA_TOKEN:
-        from_secret: ci_images_token
-    commands:
-      - mkdir -p /kaniko/.docker
-      - AUTH=$(printf '%s:%s' "$GITEA_USER" "$GITEA_TOKEN" | base64 | tr -d '\n')
-      - printf '{"auths":{"gitea.renner.uno":{"auth":"%s"}}}' "$AUTH" > /kaniko/.docker/config.json
-      - TAG=$(date +%Y-%m-%d)
-      - |
-        /kaniko/executor \
-          --context . \
-          --dockerfile Containerfile \
-          --destination gitea.renner.uno/gitea_admin/bird:$TAG \
-          --destination gitea.renner.uno/gitea_admin/bird:latest
@@ -26,16 +26,8 @@ edge_open3e_poll_seconds: 30
 # entries when added), so the multi-frame read is cheap at a 30 s poll.
 # Records faults/info events only (NormalOperation, LagDeviceDtcReported,
 # LegionellaProtectionTemperatureReached) — NOT setpoint or program changes.
-# onebase.*: the 29 paired ZigBee slots on the One Base gateway (ECU 0x6c3) --
-# 8 radiator valves, 9 climate sensors, 9 underfloor actuators, 3 floor
-# thermostats. Per-device room temperature, active setpoint, humidity, valve
-# opening/demand, battery and RSSI; the local replacement for the ViCare cloud
-# room sensors. Needs the onebase overlay (files/open3e-onebase.py) -- the stock
-# codec declares 57 bytes and a paired slot answers with 68.
-# Only CurrentValues are polled. The room and device registries (1884.., 2084..)
-# are static configuration and are read on demand, not every 30 s.
 edge_open3e_datapoints: >-
-  vitocal.259,vitocal.260,vitocal.268,vitocal.269,vitocal.271,vitocal.274,vitocal.286,vitocal.288,vitocal.318,vitocal.321,vitocal.324,vitocal.335,vitocal.336,vitocal.396,vitocal.402,vitocal.403,vitocal.476,vitocal.477,vitocal.543,vitocal.548,vitocal.565,vitocal.1007,vitocal.1043,vitocal.1190,vitocal.1211,vitocal.1391,vitocal.1416,vitocal.1417,vitocal.2320,vitocal.2346,vitocal.2350,vitocal.2351,vitocal.2352,vitocal.2369,vitocal.2370,vitocal.2486,vitocal.2487,vitocal.2488,vitocal.2543,vitocal.2560,vitocal.2622,vitocal.2624,vitocal.2625,vitocal.2735,vitocal.3016,vx3.266,vx3.378,vx3.379,vx3.380,vx3.535,vx3.1603,vx3.1664,vx3.1690,vx3.1801,vx3.1802,vx3.1810,vx3.1828,vx3.1830,vx3.1577,vx3.1578,vx3.1579,vx3.1580,vx3.1581,vx3.1582,vx3.1834,vx3.1836,vx3.2214,vx3.2240,vx3.3085,vx3.3086,vx3.3087,vx3.3088,vx3.3089,vx3.3090,vcmu.320,vcmu.322,vcmu.325,vcmu.355,vcmu.389,vcmu.391,vcmu.1772,vcmu.1773,vcmu.1775,vcmu.1776,vcmu.2256,vcmu.2259,vcmu.2333,vcmu.2334,vcmu.2569,vcmu.2571,vitocal.2791,onebase.2086,onebase.2089,onebase.2092,onebase.2095,onebase.2098,onebase.2101,onebase.2104,onebase.2107,onebase.2110,onebase.2113,onebase.2116,onebase.2119,onebase.2122,onebase.2125,onebase.2128,onebase.2131,onebase.2134,onebase.2137,onebase.2140,onebase.2143,onebase.2262,onebase.2265,onebase.2268,onebase.2271,onebase.2274,onebase.2277,onebase.2280,onebase.2283,onebase.2286
+  vitocal.259,vitocal.260,vitocal.268,vitocal.269,vitocal.271,vitocal.274,vitocal.286,vitocal.288,vitocal.318,vitocal.321,vitocal.324,vitocal.335,vitocal.336,vitocal.396,vitocal.402,vitocal.403,vitocal.476,vitocal.477,vitocal.543,vitocal.548,vitocal.565,vitocal.1007,vitocal.1043,vitocal.1190,vitocal.1211,vitocal.1391,vitocal.1416,vitocal.1417,vitocal.2320,vitocal.2346,vitocal.2350,vitocal.2351,vitocal.2352,vitocal.2369,vitocal.2370,vitocal.2486,vitocal.2487,vitocal.2488,vitocal.2543,vitocal.2560,vitocal.2622,vitocal.2624,vitocal.2625,vitocal.2735,vitocal.3016,vx3.266,vx3.378,vx3.379,vx3.380,vx3.535,vx3.1603,vx3.1664,vx3.1690,vx3.1801,vx3.1802,vx3.1810,vx3.1828,vx3.1830,vx3.1577,vx3.1578,vx3.1579,vx3.1580,vx3.1581,vx3.1582,vx3.1834,vx3.1836,vx3.2214,vx3.2240,vx3.3085,vx3.3086,vx3.3087,vx3.3088,vx3.3089,vx3.3090,vcmu.320,vcmu.322,vcmu.325,vcmu.355,vcmu.389,vcmu.391,vcmu.1772,vcmu.1773,vcmu.1775,vcmu.1776,vcmu.2256,vcmu.2259,vcmu.2333,vcmu.2334,vcmu.2569,vcmu.2571,vitocal.2791
 # ECU addresses; -dev selects codec set, tx selects the UDS address.
 # vcmu (refrigerant-circuit controller) has no dedicated codec set —
 # omitting dpList makes open3e fall back to the generic datapoint list,
@@ -46,9 +38,6 @@ edge_open3e_devices:
  # files/open3e-vx3plus.py) — die Stock-vx3-Liste filtert sie sonst weg.
  vx3: { tx: "0x6a1", dpList: "/config/vx3plus.py" }
  vcmu: { tx: "0x68c" }
-  # One Base ZigBee gateway. Overlay re-declares the paired ZigBee slots at
-  # their real 68-byte length (see files/open3e-onebase.py).
-  onebase: { tx: "0x6c3", dpList: "/config/onebase.py" }

 edge_e3oncan_gap_seconds: 10
 edge_e3oncan_mqtt_format: "{device}_{didNumber:04d}_{didName}"
@@ -1,54 +0,0 @@
-# Managed by ansible (roles/heating-can). Datapoint overlay for the One Base
-# ZigBee gateway on ECU 0x6c3 (BACKENDGATEWAY).
-#
-# Upstream declares ZigBee<N>DeviceCurrentValues as 57 bytes. A PAIRED device
-# slot on this gateway answers with 68: the 57 known bytes plus the parent
-# device serial and three trailing bytes. udsoncan reads only the declared
-# length and then treats the next two bytes -- the first half of that serial --
-# as a foreign data identifier, so every paired slot fails to decode with
-# "Server returned values for data identifier 0x.... that was not requested".
-# The bogus identifier differs per device because the serial does.
-#
-# This overlay re-declares the 29 paired slots at 68 bytes, reusing upstream's
-# sub-type list so the per-device-type O3ESwitch (climate sensor / TRV / floor
-# thermostat / actuator, incl. DeviceTemperatureSetpoint) is inherited rather
-# than copied.
-#
-# EMPTY slots answer with 57 and decode against the stock codec, so they are
-# left alone and are not polled. A slot that is later unpaired therefore starts
-# reporting a decode error on its topic instead of a value.
-#
-# Verified on ECU 0x6c3, 2026-09-17: 29 paired slots, all 68 bytes.
-from open3e.Open3Ecodecs import O3EComplexType, RawCodec
-from open3e.Open3Edatapoints import dataIdentifiers as _master
-
-# ZigBee<N>DeviceCurrentValues of the paired slots: devices 1-20 run 2086..2143
-# in steps of 3, devices 21-29 continue at 2262..2286.
-_PAIRED_CURRENT_VALUES = [2086 + 3 * n for n in range(20)] + [
-    2262 + 3 * n for n in range(9)
-]
-
-
-def _with_parent_serial(did):
-    stock = _master["dids"][did]
-    return O3EComplexType(
-        68,
-        stock.id,
-        list(stock.subTypes)
-        + [
-            RawCodec(8, "ParentDeviceSerialnumber"),
-            RawCodec(2, "Unknown_65_66"),
-            RawCodec(1, "Unknown_67"),
-        ],
-        desc=stock.desc,
-        info=stock.info,
-        acc=stock.acc,
-    )
-
-
-dataIdentifiers = {
-    "dids": {
-        **_master["dids"],
-        **{did: _with_parent_serial(did) for did in _PAIRED_CURRENT_VALUES},
-    }
-}
@@ -129,13 +129,6 @@
    mode: "0644"
  notify: Restart heating-can

- name: Deploy open3e One Base datapoint overlay (68-byte ZigBee slots)
-  ansible.builtin.copy:
-    src: open3e-onebase.py
-    dest: "{{ heating_can_base_dir }}/onebase.py"
-    mode: "0644"
-  notify: Restart heating-can
-
 # --- Compose project ---------------------------------------------------------
 - name: Deploy the broker credential env file
  ansible.builtin.template:
@@ -17,7 +17,6 @@ services:
    volumes:
      - ./devices.json:/config/devices.json:ro
      - ./vx3plus.py:/config/vx3plus.py:ro
-      - ./onebase.py:/config/onebase.py:ro
    env_file:
      # Broker credential, 0600 -- see heating-can.env.j2.
      - heating-can.env
@@ -1,22 +0,0 @@
-FROM alpine:3.22@sha256:14358309a308569c32bdc37e2e0e9694be33a9d99e68afb0f5ff33cc1f695dce
-
-# bird: BGP daemon for the pod-network mesh over pod0. Distro package, no build
-# step — the image is the binary plus its runtime.
-#
-# 3.22 rather than the 3.21 the other images here pin: bird2 first appears in
-# 3.22/community.
-#
-# Config is read from /etc/bird/bird.conf, not the package default
-# /etc/bird.conf, so the mount cannot shadow what the package ships.
-#
-# kubectl renders the init container's per-node config: go-template-file for
-# the templating, a field selector for excluding the node itself, in-cluster
-# config for the auth. From the distro, so there is no version literal to drift
-# and no download to trust. It is only ever used to read core/v1 Nodes.
-RUN apk add --no-cache bird2 kubectl
-
-# Fails the build rather than the first pod if a binary or its path is wrong.
-RUN /usr/sbin/bird --version \
- && kubectl version --client > /dev/null
-
-ENTRYPOINT ["/usr/sbin/bird", "-f", "-c", "/etc/bird/bird.conf"]
@@ -1,10 +0,0 @@
-# wiki-index: embedding index over the Notes repo's wiki/, fed from Gitea,
-# queried by ~/Notes/bin/wiki_rank.py. Standard library only; runs unprivileged.
-FROM python:3.13-slim
-
-RUN useradd --system --no-create-home --shell /usr/sbin/nologin wikiindex
-COPY wiki_index.py /opt/wiki_index.py
-
-ENV PYTHONUNBUFFERED=1
-USER wikiindex
-ENTRYPOINT ["python3", "/opt/wiki_index.py"]
@@ -1,178 +0,0 @@
-#!/usr/bin/env python3
-"""Unit tests for wiki_index. Run: python3 -m unittest test_wiki_index"""
-
-import hashlib
-import hmac
-import json
-import tempfile
-import unittest
-from pathlib import Path
-
-import wiki_index as wi
-
-NOTE = """---
-title: 'UDM intercepts DNS'
-type: wiki
-tags: [dns, unifi]
-aliases: [udm dns interception, dig 1.1.1.1 wrong]
---
-
-# UDM intercepts DNS
-
-The UDM answers every UDP/53 query itself.
-
-## Details
-
-More.
-"""
-
-
-class ParseTests(unittest.TestCase):
-    def test_note_text(self):
-        title, text = wi.note_text("udm-dns", NOTE)
-        self.assertEqual(title, "UDM intercepts DNS")
-        self.assertIn("udm dns UDM intercepts DNS udm dns interception", text)
-        self.assertIn("dns unifi", text)
-        self.assertTrue(text.endswith("The UDM answers every UDP/53 query itself."))
-        self.assertLessEqual(len(text), wi.EMBED_CHARS)
-
-    def test_credentials_are_stripped(self):
-        original = wi.http
-        wi.http = lambda method, url, body=None, headers=None, timeout=60.0: (
-            {"auth": {"client_token": "t"}} if url.endswith("/login")
-            else {"data": {"data": {"gitea_token": "abc\n", "webhook_secret": " s3cret\n"}}})
-        original_sa = wi.SA_TOKEN
-        try:
-            with tempfile.TemporaryDirectory() as d:
-                wi.SA_TOKEN = Path(d) / "token"
-                wi.SA_TOKEN.write_text("jwt\n")
-                self.assertEqual(wi.Credentials().fetch(), {"gitea_token": "abc", "webhook_secret": "s3cret"})
-        finally:
-            wi.http, wi.SA_TOKEN = original, original_sa
-
-    def test_tree_paging_stops_on_short_last_page(self):
-        # Gitea: truncated=true on every page including the last, "tree": null past it.
-        pages = {
-            1: {"tree": [{"path": f"wiki/n{i}.md", "type": "blob", "sha": f"s{i}"} for i in range(1000)], "truncated": True, "total_count": 1192},
-            2: {"tree": [{"path": f"wiki/n{i}.md", "type": "blob", "sha": f"s{i}"} for i in range(1000, 1192)] + [{"path": "raw/x.md", "type": "blob", "sha": "r"}, {"path": "wiki/sub/y.md", "type": "blob", "sha": "y"}], "truncated": True, "total_count": 1192},
-            3: {"tree": None, "truncated": False, "total_count": 1192},
-        }
-        calls = []
-        original = wi.http
-
-        def fake_http(method, url, body=None, headers=None, timeout=60.0):
-            page = int(url.rsplit("page=", 1)[1])
-            calls.append(page)
-            return pages[page]
-
-        wi.http = fake_http
-        try:
-            tree = wi.Gitea("t").tree("abc")
-        finally:
-            wi.http = original
-        self.assertEqual(calls, [1, 2])
-        self.assertEqual(len(tree), 1192)
-        self.assertNotIn("raw/x.md", tree)
-        self.assertNotIn("wiki/sub/y.md", tree)
-        self.assertEqual(tree["wiki/n1191.md"], "s1191")
-
-    def test_signature(self):
-        body = b'{"ref":"refs/heads/master"}'
-        sig = hmac.new(b"s3cret", body, hashlib.sha256).hexdigest()
-        self.assertTrue(wi.verify_signature("s3cret", body, sig))
-        self.assertFalse(wi.verify_signature("other", body, sig))
-        self.assertFalse(wi.verify_signature("s3cret", body, ""))
-
-
-class FakeGitea:
-    def __init__(self, head, tree, files):
-        self._head, self._tree, self.files = head, tree, files
-        self.raw_calls = []
-
-    def head(self):
-        return self._head
-
-    def tree(self, commit):
-        return self._tree
-
-    def raw(self, path, ref):
-        self.raw_calls.append(path)
-        return self.files[path]
-
-
-class IndexTests(unittest.TestCase):
-    def setUp(self):
-        self.tmp = tempfile.TemporaryDirectory()
-        self.path = Path(self.tmp.name) / "index.json"
-        self.original_embed = wi.embed
-        self.embed_calls = []
-
-        def fake_embed(texts, batch=2):
-            self.embed_calls.append(list(texts))
-            return [wi.unit([1.0, float(len(t) % 7)]) for t in texts]
-
-        wi.embed = fake_embed
-
-    def tearDown(self):
-        wi.embed = self.original_embed
-        self.tmp.cleanup()
-
-    def test_sync_is_incremental_and_persists(self):
-        files = {"wiki/udm-dns.md": NOTE, "wiki/ceph.md": "---\ntitle: Ceph\n---\n# Ceph\n\nOSD.\n", "wiki/_index.md": "---\ntitle: Wiki\n---\n"}
-        tree = {"wiki/udm-dns.md": "aaa", "wiki/ceph.md": "bbb", "wiki/_index.md": "ccc"}
-        g = FakeGitea("c1", tree, files)
-        idx = wi.Index(self.path)
-        s = idx.sync(g)
-        self.assertEqual((s["changed"], s["removed"], s["notes"]), (2, 0, 2))
-        self.assertEqual(set(idx.notes), {"udm-dns", "ceph"})
-        self.assertEqual(idx.notes["udm-dns"]["title"], "UDM intercepts DNS")
-        self.assertTrue(self.path.exists())
-
-        # head unchanged: nothing fetched
-        s = idx.sync(g)
-        self.assertEqual(s.get("skipped"), "head unchanged")
-        self.assertEqual(len(g.raw_calls), 2)
-
-        # one note changed, one removed, new head
-        g2 = FakeGitea("c2", {"wiki/udm-dns.md": "aaa2"}, {"wiki/udm-dns.md": NOTE.replace("itself", "itself!")})
-        s = idx.sync(g2)
-        self.assertEqual((s["changed"], s["removed"], s["notes"]), (1, 1, 1))
-        self.assertEqual(g2.raw_calls, ["wiki/udm-dns.md"])
-
-        # restart loads the persisted index
-        idx2 = wi.Index(self.path)
-        self.assertEqual(idx2.commit, "c2")
-        self.assertEqual(set(idx2.notes), {"udm-dns"})
-
-    def test_rank_orders_by_cosine(self):
-        idx = wi.Index(self.path)
-        idx.notes = {"a": {"sha": "1", "title": "A", "vector": wi.unit([1.0, 0.0])},
-                     "b": {"sha": "2", "title": "B", "vector": wi.unit([0.0, 1.0])}}
-        wi.embed = lambda texts, batch=2: [wi.unit([0.2, 1.0])]
-        rows = idx.rank("q", k=5)
-        self.assertEqual([r["slug"] for r in rows], ["b", "a"])
-        self.assertGreater(rows[0]["score"], rows[1]["score"])
-
-    def test_status_reports_age_and_errors(self):
-        idx = wi.Index(self.path)
-        self.assertFalse(idx.status()["ready"])
-        g = FakeGitea("c1", {"wiki/a.md": "x"}, {"wiki/a.md": "---\ntitle: A\n---\n# A\n\nt\n"})
-        idx.sync(g)
-        st = idx.status()
-        self.assertTrue(st["ready"])
-        self.assertEqual(st["notes"], 1)
-        self.assertEqual(st["commit"], "c1")
-
-        class Broken(FakeGitea):
-            def head(self):
-                raise wi.ApiError("gitea down")
-
-        with self.assertRaises(wi.ApiError):
-            idx.sync(Broken("c1", {}, {}))
-        self.assertIn("gitea down", idx.status()["last_error"])
-        self.assertEqual(idx.status()["notes"], 1)  # last good index kept
-        self.assertTrue(idx.status()["ready"])  # and still served
-
-
-if __name__ == "__main__":
-    unittest.main()
@@ -1,408 +0,0 @@
-#!/usr/bin/env python3
-"""wiki-index: embedding index over the Notes repo's wiki/, fed from Gitea.
-
-Source of truth is the notes repository on Gitea. The index holds, per wiki
-note, the git blob sha, title and a bge-m3 vector of slug/title/aliases/tags/
-first paragraph. A sync fetches the repo tree, embeds every note whose blob sha
-changed (via wiki-embed, in-cluster) and drops notes that left the tree; it
-runs at start, on every push webhook from Gitea, and every RESYNC_SECONDS.
-
-HTTP:
-  POST /rank     {"query": str, "k": int}   -> {"rows": [{"slug","title","score"}], "index": {...}}
-  GET  /health   -> 200 always (liveness) with {"ready", "notes", "commit", "synced_at", "age_s", "syncing", "last_error"}
-  GET  /ready    -> 200 when an index exists, else 503 (readiness)
-  POST /webhook  Gitea push event, HMAC-SHA256 in X-Gitea-Signature -> {"queued": true}
-
-Credentials (Gitea token, webhook secret) are read from OpenBao through the
-Kubernetes auth role on every use; they never land in a Secret or a log line.
-Standard library only.
-"""
-
-from __future__ import annotations
-
-import hashlib
-import hmac
-import json
-import logging
-import os
-import re
-import sys
-import threading
-import time
-import urllib.error
-import urllib.parse
-import urllib.request
-from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
-from pathlib import Path
-
-LOG = logging.getLogger("wiki-index")
-
-GITEA_URL = os.environ.get("GITEA_URL", "https://gitea.renner.uno").rstrip("/")
-NOTES_REPO = os.environ.get("NOTES_REPO", "k.renner/notes")
-NOTES_BRANCH = os.environ.get("NOTES_BRANCH", "master")
-WIKI_DIR = os.environ.get("WIKI_DIR", "wiki").strip("/")
-EMBED_URL = os.environ.get("EMBED_URL", "http://wiki-embed.ai.svc.cluster.local:8080").rstrip("/")
-EMBED_MODEL = os.environ.get("EMBED_MODEL", "bge-m3")
-EMBED_CHARS = int(os.environ.get("EMBED_CHARS", "1500"))
-INDEX_PATH = Path(os.environ.get("INDEX_PATH", "/data/index.json"))
-RESYNC_SECONDS = int(os.environ.get("RESYNC_SECONDS", "1800"))
-LISTEN_PORT = int(os.environ.get("LISTEN_PORT", "8080"))
-BAO_ADDR = os.environ.get("BAO_ADDR", "http://openbao.openbao.svc.cluster.local:8200").rstrip("/")
-BAO_ROLE = os.environ.get("BAO_ROLE", "wiki-index")
-BAO_SECRET_PATH = os.environ.get("BAO_SECRET_PATH", "homelab/data/apps/ai/wiki-index")
-SA_TOKEN = Path(os.environ.get("SA_TOKEN", "/var/run/secrets/kubernetes.io/serviceaccount/token"))
-
-
-class ApiError(Exception):
-    pass
-
-
-# ---------------------------------------------------------------- http helpers
-
-
-def http(method: str, url: str, body=None, headers=None, timeout: float = 60.0):
-    data = None
-    hdrs = dict(headers or {})
-    if body is not None:
-        data = json.dumps(body).encode("utf-8")
-        hdrs.setdefault("Content-Type", "application/json")
-    req = urllib.request.Request(url, data=data, method=method, headers=hdrs)
-    try:
-        with urllib.request.urlopen(req, timeout=timeout) as resp:
-            raw = resp.read()
-            ctype = resp.headers.get("Content-Type", "")
-    except urllib.error.HTTPError as e:
-        raise ApiError(f"{method} {url}: HTTP {e.code}: {e.read().decode('utf-8', 'replace')[:300]}") from e
-    except urllib.error.URLError as e:
-        raise ApiError(f"{method} {url}: {e.reason}") from e
-    if "json" in ctype:
-        return json.loads(raw.decode("utf-8"))
-    return raw
-
-
-# ---------------------------------------------------------------- credentials
-
-
-class Credentials:
-    """Gitea token and webhook secret from OpenBao, fetched per use."""
-
-    def __init__(self, static: dict | None = None):
-        self.static = static  # tests / local runs
-
-    def fetch(self) -> dict:
-        if self.static is not None:
-            return self.static
-        jwt = SA_TOKEN.read_text(encoding="utf-8").strip()
-        login = http("POST", f"{BAO_ADDR}/v1/auth/kubernetes/login", {"role": BAO_ROLE, "jwt": jwt})
-        token = login["auth"]["client_token"]
-        kv = http("GET", f"{BAO_ADDR}/v1/{BAO_SECRET_PATH}", headers={"X-Vault-Token": token})
-        data = {k: str(v).strip() for k, v in kv["data"]["data"].items()}  # `bao kv put key=@file` keeps the file's newline
-        for key in ("gitea_token", "webhook_secret"):
-            if not data.get(key):
-                raise ApiError(f"OpenBao {BAO_SECRET_PATH} has no key {key}")
-        return data
-
-
-# ---------------------------------------------------------------- gitea
-
-
-class Gitea:
-    def __init__(self, token: str, base: str = GITEA_URL, repo: str = NOTES_REPO, branch: str = NOTES_BRANCH):
-        self.base, self.repo, self.branch = base, repo, branch
-        self.headers = {"Authorization": f"token {token}"}
-
-    def api(self, path: str, **params):
-        q = ("?" + urllib.parse.urlencode(params)) if params else ""
-        return http("GET", f"{self.base}/api/v1/repos/{self.repo}{path}{q}", headers=self.headers)
-
-    def head(self) -> str:
-        return self.api(f"/branches/{urllib.parse.quote(self.branch, safe='')}")["commit"]["id"]
-
-    def tree(self, commit: str) -> dict[str, str]:
-        """path -> blob sha for every wiki note at that commit."""
-        out: dict[str, str] = {}
-        page, per_page, seen = 1, 1000, 0
-        while True:
-            t = self.api(f"/git/trees/{commit}", recursive="true", per_page=per_page, page=page)
-            entries = t.get("tree") or []  # Gitea answers "tree": null past the last page
-            for e in entries:
-                p = e.get("path", "")
-                if e.get("type") == "blob" and p.startswith(WIKI_DIR + "/") and p.endswith(".md") and "/" not in p[len(WIKI_DIR) + 1:]:
-                    out[p] = e["sha"]
-            seen += len(entries)
-            total = int(t.get("total_count") or 0)
-            if not entries or len(entries) < per_page or (total and seen >= total):
-                break
-            page += 1
-            if page > 50:
-                raise ApiError("tree paging runaway")
-        return out
-
-    def raw(self, path: str, ref: str) -> str:
-        data = http("GET", f"{self.base}/api/v1/repos/{self.repo}/raw/{urllib.parse.quote(path)}?ref={urllib.parse.quote(ref, safe='')}",
-                    headers=self.headers)
-        return data.decode("utf-8") if isinstance(data, bytes) else json.dumps(data)
-
-
-# ---------------------------------------------------------------- notes
-
-
-def parse_frontmatter(text: str) -> tuple[dict, str]:
-    m = re.match(r"^---[ \t]*\n(.*?)\n---[ \t]*(?:\n|$)", text, re.DOTALL)
-    if not m:
-        return {}, text
-    fm: dict = {}
-    for line in m.group(1).splitlines():
-        mm = re.match(r"^([A-Za-z_][\w-]*):\s*(.*)$", line)
-        if mm:
-            fm[mm.group(1)] = mm.group(2).strip()
-    return fm, text[m.end():]
-
-
-def as_list(value: str) -> list[str]:
-    s = (value or "").strip()
-    if s.startswith("[") and s.endswith("]"):
-        s = s[1:-1]
-    return [p.strip().strip("'\"") for p in s.split(",") if p.strip()]
-
-
-def first_paragraph(body: str) -> str:
-    para: list[str] = []
-    seen_h1 = False
-    for line in body.splitlines():
-        if line.startswith("# "):
-            seen_h1 = True
-            continue
-        if not seen_h1:
-            continue
-        if line.strip() == "" or line.startswith("#"):
-            if para:
-                break
-            continue
-        para.append(line.strip())
-    return " ".join(para)
-
-
-def note_text(slug: str, text: str) -> tuple[str, str]:
-    """(title, embedding text) for one note file."""
-    fm, body = parse_frontmatter(text)
-    title = (fm.get("title") or slug).strip().strip("'\"")
-    parts = [slug.replace("-", " "), title, " ".join(as_list(fm.get("aliases", ""))),
-             " ".join(as_list(fm.get("tags", ""))), first_paragraph(body)]
-    return title, " ".join(parts)[:EMBED_CHARS]
-
-
-# ---------------------------------------------------------------- embeddings
-
-
-def embed(texts: list[str], batch: int = 2) -> list[list[float]]:
-    out: list[list[float]] = []
-    for i in range(0, len(texts), batch):
-        resp = http("POST", f"{EMBED_URL}/v1/embeddings", {"model": EMBED_MODEL, "input": texts[i:i + batch]}, timeout=120)
-        rows = sorted(resp["data"], key=lambda d: d["index"])
-        out.extend(unit(d["embedding"]) for d in rows)
-    return out
-
-
-def unit(v: list[float]) -> list[float]:
-    n = sum(x * x for x in v) ** 0.5 or 1.0
-    return [round(x / n, 5) for x in v]
-
-
-# ---------------------------------------------------------------- index
-
-
-class Index:
-    def __init__(self, path: Path):
-        self.path = path
-        self.lock = threading.Lock()
-        self.notes: dict[str, dict] = {}  # slug -> {sha, title, vector}
-        self.commit = ""
-        self.synced_at = 0.0
-        self.syncing = False
-        self.last_error = ""
-        if path.exists():
-            try:
-                d = json.loads(path.read_text(encoding="utf-8"))
-                self.notes, self.commit, self.synced_at = d.get("notes", {}), d.get("commit", ""), d.get("synced_at", 0.0)
-                LOG.info("loaded index: %d notes at %s", len(self.notes), self.commit[:8])
-            except (OSError, ValueError) as e:
-                LOG.warning("index unreadable (%s); starting empty", e)
-
-    def save(self) -> None:
-        self.path.parent.mkdir(parents=True, exist_ok=True)
-        tmp = self.path.with_suffix(".tmp")
-        tmp.write_text(json.dumps({"notes": self.notes, "commit": self.commit, "synced_at": self.synced_at}), encoding="utf-8")
-        tmp.replace(self.path)
-
-    def sync(self, gitea: Gitea) -> dict:
-        """Bring the index to the branch head; returns a summary."""
-        with self.lock:
-            self.syncing = True
-        try:
-            head = gitea.head()
-            if head == self.commit and self.notes:
-                summary = {"commit": head, "changed": 0, "removed": 0, "notes": len(self.notes), "skipped": "head unchanged"}
-                self.synced_at = time.time()
-                self.save()
-                return summary
-            tree = gitea.tree(head)
-            slugs = {p[len(WIKI_DIR) + 1:-3]: (p, sha) for p, sha in tree.items() if not p.endswith("/_index.md")}
-            todo = [s for s, (p, sha) in slugs.items() if self.notes.get(s, {}).get("sha") != sha]
-            removed = [s for s in list(self.notes) if s not in slugs]
-            LOG.info("sync %s: %d notes, %d to embed, %d removed", head[:8], len(slugs), len(todo), len(removed))
-            fresh: dict[str, dict] = {}
-            for i in range(0, len(todo), 20):
-                chunk = todo[i:i + 20]
-                texts, titles = [], {}
-                for s in chunk:
-                    title, text = note_text(s, gitea.raw(slugs[s][0], head))
-                    titles[s] = title
-                    texts.append(text)
-                for s, vec in zip(chunk, embed(texts)):
-                    fresh[s] = {"sha": slugs[s][1], "title": titles[s], "vector": vec}
-            with self.lock:
-                for s in removed:
-                    self.notes.pop(s, None)
-                self.notes.update(fresh)
-                self.commit = head
-                self.synced_at = time.time()
-                self.last_error = ""
-                self.save()
-            return {"commit": head, "changed": len(todo), "removed": len(removed), "notes": len(self.notes)}
-        except Exception as e:
-            with self.lock:
-                self.last_error = str(e)[:300]
-            raise
-        finally:
-            with self.lock:
-                self.syncing = False
-
-    def rank(self, query: str, k: int = 60) -> list[dict]:
-        q = embed([query])[0]
-        with self.lock:
-            rows = [(s, n["title"], sum(a * b for a, b in zip(q, n["vector"]))) for s, n in self.notes.items()]
-        rows.sort(key=lambda r: -r[2])
-        return [{"slug": s, "title": t, "score": round(sc, 4)} for s, t, sc in rows[:k]]
-
-    def status(self) -> dict:
-        """`ready` = an index exists and can answer /rank; a failed sync keeps the last good index."""
-        with self.lock:
-            return {"ready": bool(self.notes), "notes": len(self.notes), "commit": self.commit,
-                    "synced_at": self.synced_at, "age_s": round(time.time() - self.synced_at) if self.synced_at else None,
-                    "syncing": self.syncing, "last_error": self.last_error or None}
-
-
-# ---------------------------------------------------------------- syncer
-
-
-class Syncer(threading.Thread):
-    """One worker: syncs on demand (webhook) and every RESYNC_SECONDS."""
-
-    def __init__(self, index: Index, creds: Credentials, interval: int):
-        super().__init__(daemon=True, name="syncer")
-        self.index, self.creds, self.interval = index, creds, interval
-        self.wake = threading.Event()
-
-    def trigger(self) -> None:
-        self.wake.set()
-
-    RETRY_SECONDS = 60
-
-    def run(self) -> None:
-        while True:
-            try:
-                data = self.creds.fetch()
-                summary = self.index.sync(Gitea(data["gitea_token"]))
-                LOG.info("sync done: %s", summary)
-                wait = self.interval
-            except Exception as e:  # keep serving the last good index; retry soon, not next interval
-                LOG.error("sync failed: %s", e)
-                wait = self.RETRY_SECONDS
-            self.wake.wait(wait)
-            self.wake.clear()
-
-
-def verify_signature(secret: str, body: bytes, signature: str) -> bool:
-    digest = hmac.new(secret.encode("utf-8"), body, hashlib.sha256).hexdigest()
-    return hmac.compare_digest(digest, (signature or "").strip())
-
-
-# ---------------------------------------------------------------- server
-
-
-def make_handler(index: Index, syncer: Syncer, creds: Credentials):
-    class Handler(BaseHTTPRequestHandler):
-        def log_message(self, fmt, *args):
-            LOG.info("%s %s", self.address_string(), fmt % args)
-
-        def send_json(self, code: int, obj) -> None:
-            data = json.dumps(obj).encode("utf-8")
-            self.send_response(code)
-            self.send_header("Content-Type", "application/json")
-            self.send_header("Content-Length", str(len(data)))
-            self.end_headers()
-            self.wfile.write(data)
-
-        def body(self) -> bytes:
-            n = int(self.headers.get("Content-Length") or 0)
-            return self.rfile.read(n) if n else b""
-
-        def do_GET(self):
-            if self.path == "/health":  # liveness: the process serves
-                return self.send_json(200, index.status())
-            if self.path == "/ready":  # readiness: an index exists
-                st = index.status()
-                return self.send_json(200 if st["ready"] else 503, st)
-            self.send_json(404, {"error": "not found"})
-
-        def do_POST(self):
-            if self.path == "/rank":
-                try:
-                    req = json.loads(self.body() or b"{}")
-                    query = str(req.get("query", "")).strip()
-                    k = max(1, min(int(req.get("k", 60)), 500))
-                    if not query:
-                        return self.send_json(400, {"error": "query required"})
-                    rows = index.rank(query, k)
-                    return self.send_json(200, {"rows": rows, "index": index.status()})
-                except ApiError as e:
-                    return self.send_json(502, {"error": str(e)})
-                except (ValueError, TypeError) as e:
-                    return self.send_json(400, {"error": str(e)})
-            if self.path == "/webhook":
-                raw = self.body()
-                try:
-                    secret = creds.fetch()["webhook_secret"]
-                except ApiError as e:
-                    return self.send_json(503, {"error": str(e)})
-                if not verify_signature(secret, raw, self.headers.get("X-Gitea-Signature", "")):
-                    return self.send_json(403, {"error": "bad signature"})
-                try:
-                    event = json.loads(raw or b"{}")
-                except ValueError:
-                    return self.send_json(400, {"error": "bad json"})
-                ref = event.get("ref", "")
-                if ref and ref != f"refs/heads/{NOTES_BRANCH}":
-                    return self.send_json(200, {"queued": False, "ref": ref})
-                syncer.trigger()
-                return self.send_json(200, {"queued": True})
-            self.send_json(404, {"error": "not found"})
-
-    return Handler
-
-
-def main() -> int:
-    logging.basicConfig(level=logging.INFO, format="%(asctime)s %(levelname)s %(message)s", stream=sys.stdout)
-    index = Index(INDEX_PATH)
-    creds = Credentials()
-    syncer = Syncer(index, creds, RESYNC_SECONDS)
-    syncer.start()
-    server = ThreadingHTTPServer(("", LISTEN_PORT), make_handler(index, syncer, creds))
-    LOG.info("listening on :%d, repo %s@%s, embed %s", LISTEN_PORT, NOTES_REPO, NOTES_BRANCH, EMBED_URL)
-    server.serve_forever()
-    return 0
-
-
-if __name__ == "__main__":
-    sys.exit(main())
@@ -1,117 +0,0 @@
-# Traffic classes on the k8s nodes
-
-Issue #576. Three of the four classes are separated; management is not.
-
-## 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 |
-
-Hard constraint, set by the owner: inter-node and pod traffic never traverses
-the UDM. Everything above stays on the 10G switch.
-
-## Why policy routing rather than moving the node IP
-
-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.
-
-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.
-
-## Rule order
-
-Lower priority runs first.
-
-| 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 |
-
-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.
-
-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.
-
-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.
-
-## 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.
-
-## 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.
@@ -0,0 +1,194 @@
+# Proposal — separate the four traffic classes onto their own VLANs
+
+Status: draft. The question this started with is **answered** — from the Cilium
+source and flag reference, not from a test — and the answer changes the shape
+of the proposal rather than confirming it.
+
+Supersedes the framing in issue #576, which described this as "give node
+management its own NIC, and Ceph public its own VLAN". That was two of four
+classes and one of them turned out already solved.
+
+## The four classes
+
+| class | what it is | where it goes today |
+|---|---|---|
+| **Ceph replication** | OSD-to-OSD heartbeat, replication, recovery | VLAN 49, its own X710 port — **separated** |
+| **cluster-internal** | pod-to-pod east-west, including Ceph client I/O | pod network, VXLAN over `mgmt0` |
+| **cluster-external** | ingress to LoadBalancer VIPs, egress to LAN and internet | `mgmt0` |
+| **management** | Talos API, etcd, kube-apiserver, kubelet | `mgmt0` |
+
+Three of four share one interface. That is not a design decision — until the
+bare-metal conversion there was one NIC to work with.
+
+Ceph client I/O living on the pod network is correct rather than a leftover:
+that traffic *is* cluster-internal. Ceph is done.
+
+## Principle
+
+**Each class gets its own VLAN. Which physical interface a VLAN lands on is a
+separate, changeable decision.**
+
+Defining classes by VLAN means the separation survives recabling, port failures
+and capacity changes, and it does not require one port per class. Defining them
+by interface would freeze a capacity decision into the architecture.
+
+## The finding that shapes everything else
+
+The original open question was whether `nodePort.directRoutingDevice` moves the
+next-hop of Cilium's direct routes, or only selects an interface. It is
+answerable from the source, and the answer is: **only the interface.**
+
+`installDirectRoute` takes the next-hop as a parameter:
+
+```go
+func installDirectRoute(log *slog.Logger, prefix netip.Prefix, nodeIP net.IP,
+    skipUnreachable bool) (routeSpec *netlink.Route, err error)
+```
+
+and `nodeUpdate` supplies it as `newNode.GetNodeIP(false)`. `GetNodeIP` returns,
+in priority order, `NodeInternalIP` → `NodeExternalIP` → anything else. The
+same `newIP4` is what tunnel mode uses as the VXLAN endpoint.
+
+`NodeInternalIP` comes from the Kubernetes Node object. `--ipv4-node` looks like
+an override but is not one: `initFromConfig()` calls `SetNodeInternalIP` and
+`initFromK8s()` runs **after** it and overwrites the value. In a Kubernetes
+cluster there is no path by which Cilium's node IP differs from the kubelet's
+`InternalIP`.
+
+`--direct-routing-device` is documented as "Device name used to connect nodes in
+direct routing mode" — a device, not an address.
+
+**So cluster-internal traffic is welded to the node IP, and the node IP is the
+kubelet InternalIP.** Two consequences:
+
+- The class cannot be moved off whatever VLAN the node IP lives on.
+- **Routing mode is irrelevant to this problem.** Tunnel mode sources the VXLAN
+  from the node IP; native mode uses it as the direct-route next-hop. Native
+  routing buys nothing here, so the disruptive, non-rolling step this proposal
+  previously contained is simply gone.
+
+## Consequence: draw the line the other way
+
+If cluster-internal cannot leave the node IP, then separation has to come from
+moving *management services* off it — and each one has to be checked
+individually, because they do not share a mechanism.
+
+| management service | can it leave the node IP? | lever |
+|---|---|---|
+| etcd | **yes** | `cluster.etcd.advertisedSubnets` (+ `listenSubnets`, which defaults to it) |
+| kube-apiserver endpoint | **yes** | already a VIP — `Layer2VIPConfig` with a `link:` field |
+| Talos API | **n/a** | `apid` answers on every node address (measured); a client and firewall choice, not a bind setting |
+| kubelet | **no** | it *is* the node IP, by definition |
+
+So four VLANs are reachable, but the boundary does not fall where the class
+names suggest: **kubelet stays with cluster-internal.** That is the honest
+result, and it is worth stating plainly rather than filing kubelet under
+management and hoping.
+
+The remaining classes keep the levers established earlier:
+
+- **cluster-external ingress** — `CiliumBGPNodeConfigOverride.spec.bgpInstances[].peers[].localAddress`.
+  For eBGP the session source is the next-hop, so pinning the session pins
+  where return traffic arrives. The session to the UDM has been running for
+  months.
+- **cluster-external egress** — eBPF masquerading follows the output device, so
+  this is a default-route decision plus `enable-masquerade-to-route-source`
+  (today `false`).
+- **Ceph replication** — Multus macvlan on VLAN 49, done.
+
+## The general escape hatch
+
+Ceph is separated because its OSDs got a second interface and stopped using the
+pod network for replication. That is the *only* mechanism that fully detaches a
+workload from the node IP, and it generalises: anything that must not ride the
+node IP gets its own `NetworkAttachmentDefinition`.
+
+It is also expensive per workload — an attachment, an IPAM pool, a pod spec
+change. Worth it for Ceph replication. Not worth it for east-west traffic in
+general.
+
+## Recommended mapping
+
+Two ways to arrange this, and they differ only in which VLAN gets relabelled:
+
+**A — move the node IP to a new cluster VLAN.** VLAN 40 keeps etcd, the API VIP
+and Talos API. Cleanest match to the class names, but it changes every node's
+identity, which is the one address the whole cluster is administered through.
+
+**B (recommended) — leave the node IP on VLAN 40 and move management off it.**
+VLAN 40 becomes the cluster VLAN; a new VLAN carries etcd and the API VIP.
+Identical end state, and no node ever changes its InternalIP.
+
+Under B:
+
+| VLAN | carries |
+|---|---|
+| 49 | Ceph replication |
+| 40 (relabelled *cluster*) | pod east-west, kubelet |
+| new *management* | etcd, kube-apiserver VIP, Talos API by convention |
+| new *external* | LoadBalancer ingress, egress default route |
+
+The property the user asked for — management carries only management traffic —
+holds under B. The label moves; the property is what matters.
+
+## Migration shape
+
+Every step is rolling and reversible. The routing-mode change is gone.
+
+1. **New VLANs** on the switch, trunked to the X710. Nothing uses them.
+2. **Addresses** on the new VLANs via Talos, one node at a time, `--mode=try`.
+   Still unused; this only proves the L2 works.
+3. **etcd** to the management VLAN via `advertisedSubnets`. One node at a time,
+   waiting for a healthy quorum between each.
+4. **API VIP** to the management VLAN via `Layer2VIPConfig.link`. DNS for
+   `k8s.renner.uno` follows.
+5. **cluster-external**: BGP `localAddress`, default route, masquerade flag.
+   Reversible by reverting the override; failure is visible immediately in
+   `cilium-dbg shell -- bgp/peers`.
+6. Verify VLAN 40 by what is *absent* from it, not by what still works.
+
+Step 3 is the one that deserves care — etcd is the only step where a mistake
+costs quorum rather than connectivity.
+
+## What this does not buy
+
+Management currently shares a 10 Gb link and is not contended. Nothing observed
+today is slow because of this. The case for the split is that traffic classes
+with different failure modes and security postures should not be
+indistinguishable — not that something is broken.
+
+## Sources and measurements
+
+Source, Cilium v1.20.1 (the running version):
+
+- `pkg/datapath/linux/node.go` — `installDirectRoute(…, nodeIP net.IP, …)`,
+  fed from `newIP4 = newNode.GetNodeIP(false)`; `createDirectRouteSpec` rejects
+  a next-hop that is not directly reachable.
+- `pkg/node/types/node.go` — `GetNodeIP` prefers `NodeInternalIP`.
+- `pkg/node/sync/local_node_sync.go` — `initFromConfig()` applies
+  `--ipv4-node`, `initFromK8s()` runs afterwards and overwrites it from the
+  Kubernetes Node.
+- `cilium-agent` flag reference — `--direct-routing-device`: "Device name used
+  to connect nodes in direct routing mode".
+
+Talos:
+
+- Multihoming guide — `cluster.etcd.advertisedSubnets`; `listenSubnets`
+  defaults to it. kube-apiserver and Talos API listen addresses are not
+  covered there.
+- `kubernetes/talos/patches/k8s-{1,2,3}.yaml` — `Layer2VIPConfig` for
+  `10.2.40.60` with `link: mgmt0`.
+
+Measured on the cluster:
+
+- `apid` answered on a node address that is not the InternalIP, so the Talos
+  API is not bound to one address.
+- k8s-1 currently holds one routable v4 address plus the VIP on `mgmt0`; its
+  InternalIP had changed since the last session, which is why nothing here
+  names an address as fixed.
+- Cilium runs `routing-mode: tunnel` with `enable-masquerade-to-route-source`
+  unset and no direct-routing device configured.
+
+Not verified: whether the kube-apiserver's `--advertise-address` follows the
+node IP independently of the VIP. It does not affect the recommendation, since
+clients reach the API through the VIP either way.
@@ -70,7 +70,7 @@ spec:
              mountPath: /ca
      containers:
        - name: audiobookshelf
-          image: ghcr.io/advplyr/audiobookshelf:2.36.1 # {"$imagepolicy": "flux-system:audiobookshelf"}
+          image: ghcr.io/advplyr/audiobookshelf:2.36.0 # {"$imagepolicy": "flux-system:audiobookshelf"}
          ports:
            - name: http
              containerPort: 80
@@ -40,7 +40,7 @@ metadata:
    # so this Service produces no public record — LDAP stays
    # LAN-internal, which matches the threat model of the cleartext
    # bind on 389.
-    external-dns.kubernetes.io/hostname: ldap.renner.uno
+    external-dns.alpha.kubernetes.io/hostname: ldap.renner.uno
 spec:
  type: LoadBalancer
  externalTrafficPolicy: Local
@@ -3,4 +3,3 @@ kind: Kustomization
 resources:
  - externalsecret.yaml
  - helmrelease.yaml
-  - networkpolicy.yaml
@@ -1,22 +0,0 @@
-# The apps namespace is default-deny-ingress, so the ServiceMonitor added in
-# monitoring-rules/external-dns-servicemonitor.yaml scraped into a timeout.
-# The selector matches both releases — the Cloudflare one deployed here and
-# external-dns-unifi from infrastructure/ — since they share the chart's
-# app.kubernetes.io/name and land in the same namespace.
-apiVersion: cilium.io/v2
-kind: CiliumNetworkPolicy
-metadata:
-  name: external-dns-allow-metrics-scrape
-  namespace: apps
-spec:
-  endpointSelector:
-    matchLabels:
-      app.kubernetes.io/name: external-dns
-  ingress:
-    - fromEndpoints:
-        - matchLabels:
-            k8s:io.kubernetes.pod.namespace: monitoring
-      toPorts:
-        - ports:
-            - port: "7979"
-              protocol: TCP
@@ -18,7 +18,7 @@ metadata:
    lb-pool: bgp
  annotations:
    lbipam.cilium.io/ips: "10.2.99.87,fd99::87"
-    external-dns.kubernetes.io/hostname: ha.renner.uno
+    external-dns.alpha.kubernetes.io/hostname: ha.renner.uno
 spec:
  type: LoadBalancer
  externalTrafficPolicy: Local
@@ -95,7 +95,7 @@ spec:
              memory: 32Mi
      containers:
        - name: home-assistant
-          image: ghcr.io/home-assistant/home-assistant:2026.9.3 # {"$imagepolicy": "flux-system:home-assistant"}
+          image: ghcr.io/home-assistant/home-assistant:2026.9.1 # {"$imagepolicy": "flux-system:home-assistant"}
          env:
            - name: TZ
              value: Europe/Berlin
@@ -19,7 +19,7 @@ metadata:
    # LAN can connect with `mqtt.renner.uno` (matches the broker cert
    # SAN) instead of the bare IP. No Cloudflare label — kept off the
    # public DNS; UniFi serves a split-DNS answer for LAN clients.
-    external-dns.kubernetes.io/hostname: mqtt.renner.uno
+    external-dns.alpha.kubernetes.io/hostname: mqtt.renner.uno
 spec:
  type: LoadBalancer
  externalTrafficPolicy: Local
@@ -32,7 +32,7 @@ spec:
          type: RuntimeDefault
      containers:
        - name: immich-machine-learning
-          image: ghcr.io/immich-app/immich-machine-learning:v3.2.2 # {"$imagepolicy": "flux-system:immich-machine-learning"}
+          image: ghcr.io/immich-app/immich-machine-learning:v3.1.0 # {"$imagepolicy": "flux-system:immich-machine-learning"}
          ports:
            - name: http
              containerPort: 3003
@@ -87,7 +87,7 @@ spec:
              mountPath: /ca
      containers:
        - name: immich-server
-          image: ghcr.io/immich-app/immich-server:v3.2.2 # {"$imagepolicy": "flux-system:immich-server"}
+          image: ghcr.io/immich-app/immich-server:v3.1.0 # {"$imagepolicy": "flux-system:immich-server"}
          ports:
            - name: http
              containerPort: 3001
@@ -42,11 +42,9 @@ spec:
              type: RuntimeDefault
          initContainers:
            # Seed the libation-config PVC from the bootstrap secret on
-            # first run only — sentinel file gates re-runs so DB updates
-            # from subsequent liberate passes are preserved. Refreshed
-            # Audible tokens are preserved by the container args, not by
-            # this sentinel: the entrypoint copies AccountsSettings.json
-            # into ephemeral /config-internal rather than symlinking it.
+            # first run only — sentinel file gates re-runs so token
+            # refreshes in AccountsSettings.json and DB updates from
+            # subsequent liberate passes are preserved.
            - name: seed-config
              image: busybox:1.37.0
              command: ["/bin/sh", "-eu", "-c"]
@@ -68,11 +66,6 @@ spec:
                  install -m 660 /bootstrap/AccountsSettings.json /libation-state/AccountsSettings.json
                  install -m 660 /bootstrap/Settings.json          /libation-state/Settings.json
                  install -m 660 /bootstrap/LibationContext.db     /libation-state/LibationContext.db
-                  # Without this key the seeded tokens cannot be decrypted:
-                  # TokenStorageMethod is Encrypted and the container has no
-                  # OS secret store, so a missing key file makes the
-                  # entrypoint mint a throwaway one on every run.
-                  install -m 660 /bootstrap/libation-master.key   /libation-state/libation-master.key
                  touch /libation-state/.bootstrapped
                  ls -la /libation-state
              volumeMounts:
@@ -92,23 +85,7 @@ spec:
                  type: RuntimeDefault
          containers:
            - name: libation
-              # Pinned by digest: Spegel caches mutable tag resolutions, so
-              # :latest stayed on 13.7.11 despite imagePullPolicy Always.
-              image: docker.io/rmcrackan/libation:14.2.1@sha256:6f9d8ba19dc3803dd6ba295ab5a7abfb6ff3bc06286eebf4ecd8ccbe43245888
-              # The entrypoint stages config into ephemeral /config-internal
-              # by copy (only the DB is symlinked to the PVC), so Audible
-              # tokens refreshed during a run are lost at container exit.
-              # Passing args makes the entrypoint set up the environment and
-              # then exec this instead of its own loop; the trap copies the
-              # refreshed file back to the PVC on every exit path.
-              args:
-                - /bin/sh
-                - -c
-                - |
-                  trap 'cp /config-internal/AccountsSettings.json /libation-state/AccountsSettings.json' EXIT
-                  set -e
-                  /libation/LibationCli scan
-                  /libation/LibationCli liberate
+              image: docker.io/rmcrackan/libation:latest
              env:
                - name: LIBATION_CONFIG_DIR
                  value: /libation-state
@@ -6,5 +6,3 @@ resources:
  - embed.yaml
  - gate-service.yaml
  - ingress.yaml
-  - wiki-embed.yaml
-  - wiki-embed-ingress.yaml
@@ -1,57 +0,0 @@
---
-# wiki-embed.renner.uno — LAN-only (no dns.renner.uno/public label), consumed by
-# bin/wiki_rank.py on the workstation and the nightly wiki jobs. Machine client,
-# so Traefik BasicAuth instead of the browser-only Authentik gate; same pattern
-# as kubernetes/apps/monitoring/loki-ingress.yaml.
-apiVersion: traefik.io/v1alpha1
-kind: Middleware
-metadata:
-  name: wiki-embed-basic-auth
-  namespace: ai
-spec:
-  basicAuth:
-    secret: wiki-embed-basic-auth
-    realm: wiki-embed
---
-# OpenBao key apps/ai/wiki-embed-basic-auth, field `users` = htpasswd line.
-apiVersion: external-secrets.io/v1
-kind: ExternalSecret
-metadata:
-  name: wiki-embed-basic-auth
-  namespace: ai
-spec:
-  refreshInterval: 1h
-  secretStoreRef:
-    kind: ClusterSecretStore
-    name: openbao
-  target:
-    name: wiki-embed-basic-auth
-  dataFrom:
-    - extract:
-        key: apps/ai/wiki-embed-basic-auth
---
-apiVersion: networking.k8s.io/v1
-kind: Ingress
-metadata:
-  name: wiki-embed
-  namespace: ai
-  annotations:
-    cert-manager.io/cluster-issuer: letsencrypt-cluster
-    traefik.ingress.kubernetes.io/router.middlewares: ai-wiki-embed-basic-auth@kubernetescrd
-spec:
-  ingressClassName: traefik
-  tls:
-    - hosts:
-        - wiki-embed.renner.uno
-      secretName: wiki-embed-tls
-  rules:
-    - host: wiki-embed.renner.uno
-      http:
-        paths:
-          - path: /
-            pathType: Prefix
-            backend:
-              service:
-                name: wiki-embed
-                port:
-                  number: 8080
@@ -1,117 +0,0 @@
---
-# Embedding endpoint for the ~/Notes wiki ranker (bin/wiki_rank.py). Own
-# Deployment so wiki traffic never queues behind Paperless on llama-embed.
-# Same model and PVC as llama-embed, so no second download.
-apiVersion: apps/v1
-kind: Deployment
-metadata:
-  name: wiki-embed
-  namespace: ai
-  labels:
-    app.kubernetes.io/name: wiki-embed
-spec:
-  replicas: 1
-  # Above the 600s startupProbe budget; see embed.yaml.
-  progressDeadlineSeconds: 900
-  strategy:
-    type: Recreate
-  selector:
-    matchLabels:
-      app.kubernetes.io/name: wiki-embed
-  template:
-    metadata:
-      labels:
-        app.kubernetes.io/name: wiki-embed
-    spec:
-      securityContext:
-        runAsNonRoot: true
-        runAsUser: 1000
-        runAsGroup: 1000
-        fsGroup: 1000
-        seccompProfile:
-          type: RuntimeDefault
-      containers:
-        - name: wiki-embed
-          image: ghcr.io/ggml-org/llama.cpp:server-vulkan-b10731
-          args:
-            - --hf-repo
-            - gpustack/bge-m3-GGUF:Q8_0
-            - --embeddings
-            - --alias
-            - bge-m3
-            # bge-m3 is trained for CLS pooling; the GGUF default is not.
-            - --pooling
-            - cls
-            - --n-gpu-layers
-            - "99"
-            - --threads
-            - "2"
-            # wiki_rank.py caps each note at 1500 chars and sends 2 per request.
-            - --batch-size
-            - "1536"
-            - --ubatch-size
-            - "1536"
-            - --ctx-size
-            - "4096"
-            - --no-webui
-          env:
-            - name: LLAMA_CACHE
-              value: /models
-          ports:
-            - name: http
-              containerPort: 8080
-          securityContext:
-            allowPrivilegeEscalation: false
-            capabilities:
-              drop:
-                - ALL
-          # Same sizing as llama-embed: 1.6 GiB measured peak, headroom for slots.
-          resources:
-            requests:
-              cpu: 250m
-              memory: 2Gi
-              gpu.intel.com/i915: "1"
-            limits:
-              cpu: "2"
-              memory: 8Gi
-              gpu.intel.com/i915: "1"
-          startupProbe:
-            httpGet:
-              path: /health
-              port: http
-            periodSeconds: 10
-            failureThreshold: 60
-          readinessProbe:
-            httpGet:
-              path: /health
-              port: http
-            periodSeconds: 10
-          livenessProbe:
-            httpGet:
-              path: /health
-              port: http
-            periodSeconds: 30
-            timeoutSeconds: 10
-            failureThreshold: 6
-          volumeMounts:
-            - name: models
-              mountPath: /models
-      volumes:
-        - name: models
-          persistentVolumeClaim:
-            claimName: llama-server-models
---
-apiVersion: v1
-kind: Service
-metadata:
-  name: wiki-embed
-  namespace: ai
-  labels:
-    app.kubernetes.io/name: wiki-embed
-spec:
-  selector:
-    app.kubernetes.io/name: wiki-embed
-  ports:
-    - name: http
-      port: 8080
-      targetPort: http
@@ -1,25 +0,0 @@
-# Both external-dns releases (Cloudflare and the UniFi webhook) carry
-# app.kubernetes.io/name: external-dns and differ only by instance, so one
-# monitor covers both. jobLabel keeps them apart as separate job values.
-apiVersion: monitoring.coreos.com/v1
-kind: ServiceMonitor
-metadata:
-  name: external-dns
-  namespace: monitoring
-  labels:
-    app.kubernetes.io/name: external-dns
-    release: kube-prometheus-stack
-spec:
-  jobLabel: app.kubernetes.io/instance
-  namespaceSelector:
-    matchNames:
-      - apps
-  selector:
-    matchLabels:
-      app.kubernetes.io/name: external-dns
-  endpoints:
-    # The controller's own metrics. Port 8080 on the unifi release is the
-    # webhook sidecar and exports no external_dns_* series.
-    - port: http
-      path: /metrics
-      interval: 60s
@@ -21,14 +21,6 @@ resources:
  # alerted if that stopped — HetznerBackupEndpointDown only probes the S3
  # endpoint, which proves reachability, not that anything was written.
  - prometheusrule-cnpg-backup.yaml
-  # DNS publication. Nothing scraped external-dns, so it deleted six LAN
-  # records on 2026-09-12 and stayed silent for five days.
-  - prometheusrule-external-dns.yaml
-  - external-dns-servicemonitor.yaml
-  # Outcome check for the same failure: the names themselves, asked of the
-  # UDM. Catches a cause external-dns cannot see, e.g. records removed there.
-  - prometheusrule-dns-resolution.yaml
-  - probes-dns.yaml
  - alloy-servicemonitor.yaml
  - cilium-servicemonitor.yaml
  - cnpg-podmonitors.yaml
@@ -1,143 +0,0 @@
-# DNS probes for the names external-dns publishes to the UDM from
-# LoadBalancer Services. One Probe per name because the dns prober's
-# query_name is module-scoped; the target is the resolver, queried on both
-# families so a single-stack blip does not read as a missing record.
-#
-# The UDM is queried directly rather than through the cluster resolver:
-# k8s-gateway caches these with `serve_stale 1h immediate`, which would keep
-# a deleted record green for an hour.
-#
-# `release` is load-bearing — probeSelector matches on it.
-apiVersion: monitoring.coreos.com/v1
-kind: Probe
-metadata:
-  name: dns-lan-ha
-  namespace: monitoring
-  labels:
-    release: kube-prometheus-stack
-spec:
-  jobName: blackbox-dns
-  prober:
-    url: blackbox-exporter-prometheus-blackbox-exporter.monitoring.svc.cluster.local:9115
-  module: dns_lan_ha
-  interval: 60s
-  scrapeTimeout: 10s
-  targets:
-    staticConfig:
-      static:
-        - "[fd00::1]:53"
-        - "10.2.40.1:53"
-      # instance is the resolver and identical across these Probes, so the
-      # queried name has to come from a label.
-      labels:
-        dns_name: ha.renner.uno
---
-apiVersion: monitoring.coreos.com/v1
-kind: Probe
-metadata:
-  name: dns-lan-mqtt
-  namespace: monitoring
-  labels:
-    release: kube-prometheus-stack
-spec:
-  jobName: blackbox-dns
-  prober:
-    url: blackbox-exporter-prometheus-blackbox-exporter.monitoring.svc.cluster.local:9115
-  module: dns_lan_mqtt
-  interval: 60s
-  scrapeTimeout: 10s
-  targets:
-    staticConfig:
-      static:
-        - "[fd00::1]:53"
-        - "10.2.40.1:53"
-      labels:
-        dns_name: mqtt.renner.uno
---
-apiVersion: monitoring.coreos.com/v1
-kind: Probe
-metadata:
-  name: dns-lan-ldap
-  namespace: monitoring
-  labels:
-    release: kube-prometheus-stack
-spec:
-  jobName: blackbox-dns
-  prober:
-    url: blackbox-exporter-prometheus-blackbox-exporter.monitoring.svc.cluster.local:9115
-  module: dns_lan_ldap
-  interval: 60s
-  scrapeTimeout: 10s
-  targets:
-    staticConfig:
-      static:
-        - "[fd00::1]:53"
-        - "10.2.40.1:53"
-      labels:
-        dns_name: ldap.renner.uno
---
-apiVersion: monitoring.coreos.com/v1
-kind: Probe
-metadata:
-  name: dns-lan-syslog
-  namespace: monitoring
-  labels:
-    release: kube-prometheus-stack
-spec:
-  jobName: blackbox-dns
-  prober:
-    url: blackbox-exporter-prometheus-blackbox-exporter.monitoring.svc.cluster.local:9115
-  module: dns_lan_syslog
-  interval: 60s
-  scrapeTimeout: 10s
-  targets:
-    staticConfig:
-      static:
-        - "[fd00::1]:53"
-        - "10.2.40.1:53"
-      labels:
-        dns_name: syslog.renner.uno
---
-apiVersion: monitoring.coreos.com/v1
-kind: Probe
-metadata:
-  name: dns-lan-ftp-scanner
-  namespace: monitoring
-  labels:
-    release: kube-prometheus-stack
-spec:
-  jobName: blackbox-dns
-  prober:
-    url: blackbox-exporter-prometheus-blackbox-exporter.monitoring.svc.cluster.local:9115
-  module: dns_lan_ftp_scanner
-  interval: 60s
-  scrapeTimeout: 10s
-  targets:
-    staticConfig:
-      static:
-        - "[fd00::1]:53"
-        - "10.2.40.1:53"
-      labels:
-        dns_name: ftp.paperless.renner.uno
---
-apiVersion: monitoring.coreos.com/v1
-kind: Probe
-metadata:
-  name: dns-lan-syncthing
-  namespace: monitoring
-  labels:
-    release: kube-prometheus-stack
-spec:
-  jobName: blackbox-dns
-  prober:
-    url: blackbox-exporter-prometheus-blackbox-exporter.monitoring.svc.cluster.local:9115
-  module: dns_lan_syncthing
-  interval: 60s
-  scrapeTimeout: 10s
-  targets:
-    staticConfig:
-      static:
-        - "[fd00::1]:53"
-        - "10.2.40.1:53"
-      labels:
-        dns_name: syncthing.paperless.renner.uno
@@ -1,49 +0,0 @@
-apiVersion: monitoring.coreos.com/v1
-kind: PrometheusRule
-metadata:
-  name: dns-resolution
-  namespace: monitoring
-  labels:
-    # ruleSelector matches on this label; without it the rule is created but
-    # never loaded into Prometheus.
-    release: kube-prometheus-stack
-spec:
-  groups:
-    - name: dns-resolution.rules
-      rules:
-        # min by dns_name: both resolver families are probed, and the record
-        # is only gone when neither answers. A single-family failure is the
-        # UDM's dnsmasq refusing connections in bursts (k8s-gateway carries
-        # the same note), which is why the window is 30m rather than the 10m
-        # used for HTTP probes.
-        - alert: LanDnsRecordMissing
-          expr: min by (dns_name) (probe_success{job="blackbox-dns"}) == 0
-          for: 30m
-          labels:
-            severity: critical
-          annotations:
-            summary: "{{ $labels.dns_name }} no longer resolves to its LAN address"
-            description: >-
-              The UDM stopped answering for a name external-dns publishes from
-              a LoadBalancer Service, or answered with the public
-              *.renner.uno wildcard instead of the record. Clients that use
-              the name rather than the address lose the service while
-              everything in the cluster stays healthy — nothing else in the
-              stack reports this. Check the external-dns alerts first, then
-              the record on the UDM.
-
-        # Dead-man's-switch, mirroring BlackboxProbesAbsent: without probe
-        # results the rule above cannot fire.
-        - alert: LanDnsProbesAbsent
-          expr: absent(probe_success{job="blackbox-dns"})
-          for: 15m
-          labels:
-            severity: critical
-          annotations:
-            summary: "No LAN DNS probe results are reaching Prometheus"
-            description: >-
-              The blackbox-dns Probes stopped being discovered or the exporter
-              rejects the modules. A 400 from /probe yields no probe_* series
-              at all, which looks the same as no Probe — check that every
-              dns_lan_* module the Probes reference exists in the
-              blackbox-exporter config.
@@ -1,87 +0,0 @@
-apiVersion: monitoring.coreos.com/v1
-kind: PrometheusRule
-metadata:
-  name: external-dns
-  namespace: monitoring
-  labels:
-    # ruleSelector matches on this label; without it the rule is created but
-    # never loaded into Prometheus.
-    release: kube-prometheus-stack
-spec:
-  groups:
-    - name: external-dns.rules
-      rules:
-        # The signal that was missing on 2026-09-12: the sources stopped
-        # producing endpoints they had produced before, and policy=sync
-        # deleted the records. 10% below the 24h peak clears a single
-        # retired Ingress (2 of ~48) and catches a whole class of names.
-        - alert: ExternalDNSSourceEndpointsDropped
-          expr: >-
-            external_dns_source_endpoints_total
-              < 0.9 * max_over_time(external_dns_source_endpoints_total[24h])
-          for: 30m
-          labels:
-            severity: critical
-          annotations:
-            summary: "{{ $labels.job }} is publishing {{ $value }} endpoints, below its 24h peak"
-            description: >-
-              external-dns stopped seeing endpoints it had before. With
-              policy=sync the matching DNS records are deleted, so names stop
-              resolving without anything else failing. Compare
-              external_dns_source_endpoints_total against the last 24h, then
-              check whether the resources still carry a hostname annotation
-              the running version understands — a changed annotation prefix
-              default empties the source silently. Deliberate removals also
-              fire this and resolve themselves 24h later.
-
-        # Source errors mean the Kubernetes side failed; registry errors mean
-        # the provider did. Either way desired state is computed from partial
-        # input, which policy=sync acts on.
-        - alert: ExternalDNSSourceErrors
-          expr: increase(external_dns_source_errors_total[15m]) > 0
-          for: 15m
-          labels:
-            severity: warning
-          annotations:
-            summary: "{{ $labels.job }} is failing to read its sources"
-            description: >-
-              external-dns could not list the Kubernetes resources it derives
-              records from. Check the pod log and its RBAC.
-
-        - alert: ExternalDNSRegistryErrors
-          expr: increase(external_dns_registry_errors_total[15m]) > 0
-          for: 15m
-          labels:
-            severity: warning
-          annotations:
-            summary: "{{ $labels.job }} is failing to reach its DNS provider"
-            description: >-
-              external-dns could not read or write records at the provider.
-              For external-dns-unifi that is the UniFi webhook sidecar and the
-              controller behind it; for external-dns it is the Cloudflare API.
-
-        - alert: ExternalDNSDown
-          expr: up{job=~"external-dns.*"} == 0
-          for: 15m
-          labels:
-            severity: critical
-          annotations:
-            summary: "{{ $labels.job }} is not being scraped"
-            description: >-
-              The external-dns instance is down or unreachable. Records already
-              at the provider keep resolving, but nothing reconciles them any
-              more and every other alert in this group is blind.
-
-        # Dead-man's-switch for the group, mirroring BlackboxProbesAbsent: a
-        # dropped ServiceMonitor makes every rule above evaluate to nothing.
-        - alert: ExternalDNSMetricsMissing
-          expr: absent(external_dns_source_endpoints_total)
-          for: 15m
-          labels:
-            severity: critical
-          annotations:
-            summary: "No external-dns metrics are reaching Prometheus"
-            description: >-
-              Either both external-dns instances are gone or the ServiceMonitor
-              stopped being discovered. Verify it still carries
-              release=kube-prometheus-stack.
@@ -232,7 +232,7 @@ metadata:
    lb-pool: bgp
  annotations:
    io.cilium/lb-ipam-ips: "10.2.99.84,fd99::84"
-    external-dns.kubernetes.io/hostname: syslog.renner.uno
+    external-dns.alpha.kubernetes.io/hostname: syslog.renner.uno
 spec:
  type: LoadBalancer
  externalTrafficPolicy: Local
@@ -72,73 +72,3 @@ spec:
          tcp:
            preferred_ip_protocol: ip6
            ip_protocol_fallback: true
-        # One module per name: the dns prober takes the RESOLVER as its
-        # target and reads query_name from the module, so a name cannot be
-        # passed per scrape. The Probes in monitoring-rules/probes-dns.yaml
-        # point these at the UDM.
-        #
-        # Both answer checks are load-bearing. valid_rcodes alone passes a
-        # deleted record two labels deep (NOERROR, no answer), and the public
-        # *.renner.uno wildcard answers one label deep with a CNAME to
-        # proxy.renner.uno — NOERROR with the WAN proxy behind it. Requiring
-        # an A and rejecting a CNAME separates a real record from both.
-        dns_lan_ha:
-          prober: dns
-          timeout: 5s
-          dns:
-            query_name: ha.renner.uno
-            query_type: A
-            valid_rcodes: [NOERROR]
-            validate_answer_rrs:
-              fail_if_not_matches_regexp: ['\sIN\s+A\s']
-              fail_if_matches_regexp: ['\sIN\s+CNAME\s']
-        dns_lan_mqtt:
-          prober: dns
-          timeout: 5s
-          dns:
-            query_name: mqtt.renner.uno
-            query_type: A
-            valid_rcodes: [NOERROR]
-            validate_answer_rrs:
-              fail_if_not_matches_regexp: ['\sIN\s+A\s']
-              fail_if_matches_regexp: ['\sIN\s+CNAME\s']
-        dns_lan_ldap:
-          prober: dns
-          timeout: 5s
-          dns:
-            query_name: ldap.renner.uno
-            query_type: A
-            valid_rcodes: [NOERROR]
-            validate_answer_rrs:
-              fail_if_not_matches_regexp: ['\sIN\s+A\s']
-              fail_if_matches_regexp: ['\sIN\s+CNAME\s']
-        dns_lan_syslog:
-          prober: dns
-          timeout: 5s
-          dns:
-            query_name: syslog.renner.uno
-            query_type: A
-            valid_rcodes: [NOERROR]
-            validate_answer_rrs:
-              fail_if_not_matches_regexp: ['\sIN\s+A\s']
-              fail_if_matches_regexp: ['\sIN\s+CNAME\s']
-        dns_lan_ftp_scanner:
-          prober: dns
-          timeout: 5s
-          dns:
-            query_name: ftp.paperless.renner.uno
-            query_type: A
-            valid_rcodes: [NOERROR]
-            validate_answer_rrs:
-              fail_if_not_matches_regexp: ['\sIN\s+A\s']
-              fail_if_matches_regexp: ['\sIN\s+CNAME\s']
-        dns_lan_syncthing:
-          prober: dns
-          timeout: 5s
-          dns:
-            query_name: syncthing.paperless.renner.uno
-            query_type: A
-            valid_rcodes: [NOERROR]
-            validate_answer_rrs:
-              fail_if_not_matches_regexp: ['\sIN\s+A\s']
-              fail_if_matches_regexp: ['\sIN\s+CNAME\s']
@@ -55,7 +55,7 @@ spec:
        # stdlib only; Flux keeps the tag in lockstep via the shared image
        # policy) — no extra dependency.
        - name: reconcile-publish-ip
-          image: ghcr.io/music-assistant/server:2.10.4 # {"$imagepolicy": "flux-system:music-assistant"}
+          image: ghcr.io/music-assistant/server:2.10.2 # {"$imagepolicy": "flux-system:music-assistant"}
          command: ["/app/venv/bin/python", "/scripts/reconcile.py"]
          volumeMounts:
            - name: data
@@ -65,7 +65,7 @@ spec:
              readOnly: true
      containers:
        - name: music-assistant
-          image: ghcr.io/music-assistant/server:2.10.4 # {"$imagepolicy": "flux-system:music-assistant"}
+          image: ghcr.io/music-assistant/server:2.10.2 # {"$imagepolicy": "flux-system:music-assistant"}
          env:
            - name: LOG_LEVEL
              value: info
@@ -43,7 +43,7 @@ spec:
              mountPath: /consume
      containers:
        - name: vsftpd
-          image: gitea.renner.uno/k.renner/vsftpd:build.1789630783 # {"$imagepolicy": "flux-system:vsftpd"}
+          image: gitea.renner.uno/k.renner/vsftpd:build.1784731469 # {"$imagepolicy": "flux-system:vsftpd"}
          ports:
            - name: ftp-ctrl
              containerPort: 21
@@ -70,19 +70,24 @@ spec:
            - name: consume
              mountPath: /home/scanner
              subPath: scanner
-          # Asserts the control socket is listening without opening an FTP
-          # connection: vsftpd 3.0.5 segfaults its listener on a fraction of
-          # the connections it accepts. The passive range is negotiated per
-          # transfer and is not listening at rest, so it is not probed.
+          # FTP has no health verb, so the control port accepting a connection
+          # is what can be asserted. It is enough for the failure that matters:
+          # if vsftpd dies the scanner cannot deposit anything and nothing else
+          # in the chain notices, because paperless only ever sees an empty
+          # consume directory — which is indistinguishable from nobody having
+          # scanned.
+          #
+          # Only the control port is probed. The passive range is negotiated
+          # per transfer and is not listening at rest.
          livenessProbe:
-            exec:
-              command: ["sh", "-c", "netstat -lnt | grep -q ':21 '"]
+            tcpSocket:
+              port: ftp-ctrl
            initialDelaySeconds: 15
            periodSeconds: 30
            timeoutSeconds: 5
          readinessProbe:
-            exec:
-              command: ["sh", "-c", "netstat -lnt | grep -q ':21 '"]
+            tcpSocket:
+              port: ftp-ctrl
            initialDelaySeconds: 5
            periodSeconds: 10
            timeoutSeconds: 5
@@ -119,7 +124,7 @@ metadata:
    # label — FTP must not be exposed via the cargo tunnel (clear-text
    # protocol, public exposure not appropriate). The brother scanner
    # uses this hostname instead of the LB IP.
-    external-dns.kubernetes.io/hostname: ftp.paperless.renner.uno
+    external-dns.alpha.kubernetes.io/hostname: ftp.paperless.renner.uno
 spec:
  type: LoadBalancer
  externalTrafficPolicy: Local
@@ -22,7 +22,7 @@ spec:
          type: RuntimeDefault
      containers:
        - name: gotenberg
-          image: gotenberg/gotenberg:8.37.0 # {"$imagepolicy": "flux-system:gotenberg"}
+          image: gotenberg/gotenberg:8.36.0 # {"$imagepolicy": "flux-system:gotenberg"}
          args:
            - "gotenberg"
            - "--chromium-disable-javascript=true"
@@ -25,15 +25,18 @@ resources:
 configMapGenerator:
  # vsftpd config for the ftp-scanner deployment. The content-hash name
  # suffix is intentionally LEFT ENABLED (unlike home-assistant's nginx
-  # config): vsftpd reads these into vsftpd.conf at container start, so a
-  # changed value must roll the pod, which the renamed ConfigMap does.
-  #
-  # No PASV_ADDRESS: the LB address is declared once, on the Service
-  # (io.cilium/lb-ipam-ips). The image default 0.0.0.0 tells PASV clients
-  # to reuse the control-connection address.
+  # config): vsftpd templates pasv_address from PASV_ADDRESS at container
+  # start, so a changed value must roll the pod. With the hash suffix,
+  # editing any value here renames the generated ConfigMap, which mutates
+  # the Deployment's pod template and triggers an automatic rolling
+  # restart. Without it, an IP change (e.g. an LB re-allocation) silently
+  # leaves the running pod advertising the stale PASV address, which
+  # breaks classic-PASV clients (older Brother scanners) while EPSV
+  # clients keep working.
  - name: ftp-scanner-config
    namespace: paperless
    literals:
+      - PASV_ADDRESS=10.2.99.82
      - PASV_MIN_PORT=21100
      - PASV_MAX_PORT=21104
      - FTP_USER=scanner
@@ -183,7 +183,7 @@ spec:
              drop: ["ALL"]
      containers:
        - name: paperless
-          image: ghcr.io/paperless-ngx/paperless-ngx:3.2.0 # {"$imagepolicy": "flux-system:paperless-ngx"}
+          image: ghcr.io/paperless-ngx/paperless-ngx:3.1.3 # {"$imagepolicy": "flux-system:paperless-ngx"}
          ports:
            - name: http
              containerPort: 8000
@@ -145,7 +145,7 @@ metadata:
    lb-pool: bgp
  annotations:
    io.cilium/lb-ipam-ips: "10.2.99.85,fd99::85"
-    external-dns.kubernetes.io/hostname: syncthing.paperless.renner.uno
+    external-dns.alpha.kubernetes.io/hostname: syncthing.paperless.renner.uno
 spec:
  type: LoadBalancer
  externalTrafficPolicy: Local
@@ -134,7 +134,7 @@ spec:
        # is a device-bricking class of risk. Bumps are a reviewed, manual
        # act — see kubernetes/infrastructure/image-automation/README.md.
        - name: terminus
-          image: ghcr.io/usetrmnl/terminus:0.74.0 # {"$imagepolicy": "flux-system:terminus"}
+          image: ghcr.io/usetrmnl/terminus:0.72.0 # {"$imagepolicy": "flux-system:terminus"}
          ports:
            - name: http
              containerPort: 2345
@@ -197,7 +197,7 @@ spec:
            capabilities:
              drop: ["ALL"]
        - name: sidekiq
-          image: ghcr.io/usetrmnl/terminus:0.74.0 # {"$imagepolicy": "flux-system:terminus"}
+          image: ghcr.io/usetrmnl/terminus:0.72.0 # {"$imagepolicy": "flux-system:terminus"}
          command: ["bundle", "exec", "sidekiq", "-r", "./config/sidekiq.rb"]
          envFrom:
            - configMapRef:
@@ -18,7 +18,7 @@ spec:
    spec:
      containers:
        - name: whisper
-          image: rhasspy/wyoming-whisper:3.8.1 # {"$imagepolicy": "flux-system:wyoming-whisper"}
+          image: rhasspy/wyoming-whisper:3.7.0 # {"$imagepolicy": "flux-system:wyoming-whisper"}
          args:
            - --model=base-int8
            - --language=de
@@ -1,137 +0,0 @@
---
-# ServiceAccount is the OpenBao identity (auth/kubernetes/role/wiki-index); no
-# Kubernetes API access needed.
-apiVersion: v1
-kind: ServiceAccount
-metadata:
-  name: wiki-index
-  namespace: ai
-  labels:
-    app.kubernetes.io/name: wiki-index
---
-# Index (~17 MB JSON for 1700 notes). Survives restarts so a pod start does not
-# re-embed the wiki; a lost volume only costs one full sync (~5 min).
-apiVersion: v1
-kind: PersistentVolumeClaim
-metadata:
-  name: wiki-index-data
-  namespace: ai
-spec:
-  accessModes:
-    - ReadWriteOnce
-  storageClassName: ceph-block
-  resources:
-    requests:
-      storage: 1Gi
---
-# Embedding index over the Notes repo's wiki/ (ci/images/wiki-index). Syncs from
-# Gitea on push webhook and every RESYNC_SECONDS, embeds via wiki-embed, serves
-# POST /rank to ~/Notes/bin/wiki_rank.py through wiki-index.renner.uno.
-apiVersion: apps/v1
-kind: Deployment
-metadata:
-  name: wiki-index
-  namespace: ai
-  labels:
-    app.kubernetes.io/name: wiki-index
-spec:
-  replicas: 1
-  strategy:
-    type: Recreate
-  selector:
-    matchLabels:
-      app.kubernetes.io/name: wiki-index
-  template:
-    metadata:
-      labels:
-        app.kubernetes.io/name: wiki-index
-    spec:
-      serviceAccountName: wiki-index
-      securityContext:
-        runAsNonRoot: true
-        runAsUser: 65534
-        runAsGroup: 65534
-        fsGroup: 65534
-        seccompProfile:
-          type: RuntimeDefault
-      containers:
-        - name: wiki-index
-          # Flux image automation bumps this to the newest build.<unix-ts> tag.
-          image: gitea.renner.uno/gitea_admin/wiki-index:2026-09-20-4b529cd1 # {"$imagepolicy": "flux-system:wiki-index"}
-          securityContext:
-            allowPrivilegeEscalation: false
-            readOnlyRootFilesystem: true
-            capabilities:
-              drop: ["ALL"]
-          env:
-            - name: GITEA_URL
-              value: https://gitea.renner.uno
-            - name: NOTES_REPO
-              value: k.renner/notes
-            - name: NOTES_BRANCH
-              value: main
-            - name: EMBED_URL
-              value: http://wiki-embed.ai.svc.cluster.local:8080
-            - name: BAO_ADDR
-              value: http://openbao.openbao.svc.cluster.local:8200
-            - name: BAO_ROLE
-              value: wiki-index
-            - name: BAO_SECRET_PATH
-              value: homelab/data/apps/ai/wiki-index
-            - name: INDEX_PATH
-              value: /data/index.json
-            - name: RESYNC_SECONDS
-              value: "1800"
-          ports:
-            - name: http
-              containerPort: 8080
-          # Pure-Python cosine over 1700 x 1024 floats per query; the index itself is ~60 MB in memory.
-          resources:
-            requests:
-              cpu: 100m
-              memory: 256Mi
-            limits:
-              cpu: "1"
-              memory: 512Mi
-          # /health: the process serves (liveness). /ready: an index exists; a cold
-          # start on an empty PVC embeds every note (~5 min) before traffic arrives.
-          startupProbe:
-            httpGet:
-              path: /health
-              port: http
-            periodSeconds: 5
-            failureThreshold: 12
-          readinessProbe:
-            httpGet:
-              path: /ready
-              port: http
-            periodSeconds: 10
-          livenessProbe:
-            httpGet:
-              path: /health
-              port: http
-            periodSeconds: 60
-            timeoutSeconds: 10
-            failureThreshold: 3
-          volumeMounts:
-            - name: data
-              mountPath: /data
-      volumes:
-        - name: data
-          persistentVolumeClaim:
-            claimName: wiki-index-data
---
-apiVersion: v1
-kind: Service
-metadata:
-  name: wiki-index
-  namespace: ai
-  labels:
-    app.kubernetes.io/name: wiki-index
-spec:
-  selector:
-    app.kubernetes.io/name: wiki-index
-  ports:
-    - name: http
-      port: 8080
-      targetPort: http
@@ -1,56 +0,0 @@
---
-# wiki-index.renner.uno — LAN-only (no dns.renner.uno/public label), machine
-# client (bin/wiki_rank.py), so Traefik BasicAuth like wiki-embed.renner.uno.
-# Gitea's webhook does not pass here; it targets the ClusterIP Service directly.
-apiVersion: traefik.io/v1alpha1
-kind: Middleware
-metadata:
-  name: wiki-index-basic-auth
-  namespace: ai
-spec:
-  basicAuth:
-    secret: wiki-index-basic-auth
-    realm: wiki-index
---
-# OpenBao key apps/ai/wiki-index-basic-auth, field `users` = htpasswd line.
-apiVersion: external-secrets.io/v1
-kind: ExternalSecret
-metadata:
-  name: wiki-index-basic-auth
-  namespace: ai
-spec:
-  refreshInterval: 1h
-  secretStoreRef:
-    kind: ClusterSecretStore
-    name: openbao
-  target:
-    name: wiki-index-basic-auth
-  dataFrom:
-    - extract:
-        key: apps/ai/wiki-index-basic-auth
---
-apiVersion: networking.k8s.io/v1
-kind: Ingress
-metadata:
-  name: wiki-index
-  namespace: ai
-  annotations:
-    cert-manager.io/cluster-issuer: letsencrypt-cluster
-    traefik.ingress.kubernetes.io/router.middlewares: ai-wiki-index-basic-auth@kubernetescrd
-spec:
-  ingressClassName: traefik
-  tls:
-    - hosts:
-        - wiki-index.renner.uno
-      secretName: wiki-index-tls
-  rules:
-    - host: wiki-index.renner.uno
-      http:
-        paths:
-          - path: /
-            pathType: Prefix
-            backend:
-              service:
-                name: wiki-index
-                port:
-                  number: 8080
@@ -1,5 +0,0 @@
-apiVersion: kustomize.config.k8s.io/v1beta1
-kind: Kustomization
-resources:
-  - deployment.yaml
-  - ingress.yaml
@@ -34,7 +34,7 @@ spec:
          # Image automation, range 3.x (see image-automation/woodpecker.yaml).
          # MUST stay version-matched with woodpecker-server (gRPC parity) — both
          # track the same range/filter so they bump together on each release.
-          image: woodpeckerci/woodpecker-agent:v3.18.1 # {"$imagepolicy": "flux-system:woodpecker-agent"}
+          image: woodpeckerci/woodpecker-agent:v3.18.0 # {"$imagepolicy": "flux-system:woodpecker-agent"}
          env:
            - name: WOODPECKER_SERVER
              value: woodpecker.woodpecker.svc.cluster.local:9000 # in-cluster gRPC, replaces LXC :9000
@@ -40,7 +40,7 @@ spec:
          # minor + patch auto-update within 3.x; only the 4.0 major is held for
          # a reviewed bump. Safe now the DB is postgres (the old sqlite-surprise
          # -bump hazard is gone). Stays in lockstep with the agent.
-          image: woodpeckerci/woodpecker-server:v3.18.1 # {"$imagepolicy": "flux-system:woodpecker-server"}
+          image: woodpeckerci/woodpecker-server:v3.18.0 # {"$imagepolicy": "flux-system:woodpecker-server"}
          env:
            - name: WOODPECKER_HOST
              value: https://ci.renner.uno
@@ -21,7 +21,7 @@ spec:
      # Service; nothing on this path needs the host's network namespace.
      containers:
        - name: zone-configurator
-          image: everythingsmarthome/everything-presence-mmwave-configurator:2.2.5 # {"$imagepolicy": "flux-system:mmwave-configurator"}
+          image: everythingsmarthome/everything-presence-mmwave-configurator:2.2.4 # {"$imagepolicy": "flux-system:mmwave-configurator"}
          env:
            - name: HA_BASE_URL
              value: "http://home-assistant:80"
@@ -610,8 +610,8 @@ if [[ "$PHASE" == "hcloudtest" ]]; then
    # Constraints on the choice: cpx11/cpx21 are US-only and cax* are ARM, which
    # would not match the amd64 schematic this cluster installs. That leaves the
    # cx*/cpx*2 families in the German locations.
-    # Two segments only, unlike the homelab's split in
-    # NETWORK-traffic-classes.md: here the node address carries management,
+    # Today's traffic-class split, not the target one in
+    # PROPOSAL-traffic-class-separation.md: the node address carries management,
    # cluster-internal and external alike, and Ceph replication has its own.
    #
    # The order is load-bearing: it decides which kernel interface each segment
@@ -1,22 +0,0 @@
-apiVersion: kustomize.toolkit.fluxcd.io/v1
-kind: Kustomization
-metadata:
-  name: wiki-index
-  namespace: flux-system
-spec:
-  interval: 10m
-  retryInterval: 1m
-  # First start on an empty PVC embeds every note (~5 min) before /health is 200.
-  timeout: 20m
-  sourceRef:
-    kind: GitRepository
-    name: flux-system
-  path: ./kubernetes/apps/wiki-index
-  prune: true
-  wait: true
-  dependsOn:
-    - name: infrastructure
-    - name: infrastructure-crds
-    - name: infrastructure-ceph-cluster
-    # wiki-embed (the embedding backend) lives in the llama-server app.
-    - name: llama-server
@@ -1,103 +0,0 @@
-# Identical on every node and free of addresses. defines.conf and peers.conf
-# are rendered per node by the init container; the first is included above the
-# templates that reference it, the second below them.
-include "/etc/bird/defines.conf";
-
-router id from "pod0";
-log stderr all;
-
-protocol device { scan time 10; }
-
-# Everything a pod may need to reach, in one table, resolved by longest prefix:
-# the node's attached networks, the peers' pod CIDRs, and a default for the
-# rest. No rule has to carve out exceptions, because the specific routes win on
-# their own.
-#
-# The table is BIRD's alone. If this daemon stops, it empties, the rule finds
-# nothing and falls through to main — pod traffic returns to VLAN 40,
-# unseparated rather than misrouted. That is why the default route lives here
-# and not in the Talos config: a default that outlived the specific routes
-# would send pod traffic to the UDM.
-
-# The node's attached networks, taken from the interfaces rather than written
-# down. The usual advice is not to export these to the kernel because it
-# maintains them itself — true of main, not of a table the kernel never fills.
-protocol direct attached {
-  interface "mgmt0", "ingress0", "pod0";
-  ipv4 { table master4; };
-  ipv6 { table master6; };
-}
-
-# Pod egress. GW4/GW6 are the UDM on the ingress segment, published as node
-# annotations beside the address that defines them.
-protocol static egress4 {
-  ipv4 { table master4; };
-  route 0.0.0.0/0 via GW4;
-}
-protocol static egress6 {
-  ipv6 { table master6; };
-  route ::/0 via GW6;
-}
-
-# This node's own pod CIDR, via the interface Cilium serves it on. Originated
-# rather than read back out of the kernel: reading it back made BIRD's view
-# depend on the shape of Cilium's route, and the IPv4 one is rejected outright,
-# its next hop sitting inside the prefix it describes.
-#
-# It belongs in the table, not just in the BGP announcement. A reply from a pod
-# on another node is forwarded traffic with a pod source, so the rule matches it
-# too; without this route it would find only the default and leave for the UDM.
-# Measured — cross-node pod traffic, and with it cluster DNS, stopped.
-protocol static own4 {
-  ipv4 { table master4; };
-  route OWN_POD4 via "cilium_host";
-}
-protocol static own6 {
-  ipv6 { table master6; };
-  route OWN_POD6 via "cilium_host";
-}
-
-protocol kernel k4 {
-  kernel table 47;
-  metric 0;
-  ipv4 { table master4; import none; export all; };
-}
-protocol kernel k6 {
-  kernel table 47;
-  metric 0;
-  ipv6 { table master6; import none; export all; };
-}
-
-# iBGP over pod0. 64514 keeps clear of Cilium's 64513, which peers with the UDM
-# from ingress0 and is unrelated.
-#
-# direct is explicit because iBGP defaults to multihop, which makes the gateway
-# recursive — BIRD would try to resolve the peer in the pod RIB. strict bind
-# keeps the listener off mgmt0 and ingress0.
-#
-# Only the node's own CIDR is offered: the export filter drops everything the
-# static protocols and the direct protocol produce.
-template bgp pod4 {
-  local as 64514;
-  source address MY4;
-  direct;
-  strict bind;
-  ipv4 {
-    next hop self;
-    import all;
-    export filter { if net = OWN_POD4 then accept; reject; };
-  };
-}
-template bgp pod6 {
-  local as 64514;
-  source address MY6;
-  direct;
-  strict bind;
-  ipv6 {
-    next hop self;
-    import all;
-    export filter { if net = OWN_POD6 then accept; reject; };
-  };
-}
-
-include "/etc/bird/peers.conf";
@@ -1,115 +0,0 @@
-apiVersion: v1
-kind: ServiceAccount
-metadata:
-  name: bird
-  namespace: kube-system
---
-apiVersion: rbac.authorization.k8s.io/v1
-kind: ClusterRole
-metadata:
-  name: bird
-rules:
-  # The peer list is read from the nodes' pod0 annotations, so no address is
-  # written into this repo twice.
-  - apiGroups: [""]
-    resources: ["nodes"]
-    verbs: ["get", "list"]
---
-apiVersion: rbac.authorization.k8s.io/v1
-kind: ClusterRoleBinding
-metadata:
-  name: bird
-roleRef:
-  apiGroup: rbac.authorization.k8s.io
-  kind: ClusterRole
-  name: bird
-subjects:
-  - kind: ServiceAccount
-    name: bird
-    namespace: kube-system
---
-apiVersion: apps/v1
-kind: DaemonSet
-metadata:
-  name: bird
-  namespace: kube-system
-  labels:
-    tier: node
-    app: bird
-spec:
-  selector:
-    matchLabels:
-      app: bird
-  updateStrategy:
-    type: RollingUpdate
-  template:
-    metadata:
-      labels:
-        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.
-      priorityClassName: system-node-critical
-      # Writes host routes and peers on host addresses.
-      hostNetwork: true
-      serviceAccountName: bird
-      # Every node here is a control plane node, so this is not optional.
-      tolerations:
-        - operator: Exists
-          effect: NoSchedule
-      initContainers:
-        - name: render-config
-          # By digest, not by the dated tag: the tag is rebuilt within the day
-          # it names, and the cluster's image path keeps resolving it to the
-          # first build of that day — imagePullPolicy: Always does not help,
-          # because the tag itself resolves stale. Update this when the image
-          # is rebuilt.
-          image: gitea.renner.uno/gitea_admin/bird@sha256:3bc8500613eeb45b2d8858b5445198bae3990ecf64f6d891aa94797f5a09dc1d
-          command: ["/bin/sh", "-c"]
-          args:
-            - |
-              set -eu
-              cp /config/bird.conf /etc/bird/bird.conf
-              kubectl get node "$NODE_NAME" \
-                -o go-template-file=/config/defines.tmpl > /etc/bird/defines.conf
-              kubectl get nodes --field-selector "metadata.name!=$NODE_NAME" \
-                -o go-template-file=/config/peers.tmpl > /etc/bird/peers.conf
-              # BIRD's own parser is the check: a node whose annotation is
-              # missing renders as <no value> and fails here rather than
-              # crashlooping the daemon or forming a mesh with a hole.
-              bird -p -c /etc/bird/bird.conf
-          env:
-            - name: NODE_NAME
-              valueFrom:
-                fieldRef:
-                  fieldPath: spec.nodeName
-          volumeMounts:
-            - name: config
-              mountPath: /config
-            - name: etc-bird
-              mountPath: /etc/bird
-      containers:
-        - name: bird
-          image: gitea.renner.uno/gitea_admin/bird@sha256:3bc8500613eeb45b2d8858b5445198bae3990ecf64f6d891aa94797f5a09dc1d
-          resources:
-            requests:
-              cpu: "50m"
-              memory: "32Mi"
-            limits:
-              cpu: "200m"
-              memory: "128Mi"
-          securityContext:
-            capabilities:
-              drop: ["ALL"]
-              # NET_ADMIN writes the routes; NET_BIND_SERVICE binds 179.
-              add: ["NET_ADMIN", "NET_BIND_SERVICE"]
-          volumeMounts:
-            - name: etc-bird
-              mountPath: /etc/bird
-      volumes:
-        - name: config
-          configMap:
-            name: bird-config
-        - name: etc-bird
-          emptyDir: {}
@@ -1,9 +0,0 @@
-define MY6 = {{ index .metadata.annotations "net.renner.uno/pod0-v6" }};
-define MY4 = {{ index .metadata.annotations "net.renner.uno/pod0-v4" }};
-define GW6 = {{ index .metadata.annotations "net.renner.uno/egress-gw-v6" }};
-define GW4 = {{ index .metadata.annotations "net.renner.uno/egress-gw-v4" }};
-{{/* podCIDRs follows the cluster's podSubnets order, v4 first. A flip puts an
-     IPv6 prefix into an ipv4 protocol, which fails the parse in the init
-     container rather than passing silently. */}}
-define OWN_POD4 = {{ index .spec.podCIDRs 0 }};
-define OWN_POD6 = {{ index .spec.podCIDRs 1 }};
@@ -1,12 +0,0 @@
-apiVersion: kustomize.config.k8s.io/v1beta1
-kind: Kustomization
-resources:
-  - daemonset.yaml
-configMapGenerator:
-  # Hashed name, so editing any of these rolls the DaemonSet.
-  - name: bird-config
-    namespace: kube-system
-    files:
-      - bird.conf
-      - defines.tmpl
-      - peers.tmpl
@@ -1,7 +0,0 @@
-{{- range $i, $n := .items -}}
-{{- $v6 := index $n.metadata.annotations "net.renner.uno/pod0-v6" -}}
-{{- $v4 := index $n.metadata.annotations "net.renner.uno/pod0-v4" -}}
-# {{ $n.metadata.name }}
-protocol bgp peer{{ $i }}_v6 from pod6 { neighbor {{ $v6 }} as 64514; }
-protocol bgp peer{{ $i }}_v4 from pod4 { neighbor {{ $v4 }} as 64514; }
-{{ end -}}
@@ -180,17 +180,3 @@ spec:
    # the supernets here cover the LAN as well, which is the point.
    ipv4NativeRoutingCIDR: 10.0.0.0/8
    ipv6NativeRoutingCIDR: fd00::/8
-    # Pod east-west leaves the node on pod0 (VLAN 48) rather than the
-    # management VLAN. Native routing hands forwarding to the kernel;
-    # autoDirectNodeRoutes installs the remote PodCIDR routes, which is
-    # allowed because all three nodes share pod0's L2 segment.
-    #
-    # directRoutingDevice is explicit on purpose: with several addressed
-    # devices Cilium falls back to the one carrying the Kubernetes InternalIP,
-    # which is mgmt0.
-    #
-    # Encapsulation is gone, so the datapath MTU goes 1450 -> 1500.
-    routingMode: native
-    autoDirectNodeRoutes: true
-    nodePort:
-      directRoutingDevice: pod0
@@ -24,6 +24,5 @@ resources:
  - valkey.yaml
  - velero-plugin-for-aws.yaml
  - vsftpd.yaml
-  - wiki-index.yaml
  - woodpecker.yaml
  - wyoming.yaml
@@ -1,26 +0,0 @@
-apiVersion: image.toolkit.fluxcd.io/v1
-kind: ImageRepository
-metadata:
-  name: wiki-index
-  namespace: flux-system
-spec:
-  image: gitea.renner.uno/gitea_admin/wiki-index
-  interval: 10m
---
-# build-ci-images.yml publishes `build.<unix-ts>` alongside the dated tag so
-# every build rolls out on its own; the dated tag is rewritten several times a
-# day and cannot be tracked.
-apiVersion: image.toolkit.fluxcd.io/v1
-kind: ImagePolicy
-metadata:
-  name: wiki-index
-  namespace: flux-system
-spec:
-  imageRepositoryRef:
-    name: wiki-index
-  filterTags:
-    pattern: '^build\.(?P<ts>\d+)$'
-    extract: '$ts'
-  policy:
-    numerical:
-      order: asc
@@ -7,7 +7,6 @@ resources:
  - cluster-settings/
  - cilium/
  - multus/
-  - bird/
  - cert-manager/
  - snapshot-controller/
  - metrics-server/
@@ -126,59 +126,3 @@ spec:
            - { type: 135, family: IPv6 }
            - { type: 136, family: IPv6 }
            - { type: 143, family: IPv6 }
---
-# wiki-embed is the one ai pod Traefik may reach: it sits behind BasicAuth
-# (apps/llama-server/wiki-embed-ingress.yaml), not behind the Authentik gate.
-# Scoped to this pod so the default-deny above keeps Traefik off llama-server.
-apiVersion: cilium.io/v2
-kind: CiliumNetworkPolicy
-metadata:
-  name: wiki-embed-from-traefik
-  namespace: ai
-spec:
-  endpointSelector:
-    matchLabels:
-      app.kubernetes.io/name: wiki-embed
-  ingress:
-    - fromEndpoints:
-        - matchLabels:
-            "k8s:io.kubernetes.pod.namespace": ingress
-            app.kubernetes.io/name: traefik
-      toPorts:
-        - ports:
-            - port: "8080"
-              protocol: TCP
---
-# wiki-index: Traefik (LAN clients behind BasicAuth) and Gitea (push webhook)
-# may reach it; it reaches OpenBao for its credentials. Gitea itself is reached
-# through Traefik, which the namespace egress above already allows.
-apiVersion: cilium.io/v2
-kind: CiliumNetworkPolicy
-metadata:
-  name: wiki-index
-  namespace: ai
-spec:
-  endpointSelector:
-    matchLabels:
-      app.kubernetes.io/name: wiki-index
-  ingress:
-    - fromEndpoints:
-        - matchLabels:
-            "k8s:io.kubernetes.pod.namespace": ingress
-            app.kubernetes.io/name: traefik
-        - matchLabels:
-            "k8s:io.kubernetes.pod.namespace": apps
-            app.kubernetes.io/name: gitea
-      toPorts:
-        - ports:
-            - port: "8080"
-              protocol: TCP
-  egress:
-    - toEndpoints:
-        - matchLabels:
-            "k8s:io.kubernetes.pod.namespace": openbao
-            app.kubernetes.io/name: openbao
-      toPorts:
-        - ports:
-            - port: "8200"
-              protocol: TCP
@@ -45,16 +45,6 @@ spec:
        - ports:
            - port: "1883"
              protocol: TCP
-    # gitea delivers the notes repo's push webhook to wiki-index (ai ns) on :8080.
-    # wiki-index's own ingress policy scopes this to the gitea pod.
-    - toEndpoints:
-        - matchLabels:
-            "k8s:io.kubernetes.pod.namespace": ai
-            app.kubernetes.io/name: wiki-index
-      toPorts:
-        - ports:
-            - port: "8080"
-              protocol: TCP
    - toEntities:
        - host
        - remote-node
@@ -31,16 +31,6 @@ spec:
        - ports:
            - port: "8200"
              protocol: TCP
-    # wiki-index (ns ai) logs in via auth/kubernetes/role/wiki-index and reads
-    # its Gitea token and webhook secret on every sync.
-    - fromEndpoints:
-        - matchLabels:
-            "k8s:io.kubernetes.pod.namespace": ai
-            app.kubernetes.io/name: wiki-index
-      toPorts:
-        - ports:
-            - port: "8200"
-              protocol: TCP
    # (Secret-rotation CronJobs run in THIS namespace since 2026-07-18, so their
    # OpenBao access is covered by the intra-namespace ingress rule above — no
    # per-app rule needed.)
@@ -1,21 +0,0 @@
-# OpenBao policy: wiki-index
-#
-# The wiki-index Deployment (namespace ai) reads its Gitea token and the
-# webhook HMAC secret on every use. Read-only on that one path.
-#
-# NOT applied by Flux. Apply ONCE with an admin token:
-#
-#   bao policy write wiki-index kubernetes/infrastructure/openbao/policies/wiki-index.hcl
-#   bao write auth/kubernetes/role/wiki-index \
-#     bound_service_account_names=wiki-index \
-#     bound_service_account_namespaces=ai \
-#     policies=wiki-index ttl=15m
-#
-# Secret: homelab/apps/ai/wiki-index, keys `gitea_token` (read:repository on
-# the notes repo) and `webhook_secret` (the Gitea webhook's secret).
-#
-# Do NOT set token_no_default_policy=true (kv-v2 preflight needs the default policy).
-
-path "homelab/data/apps/ai/wiki-index" {
-  capabilities = ["read"]
-}
@@ -14,7 +14,6 @@ spec:
  # ClusterSecretStore" and the secret never syncs. Keep this list sorted.
  conditions:
    - namespaces:
-        - ai # wiki-embed-basic-auth (apps/llama-server/wiki-embed-ingress.yaml)
        - apps
        - audiobookshelf
        - authentik
@@ -126,16 +126,6 @@ spec:
    addressRanges:
      cluster:
        - 10.99.0.0/29
-  # Reads go to the OSD in the client's own CRUSH location. crushLocationLabels
-  # is unset, so Rook's default list applies; of those the nodes carry
-  # topology.kubernetes.io/zone, one zone per node, matching the zone buckets
-  # the OSDs are created under.
-  #
-  # Passed as a krbd/mount option at map time, so a volume keeps its current
-  # behaviour until it is remounted.
-  csi:
-    readAffinity:
-      enabled: true
  # Nodes listed explicitly, not useAllNodes: under useAllNodes the operator
  # re-provisions an OSD on any node whose device still carries a BlueStore
  # signature, so purging an OSD before pulling its drive is impossible — it
@@ -52,12 +52,6 @@ machine:
    # this label instead of listing hostnames, so which boxes carry storage is
    # stated once, here, by the machine that knows.
    storage.renner.uno/ceph-osd: "true"
-  # Publishes this node's pod0 addresses, declared below in its VLANConfig, so
-  # the bird DaemonSet reads its peers from the node list rather than from a
-  # copy. Annotations, not labels: a label value cannot contain a colon.
-  nodeAnnotations:
-    net.renner.uno/pod0-v6: fd02:48::11
-    net.renner.uno/pod0-v4: 10.2.48.11
  udev:
    # Renames every port this node uses to its role at boot, matched by
    # burned-in MAC — kernel names derive from PCI topology and shift when
@@ -188,13 +182,8 @@ addresses:
 # network's clients selectively, so the client goes rather than the record.
 # DHCP is off on VLAN 47 for the same reason.
 #
-# No route here. Table 47 belongs to bird, which fills it with this node's
-# attached networks, the peers' pod CIDRs and the egress default, and empties it
-# if it stops — leaving the rule to fall through to main. A default declared
-# here would outlive the specific routes and send pod traffic to the UDM.
-#
-# accept_ra_defrtr stays 0 on this link so the RA default does not land in main,
-# where mgmt0 owns it.
+# The metric loses to mgmt0's default 1024 deliberately: this pass only adds
+# the segment. The default route moves here in the change that flips it.
 apiVersion: v1alpha1
 kind: VLANConfig
 name: ingress0
@@ -203,6 +192,9 @@ parent: uplink
 up: true
 addresses:
  - address: 10.2.47.11/24
+routes:
+  - gateway: 10.2.47.1
+    metric: 2048
 ---
 # Dedicated Ceph cluster network on X710 port 1, VLAN 49, untagged on the
 # wire. OSD-to-OSD replication and nothing else. The udev rule in the machine
@@ -33,10 +33,6 @@ machine:
    topology.kubernetes.io/zone: saturn
    # Holds a Ceph OSD drive — see k8s-1.yaml.
    storage.renner.uno/ceph-osd: "true"
-  # pod0 addresses for the bird peer list — see k8s-1.yaml.
-  nodeAnnotations:
-    net.renner.uno/pod0-v6: fd02:48::12
-    net.renner.uno/pod0-v4: 10.2.48.12
  udev:
    # Renames every port to its role at boot — see k8s-1.yaml.
    rules:
@@ -132,6 +128,9 @@ parent: uplink
 up: true
 addresses:
  - address: 10.2.47.12/24
+routes:
+  - gateway: 10.2.47.1
+    metric: 2048
 ---
 # Dedicated Ceph cluster network on X710 port 1, VLAN 49, untagged on the
 # wire. OSD-to-OSD replication only, not routed off its own segment.
@@ -29,10 +29,6 @@ machine:
    topology.kubernetes.io/zone: neptune
    # Holds a Ceph OSD drive — see k8s-1.yaml.
    storage.renner.uno/ceph-osd: "true"
-  # pod0 addresses for the bird peer list — see k8s-1.yaml.
-  nodeAnnotations:
-    net.renner.uno/pod0-v6: fd02:48::13
-    net.renner.uno/pod0-v4: 10.2.48.13
  udev:
    # Renames every port to its role at boot — see k8s-1.yaml.
    rules:
@@ -119,6 +115,9 @@ parent: uplink
 up: true
 addresses:
  - address: 10.2.47.13/24
+routes:
+  - gateway: 10.2.47.1
+    metric: 2048
 ---
 # Dedicated Ceph cluster network — see k8s-1.yaml. The node holds NO address
 # here on purpose; the link is handed whole to the OSD pod (host-device).
@@ -7,51 +7,32 @@
 # 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.
+    # Pins the node IP to pod0 (VLAN 48), which is what carries pod east-west.
+    # Cilium reaches a peer's pods at that peer's Kubernetes InternalIP and
+    # offers no way to point it elsewhere: --ipv4-node/--ipv6-node are accepted
+    # and then overwritten from the Node object, and directRoutingDevice only
+    # selects the NodePort/DSR device. So the node IP is the only control over
+    # which segment east-west uses, in both tunnel and native routing mode.
+    #
+    # Unpinned, kubelet takes the first non-loopback address, which could land
+    # on the v30iot NIC (ens19, 10.2.30.0/24).
+    #
+    # fd02:48::/64 is listed so the Node object carries an IPv6 InternalIP at
+    # all. Without one, Cilium detects the node's IPv6 itself and picks the
+    # SLAAC address from the dynamic provider prefix, then probes it until the
+    # agent restarts (#789).
    nodeIP:
      validSubnets:
-        - 10.2.40.0/24
-        - fd40::/64
-  # 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.
-  nodeAnnotations:
-    net.renner.uno/egress-gw-v6: fd02:47::1
-    net.renner.uno/egress-gw-v4: 10.2.47.1
---
-# All pod traffic goes to table 47, and the table decides — longest prefix
-# match, no exceptions to carve out. bird fills it with the node's attached
-# networks, the peers' pod CIDRs and the egress default, so a packet to a node
-# on the mgmt segment takes the connected route, a packet to a pod on another
-# node takes pod0, and anything else takes the default over ingress0.
-#
-# Matching the source is what keeps the node's own traffic out: it answers on
-# mgmt0, and a reply leaving over ingress0 would carry an mgmt0 source onto
-# VLAN 47. The routing decision runs before Cilium's masquerade hook, so the
-# rule sees the pod address and the masquerade then rewrites it.
-#
-# If bird stops, table 47 empties, the lookup misses and the rule falls through
-# to main — pod traffic returns to VLAN 40, unseparated rather than misrouted.
-#
-# Priorities are below main (32766) and clear of the rule Cilium installs at 9.
-# The two families are separate rule tables, so the priorities need not match;
-# the document name is the priority and must be unique.
-apiVersion: v1alpha1
-kind: RoutingRuleConfig
-name: "1000"
-src: fd10:244::/48
-table: "47"
---
-apiVersion: v1alpha1
-kind: RoutingRuleConfig
-name: "1001"
-src: 10.244.0.0/16
-table: "47"
+        - 10.2.48.0/24
+        - fd02:48::/64
+cluster:
+  etcd:
+    # Pins etcd to pod0 with the node IP. Unset, etcd advertises "the first
+    # routable address of the node", which is not deterministic across a
+    # re-address. listenSubnets defaults to this.
+    #
+    # pod0 is a closed 10G L2 between the three nodes. The management VLAN is
+    # about to become a single 1G NIC shared with everything else management
+    # does, which is the wrong link for quorum traffic.
+    advertisedSubnets:
+      - 10.2.48.0/24