Phase 01 -- Bundle Deploy

Deploy the hardened bundle + the octavia-pki overlay onto the freshly-prepped MAAS machines, and verify it settles to the expected PRE-vault-init state (zero errors, vault awaiting init, the TLS consumers awaiting vault certs). Vault init is phase-02.

Decisions: B5 (IP-only), D-019 (no designate -- SUPERSEDED 2026-07-10 by D-106, see the callout below), D-020 (dual provider+metal VIPs), R14 (VIPs front-loaded to .50-.60), Section-G NIC bindings. Troubleshooting: appendix-A -- R14 (VIP relocation), R15 (.10 phantom resolver), L1 (no set -e on count-gate blocks), L3 (metal-side dual-VIP eyeball check), DOCFIX-016 (maas list leak).

!!! bundle.yaml CHANGED 2026-07-10 (DOCFIX-167, VR1 DC-DC buildout, D-106/ D-108 both ADOPTED): designate/designate-bind/designate-mysql-router/ designate-hacluster and cinder-backup/ceph-rbd-mirror are now IN the same bundle.yaml this phase deploys -- the counts and "NO designate" GATE text below are updated to match. Genuinely unresolved, flagged rather than decided: this is the SAME shared bundle file used for both a plain VR0/DC0 single-DC rehearsal redeploy and for standing up VR1's DC1 (per the buildout design's own framing, VR0's testcloud IS becoming DC1, not staying a separate thing) -- so ANY phase-01 run from this point forward gets these new charms, whether the intent is a VR0-only continuation or a DC-DC DC1 build. If a future session needs a plain VR0-only redeploy WITHOUT Designate/cinder-backup/ceph-rbd-mirror, that needs an explicit decision (a bundle fragment/override removing them for that one run) -- do not assume either interpretation silently.

Prerequisites (must be true entering phase-01)

phase-00 done: 4 machines Ready/power=off; MAAS carve applied (front-loaded VIP /26 reserved, FIP pool reserved, stale iprange gone); enp8s0 data NIC linked on ALL four hosts; OSD /dev/vdb wiped blank.
overlays/${DC}-octavia-pki.yaml present (Step 1.0). PER-DC since F1 (2026-07-29) -- export DC (region-qualified, e.g. vr1-dc0) before running any step in this phase.
Hardened bundle.yaml in the working dir (channels pinned; VIPs .50-.60; reserved-host-memory 8192; image-conversion; use-policyd-override).

GATE (DOCFIX-073) -- run THE pre-deploy gate before add-model:

bash scripts/preflight.sh

Expect PREFLIGHT: PASS (repo lint + bundle invariants + Charmhub channel assert + live MAAS pre-flight, aggregated). WARN = review then decide; FAIL = do not deploy. This is the single entry point that replaced the six separate gate artifacts an operator previously had to remember.

Constants and env-literals

Host identity / system_ids: resolve LIVE via scripts/lib-hosts.sh (DOCFIX-040: system_ids are re-minted on every reenrollment -- a documented id is stale by definition; never paste one into a command).
Planes / CIDRs / VIP bands: scripts/lib-net.sh is the single authority (D-052/D-053). Resolve MAAS subnets BY CIDR (PATTERN-1), never by numeric subnet id -- ids drift across MAAS lifecycles. (DOCFIX-068 removed a stale pre-D-052 literal map that sat here.)
expected plan (as of bundle.yaml post-DOCFIX-167, 2026-07-10): 56 apps, 108 relations, 4 machines (bundle 8/9/10/11 -> juju 0/1/2/3), 24 LXD. (Was 50 apps/97 relations before the Designate + cinder-backup + ceph-rbd-mirror reactivation -- see the callout above.)

Run-location legend

# RUN: jumphost -- juju + maas admin (MAAS profile is admin; never maas list -- DOCFIX-016).

Command-label convention

Every command block below is bracketed by bold labels, so a command line is never mistaken for surrounding prose (these render in GitBucket and read clearly in a raw editor):

RUN -- LOC -- the block CHANGES state; run it at LOC (e.g. jumphost, vault/0, jumphost -> magnum/0).
CHECK (read-only) -- LOC -- a read-only verification; safe to re-run.
GATE: -- a hard stop; do NOT proceed past the block unless the stated condition holds.
Expect: -- what a passing result looks like.
> CAUTION: -- marks a destructive, secret-handling, or irreversible step.

Step 1.0 -- Octavia PKI overlay (secret-handling prereq) DISCRETE

overlays/${DC}-octavia-pki.yaml carries the 5 lb-mgmt-* PKI keys (controller CA/cert, issuing CA key+passphrase+cert). It is the ONLY overlay in the deploy command and is secret-safe + ASCII. PRIMARY path: reuse the existing validated overlay (the CAs are 10y, so it survives rebuilds). REGENERATION path (fresh CAs): run the discrete secret procedure inlined as "Step 1.0-GEN" at the end of this phase. Either way, confirm the overlay parses and contains exactly the 5 keys (sanity block below) before deploying.

CHECK (read-only) -- jumphost

# RUN: jumphost -- sanity only (does NOT print key material)
DC="${DC:?export DC=vr1-dc0|vr1-dc1 -- the PKI overlay is per-DC (F1)}"
OVL="overlays/${DC}-octavia-pki.yaml"
# Assert on CONTENT, not existence: an absent file, a wrong key count and a non-ASCII
# byte are three DIFFERENT answers and none of them may read as OK.
if [ ! -f "$OVL" ]; then
  echo "MISSING $OVL -- generate it with Step 1.0-GEN for DC=$DC"
else
  K=$(grep -cE 'lb-mgmt-' "$OVL" || true)
  [ "$K" -eq 5 ] && echo "OK 5 lb-mgmt-* keys" || echo "FAIL $K lb-mgmt-* keys (want 5)"
  LC_ALL=C grep -qP '[^\x00-\x7F]' "$OVL" && echo "FAIL non-ASCII bytes" || echo "OK ASCII clean"
fi

Step 1.1 -- Pre-deploy verify (read-only; 4 checks)

One consolidated read-only block. NO set -e (a guarded count of 0 is a valid answer, not a failure -- appendix-A: L1); count greps are || true.

CHECK (read-only) -- jumphost

( {
  echo "=== CHECK 1: bundle VIPs (quote-tolerant, octet-anchored) ==="
  grep -nE '^[[:space:]]+vip:' bundle.yaml overlays/vr1-dc0-vips.yaml
  TOT=$(grep -hcE '^[[:space:]]+vip:[[:space:]]*"?10\.12\.4\.' bundle.yaml overlays/vr1-dc0-vips.yaml 2>/dev/null | awk '{s+=$1} END{print s+0}')
  HI=$(grep -hcE '^[[:space:]]+vip:[[:space:]]*"?10\.12\.4\.(5[0-9]|6[0-2])("|$|[[:space:]])' bundle.yaml overlays/vr1-dc0-vips.yaml 2>/dev/null | awk '{s+=$1} END{print s+0}')
  LO=$(grep -hcE '^[[:space:]]+vip:[[:space:]]*"?10\.12\.4\.(1[0-9]|20)("|$|[[:space:]])' bundle.yaml overlays/vr1-dc0-vips.yaml 2>/dev/null | awk '{s+=$1} END{print s+0}')
  echo "  provider VIPs total=$TOT  in .50-.60=$HI  in .10-.20(stale)=$LO   (want 13/13/0)"
  # RULING 3 (2026-07-25): the counts span bundle.yaml AND its per-DC VIP overlay,
  # because the base is now VIP-FREE. `grep -c` over two files prints one count PER
  # FILE, so these sum with awk -- a bare $( ) would capture a two-line string and
  # every numeric test below would silently fail.
  # metal side is the second token of each dual vip; eyeball that all 11 are .8.50-.60,
  # clear of metal infra .8.10(maas)/.8.20(lxd)/.8.21(capi)/.8.30(juju) -- appendix-A: L3.

  echo "=== CHECK 2: enp8s0 data NIC linked on ALL FOUR hosts (10.12.12.0/22) ==="
  for SID in 4na83t qdbqd6 h8frng tmsafc; do
    echo -n "  $SID: "
    maas admin interfaces read "$SID" | jq -r '.[] | select(.name=="enp8s0")
      | [.links[]? | select(.subnet.cidr=="10.12.12.0/22") | .ip_address] | join(",")'
  done   # expect 10.12.12.40 / .41 / .42 / .43 (select by .subnet.cidr -> robust to id drift)

  echo "=== CHECK 3: subnet DNS resolvers ==="
  for ID in 1 2 6 7 8 9; do maas admin subnet read "$ID" | jq -c '{id,cidr,dns_servers}'; done
  # expect subnet 1 (provider) -> [10.12.4.1]; 2/6/7/8/9 -> [10.12.8.1]

  echo "=== CHECK 4a: nodes Ready / power off ==="
  maas admin machines read | jq -r '.[] | select(.system_id|IN("4na83t","qdbqd6","h8frng","tmsafc"))
    | "\(.hostname) \(.status_name) power=\(.power_state)"'
} )

CHECK (read-only) -- jumphost

# CHECK 4b: OSD /dev/vdb blank (DOCFIX-027 -- LOCAL libvirt-host loop, NOT ssh: the four
# hosts are Released/powered-off entering phase-01, and /var/lib/libvirt/images is a
# hypervisor (jumphost) path that does not exist on the hosts. RUN: jumphost (libvirt host; sudo).
for h in openstack0 openstack1 openstack2 openstack3; do
  echo "== $h =="
  sudo qemu-img info "/var/lib/libvirt/images/${h}-1.qcow2" | grep -E 'virtual size|disk size'
done   # expect virtual 512 GiB, disk ~200 KiB (sparse/blank)

GATE: VIPs 11/11/0; enp8s0 linked on all 4; subnet DNS as above; 4 nodes Ready; OSD blank.

Step 1.2 -- Dry-run (guarded)

Refuse to add a model if openstack already exists; require the overlay.

RUN -- jumphost

( {
  juju models 2>&1 | tee /tmp/jmodels.txt
  if grep -qE '(^|[[:space:]]|/)openstack([[:space:]*]|$)' /tmp/jmodels.txt; then
    echo "ABORT: an 'openstack' model already exists (teardown is phase-00)"; 
  elif [ ! -f "overlays/${DC:?export DC first}-octavia-pki.yaml" ]; then
    echo "ABORT: overlays/${DC}-octavia-pki.yaml missing (Step 1.0-GEN for DC=$DC)";
  else
    juju add-model openstack
    juju deploy ./bundle.yaml --overlay "overlays/${DC}-vips.yaml" \
      --overlay "overlays/${DC}-octavia-pki.yaml" -m openstack --dry-run
  fi
} )

GATE: (from the plan): 56 apps, 108 relations, 4 machines (8/9/10/11 -> 0/1/2/3), 24 LXD; ceph-osd/0-3 one per node; nova-compute/0-2 on machines 1/2/3 ONLY (machine 0 = OSD+LXD host, no compute); channels match the matrix; relations include octavia:certificates - vault:certificates, vault:shared-db - vault-mysql-router, mysql-innodb-cluster:certificates - vault:certificates; NO vault:ha (still absent, unchanged); designate/designate-bind/designate-mysql-router/designate- hacluster and cinder-backup/ceph-rbd-mirror NOW PRESENT (D-106/D-108, 2026-07-10 -- corrects this GATE's old "NO designate (D-019)" text, since D-019 is superseded). Only the two benign R11 warnings (L34 name, L55 variables).

Step 1.3 -- Deploy (VIP-guarded)

Re-run the VIP guard inline (the dry-run never echoes vip values), then deploy only if the guard exits 0.

THE GUARD IS NOW THE MAINTAINED CHECKER, NOT A HAND-ROLLED GREP (2026-07-29). This block used to compute a 13/13/0 triple with three grep -hcE counts anchored on the literal 10\.12\.4\.. Three defects, all measured: it hard-ABORTED on vr1-dc1, whose provider band is 10.12.64.x; it counted only IPv4, so after R2's dual-stack ADD the ruled v6 legs were checked by NOTHING (chain-audit finding 22); and its prose said "11/11/0" while its code demanded 13, a drift that appeared the moment R11 added vault .61 and designate .62. scripts/provider-bundle-check.py already encodes every one of those invariants -- per-DC bands, the VIP count, and since 2026-07-28 the dual-family ARITY coupling -- so the runbook defers to it instead of keeping a second copy that drifts. One source of truth; a new ruling updates the checker and this step inherits it.

RUN -- jumphost

( {
  DC="${DC:?export DC=vr1-dc0|vr1-dc1 first}"
  set -- ./bundle.yaml
  for OVL in "overlays/${DC}-vips.yaml" "overlays/${DC}-machines.yaml" \
             "overlays/${DC}-octavia-pki.yaml"; do
    [ -f "$OVL" ] && set -- "$@" --overlay "$OVL"
  done
  # The octavia-pki overlay is REQUIRED for deploy (it carries the 5 lb-mgmt-* keys) --
  # absence is an ABORT, not a silently-skipped overlay.
  [ -f "overlays/${DC}-octavia-pki.yaml" ] || {
    echo "ABORT: overlays/${DC}-octavia-pki.yaml missing -- run Step 1.0-GEN for DC=$DC"; exit 1; }

  echo "validating the MERGED $DC deploy input: $*"
  if python3 scripts/provider-bundle-check.py "$@" --dc "$DC"; then
    juju deploy "$@" -m "${DC_MODEL:?export DC_MODEL first}"
  else
    echo "ABORT: VIP/bundle invariants failed for $DC -- fix before deploying"; exit 1
  fi
} )

CONVERGENCE WATCH (ENHANCEMENT-1): keep two windows open for the whole deploy arc.

Window 1 (detail): juju status -m openstack --watch 5s (always explicit -m; on a 50-app model the table exceeds one screen, so this is a slice -- optionally filter, e.g. ... 'mysql*' vault).
Window 2 (signal): scripts/deploy-watch.sh openstack 15 -- compact machine/unit state counts
- named error/blocked units. Health-at-a-glance: the error/blocked section stays EMPTY until the expected late blocks (vault needs-init, octavia awaiting-configure). Neither window descends into subordinates; neither replaces the phase gates.

Step 1.4 -- DNS gate during deploy (as machines come up)

Run when machine 0 reaches started, then per LXD unit as they appear (flag BEFORE the target; logic inside the remote quotes; no outer 2>/dev/null):

CHECK (read-only) -- jumphost

juju ssh -m openstack 0 -- 'resolvectl status | grep -i "DNS Server"; getent hosts api.snapcraft.io && echo OK || echo FAIL'
# repeat for ceph-mon/0, mysql-innodb-cluster/0 as they appear

GATE: each returns OK (api.snapcraft.io resolves -> the snap install storm proceeds clean). FINDING (non-blocking, R15): the unreachable region resolver 10.12.8.10 (MAAS region/rack controller, advertised on the metal VLAN independent of the subnet field) may still appear in a node's resolver list -- resolution succeeds because systemd-resolved deprioritizes .10 and falls through to .1. Latent fragility if .1 ever drops; understand/eliminate for Roosevelt. (appendix-A: R15.)

EXIT GATE (phase-01 complete)

Deploy settled to the PRE-vault-init end state:
- ZERO units in error.
- mysql-innodb-cluster x3 ACTIVE ("Cluster is ONLINE").
- vault/0 BLOCKED "Vault needs to be initialized" (the phase-02 trigger, not a fault).
- Waiting on vault certs (expected pre-init): ovn-central x3, ovn-chassis x3 (incl nova-compute subordinates), ovn-chassis-octavia, neutron-api-plugin-ovn, barbican-vault.
- octavia BLOCKED "Awaiting configure-resources" (D-021); gss unknown (pre-run).
- NOT YET OBSERVED LIVE (2026-07-10, D-106/D-108 reactivation) -- add to this checklist once a real deploy confirms it: designate and designate-bind have a certificates: metal-internal relation to vault (per bundle.yaml), so BY THE SAME PATTERN as the apps above they are LIKELY also waiting on vault certs pre-init -- inferred from the relation graph, not confirmed against a real deploy. cinder-backup (subordinate) and ceph-rbd-mirror have no certificates relation, so no equivalent expectation for them. If a real deploy shows anything different for designate/designate-bind, correct this list rather than trusting this inference.
- magnum/0 BLOCKED "Ports which should be open, but are not: 9501" -- pre-vault posture: magnum-api is loopback-bound ([api] host not yet templated) and haproxy backends target unit IPs. EXPECTED phase-01 end-state; self-resolves at the phase-02 cert rollout (apache2 takes *:9501). Confirmed self-resolving 2026-06-12 (FINDING-2); verify in the phase-02 post-init sweep.
- keystone/0 "PO (broken): Unit is ready" -- expected while use-policyd-override=true with no policy zip attached (FINDING-1); keystone runs the DEFAULT policy. No mutation this arc.
Section-G NIC payoff confirmed (no subset/binding errors): ceph-mon -> storage 10.12.16.x; octavia -> data 10.12.12.1; nova-compute -> data 10.12.12.4x; vault -> metal 10.12.8.x.
Proceed to phase-02 (vault init).

As-built reference (2026-06-03 second redeploy -- audit trail)

juju deploy ./bundle.yaml --overlay overlays/octavia-pki.yaml -m openstack on maas/default (cred maas-api).
Plan: 50 apps / 97 relations / 4 machines / 24 LXD; placement as above.
Pre-deploy verify: VIPs 11/11/0; enp8s0 -> 10.12.12.40-43 (all 4); subnet DNS as above; nodes Ready; OSD blank.
Settled: zero errors; mysql /0 R/W (10.12.8.173), /1 (.179) /2 (.185) R/O; vault blocked needs-init.

Balance / stability observations (Roosevelt rebalancing inputs -- post-deploy item 6)

Quorum triads (mysql-innodb-cluster, ovn-central, ceph-mon) all on machines 0/1/2: correct anti-affinity; machine 3 loss breaks no quorum; any single loss of 0/1/2 leaves a 2-of-3 majority.
Machine 0 = no-compute control host, largest container count: prefigures the Roosevelt role split.
FLAG: machine 3 concentrates six singletons (vault, glance, nova-cloud-controller, octavia, placement, barbican) + compute + OSD. Acceptable on testcloud; Roosevelt answer is role split + HA, informed by measured footprints at the 32 GiB envelope (DEVIATION-2 caveat).
rabbitmq-server single unit: messaging SPOF, as designed for v1.

PATTERN-1 (standing convention) -- dynamic lookup vs. pinned identifiers

READ/VERIFY ops discover values at runtime (never hardcode what resolves: hostname->system_id via maas admin machines read | jq; subnet id by CIDR). DESTRUCTIVE/IRREVERSIBLE ops discover dynamically, ASSERT against a pinned EXPECTED set, ABORT on mismatch, then operate on the pinned values (a filter bug or an unexpected new machine must not become collateral damage). Retrofit candidates (apply as fixture-tested gated blocks, NOT bulk edits): phase-00 release/host loops -> discover-assert-pin; subnet ids -> resolve by CIDR; octet maps -> derive from hostname index. Canonical statement in runbooks/README.md.

phase-02 -- vault bring-up.

Step 1.0-GEN -- Octavia management-PKI generation (regeneration path) DISCRETE / SECRET

CAUTION: SECRET step -- generates Octavia CA private keys + passphrases. Do NOT echo or log key material (only cert dates/subjects + verify-OK are printed). The overlay it writes is gitignored.

Run ONLY if this DC has no PKI yet -- 1.0-GEN.0 REFUSES if overlays/${DC}-octavia-pki.yaml or this DC's workspace already exists, because regenerating invalidates every amphora already issued against that CA. Each DC is generated separately. Produces the two-tier EC PKI for Charmed Octavia's amphora trust domain and writes the overlay. Decisions (Workstream 3a, 2026-05-22): fresh generation; EC P-384 CAs (SHA-384, 10y); EC P-256 controller cert (2y); overlay-file distribution (gitignored); artifacts under $HOME/octavia-pki/; passphrases = 32 random bytes base64 (44 chars). SECRET step -- do NOT echo key material; the only printed values are cert dates/subjects and verify OK.

The five octavia charm options the overlay sets:

lb-mgmt-issuing-cacert = base64(issuing CA cert)
lb-mgmt-issuing-ca-private-key = base64(issuing CA ENCRYPTED key)
lb-mgmt-issuing-ca-key-passphrase = the issuing CA passphrase (PLAIN string, NOT base64)
lb-mgmt-controller-cacert = base64(controller CA cert)
lb-mgmt-controller-cert = base64(controller cert + key, concatenated)

1.0-GEN.0 -- select the DC, then the workspace (openssl 3.x; $HOME only -- snap home-confinement, never /tmp)

EVERY PATH IN THIS STEP IS PER-DC (F1, 2026-07-29). The D-109 amendment rules "each DC gets its own Octavia CA; no cross-DC amphora root-of-trust". The 2026-07-27 R7 work parameterised the CA SUBJECT and the VIP gate but NOT the paths, so the workspace and the overlay were single fixed locations: generating a second DC's PKI overwrote the first DC's issuing-CA key, controller-CA key and both passphrases, and left the one fixed-name overlay carrying the wrong DC's CA -- which a later redeploy of the first DC would then apply to it. Set DC ONCE here; everything below derives from it, paths included.

RUN -- jumphost

# RUN: jumphost
export DC="${DC:-vr1-dc0}"          # vr1-dc0 | vr1-dc1 -- region-qualified, never bare dcN
export DC_LABEL="$(printf '%s' "$DC" | tr 'a-z-' 'A-Z ')"   # e.g. "VR1 DC0"
export REPO="${REPO:?REPO not set -- export REPO=<path to your clone> (see runbooks/README.md Conventions)}"
export VIP_OVERLAY="$REPO/overlays/${DC}-vips.yaml"
export OCTAVIA_PKI_OVERLAY="$REPO/overlays/${DC}-octavia-pki.yaml"
[ -r "$VIP_OVERLAY" ] || { echo "ABORT: no VIP overlay at $VIP_OVERLAY -- is DC=$DC correct?"; exit 1; }

WORKDIR="$HOME/octavia-pki/$DC"     # PER-DC (F1) -- never $HOME/octavia-pki directly
mkdir -p "$WORKDIR"/issuing-ca "$WORKDIR"/controller-ca "$WORKDIR"/controller
chmod 700 "$HOME/octavia-pki" "$WORKDIR"
openssl version    # expect OpenSSL 3.x
echo "DC=$DC  WORKDIR=$WORKDIR  overlay=$OCTAVIA_PKI_OVERLAY"

REFUSE-IF-PRESENT gate -- run before generating anything. Regeneration destroys the CA this DC's already-deployed amphorae trust, so it must be a deliberate act, not the default outcome of re-running a step.

# RUN: jumphost
if [ -e "$WORKDIR/issuing-ca/issuing-ca.key.enc" ] || [ -e "$OCTAVIA_PKI_OVERLAY" ]; then
  echo "REFUSING: PKI material already exists for DC=$DC."
  echo "  workspace: $WORKDIR"
  echo "  overlay:   $OCTAVIA_PKI_OVERLAY"
  echo "Regenerating invalidates every amphora already issued against this CA."
  echo "To deliberately regenerate: move the existing material aside first, gated."
  exit 1
fi
echo "OK: no existing PKI for DC=$DC -- safe to generate"

1.0-GEN.a -- Issuing CA (EC P-384, AES-256 encrypted key, self-signed 10y)

RUN -- jumphost

R7 (D-109 AMENDMENT, RULED 2026-07-27) -- this generator is PER-DC. It used to bake VR0 DC0 into both CA subjects and gate the VIP on ^10\.12\.4\., so it could only ever produce a dc0 artifact -- and phase-01 hard-ABORTs without the overlay, which made every other DC undeployable. Reuse across DCs was refused on posture: the overlay carries CA private keys plus a plaintext passphrase in a repo SEC-004 records as PUBLIC, so one shared amphora CA across two clouds D-100 defines as independent would widen an existing exposure. DC, DC_LABEL, VIP_OVERLAY and OCTAVIA_PKI_OVERLAY are exported ONCE in 1.0-GEN.0 above -- along with the per-DC WORKDIR that F1 added -- and every block below derives from them.

( {
  WORKDIR="$HOME/octavia-pki/${DC:?DC not set -- run 1.0-GEN.0 first}"
  cd "$WORKDIR/issuing-ca" || exit 1   # dir from 1.0-GEN.0
  openssl rand -base64 32 | tr -d '\n' > passphrase.txt
  chmod 600 passphrase.txt
  test "$(wc -c < passphrase.txt)" -eq 44 || { echo "ABORT: issuing passphrase length != 44"; exit 1; }
  openssl genpkey -algorithm EC -pkeyopt ec_paramgen_curve:P-384 \
    -aes-256-cbc -pass file:passphrase.txt -out issuing-ca.key.enc
  chmod 600 issuing-ca.key.enc
  openssl req -new -x509 -sha384 -key issuing-ca.key.enc -passin file:passphrase.txt \
    -days 3650 -subj "/CN=${DC_LABEL} Omega Cloud Octavia Issuing CA/O=Neumatrix" \
    -out issuing-ca.cert.pem
  openssl x509 -in issuing-ca.cert.pem -noout -dates -subject
  openssl verify -CAfile issuing-ca.cert.pem issuing-ca.cert.pem    # expect: OK
} )

1.0-GEN.b -- Controller CA (EC P-384, AES-256 encrypted key, self-signed 10y; own passphrase)

The controller CA key is encrypted (its own passphrase) for future controller-cert rotation -- Octavia never receives this key, only the controller CA cert.

RUN -- jumphost

( {
  WORKDIR="$HOME/octavia-pki/${DC:?DC not set -- run 1.0-GEN.0 first}"
  cd "$WORKDIR/controller-ca" || exit 1   # dir from 1.0-GEN.0
  openssl rand -base64 32 | tr -d '\n' > passphrase.txt
  chmod 600 passphrase.txt
  test "$(wc -c < passphrase.txt)" -eq 44 || { echo "ABORT: controller passphrase length != 44"; exit 1; }
  openssl genpkey -algorithm EC -pkeyopt ec_paramgen_curve:P-384 \
    -aes-256-cbc -pass file:passphrase.txt -out controller-ca.key.enc
  chmod 600 controller-ca.key.enc
  openssl req -new -x509 -sha384 -key controller-ca.key.enc -passin file:passphrase.txt \
    -days 3650 -subj "/CN=${DC_LABEL} Omega Cloud Octavia Controller CA/O=Neumatrix" \
    -out controller-ca.cert.pem
  openssl x509 -in controller-ca.cert.pem -noout -dates -subject
  openssl verify -CAfile controller-ca.cert.pem controller-ca.cert.pem    # expect: OK
} )

1.0-GEN.c -- Controller cert (EC P-256 UNENCRYPTED, SAN, signed by Controller CA, 2y)

The P-256 key is unencrypted -- Octavia reads it at startup. SAN carries the controller FQDN, the octavia API FQDN, and the Octavia API VIP (derived from the bundle at generation time -- DOCFIX-067; never a baked literal).

RUN -- jumphost

( {
  WORKDIR="$HOME/octavia-pki/${DC:?DC not set -- run 1.0-GEN.0 first}"
  cd "$WORKDIR/controller" || exit 1   # dir from 1.0-GEN.0
  openssl genpkey -algorithm EC -pkeyopt ec_paramgen_curve:P-256 -out controller.key
  chmod 600 controller.key
  cat > controller.cnf <<'CNF'
[req]
distinguished_name = req_distinguished_name
req_extensions = v3_req
prompt = no

[req_distinguished_name]
CN = octavia-controller.omega.dc0.vr0.cloud.neumatrix.local
O = Neumatrix

[v3_req]
keyUsage = critical, digitalSignature, keyEncipherment
extendedKeyUsage = clientAuth, serverAuth
subjectAltName = @alt_names

[alt_names]
DNS.1 = octavia-controller.omega.dc0.vr0.cloud.neumatrix.local
DNS.2 = octavia.omega.dc0.vr0.cloud.neumatrix.local
CNF
  # NOTE (logged 2026-07-29, NOT changed -- outside R7's ruled scope): the CN and the two
  # DNS SANs above still carry the `dc0.vr0` identity. R7 ruled only the CA SUBJECT and
  # the VIP gate, and recorded that the controller cert's SAN is "already DERIVED per-DC
  # by design (DOCFIX-067)" -- which is true of the IP SAN, not of these DNS names. They
  # are inert while `os-public-hostname` is unset on every API charm (bundle.yaml:11,
  # IP-ONLY endpoints per B5), so nothing resolves them today. Revisit with D-106's
  # hostname work, not here.
  #
  # (The "no IPv6 IP SAN" gap this block used to log was RULED and CLOSED 2026-07-29 --
  # operator: "Yes, add the v6 IP sans" -- see the D-109 ruling note. IP.2 is emitted
  # below. Still true and still worth knowing: amphorae reach the controller on o-hm0's
  # charm-generated fc00::/64 ULA, not on any VIP leg, so neither SAN matches THAT path;
  # settle by inspection at the Octavia step rather than assuming.)
  #
  # DOCFIX-067: the SAN IP is DERIVED from the bundle (provider leg of octavia's
  # vip triple), never a literal -- the old baked 10.12.4.233 went stale when R14
  # relocated the VIPs to .50-.60. Measured value is echoed; confirm it before
  # the sign. BUNDLE may be overridden if the hardened bundle lives elsewhere.
  REPO="${REPO:?REPO not set -- export REPO=<path to your clone> (see runbooks/README.md Conventions)}"
  BUNDLE="${BUNDLE:-$REPO/bundle.yaml}"
  [ -r "$BUNDLE" ] || { echo "FAIL: bundle not readable at $BUNDLE (set BUNDLE=)"; exit 1; }
  # RULING 3 (2026-07-25): octavia's vip lives in the per-DC OVERLAY now, not in
  # $BUNDLE -- reading the base alone yields a KeyError that `|| true` would swallow
  # into an empty VIP, so this reads the MERGED deploy input explicitly.
  #
  # R7 (2026-07-27) carried an explicit caveat: "the generator's VIP gate must read the
  # MERGED deploy input, not bundle.yaml, or it breaks again the moment ruling-3's VIP
  # extraction and R11's .61/.62 land." Both have since landed, so the overlay IS the
  # merged input for this key, and it is selected by $DC rather than hardcoded to dc0.
  #
  # The gate was `^10\.12\.4\.` -- dc0's provider prefix -- so it hard-ABORTED on dc1's
  # 10.12.64.x. It now derives the expected prefix from the SAME overlay it read the VIP
  # from, so the check still bites (a VIP outside its own DC's provider plane fails) but
  # is no longer frozen to one DC. `.split()[0]` still takes the PROVIDER v4 leg, which
  # is the SAN this cert wants; the v6 legs are deliberately not added here (see the
  # note below).
  [ -r "$VIP_OVERLAY" ] || { echo "FAIL: VIP overlay not readable at $VIP_OVERLAY (set DC=)"; exit 1; }
  OCTAVIA_VIP=$(python3 -c "import yaml,sys;print(yaml.safe_load(open(sys.argv[1]))['applications']['octavia']['options']['vip'].split()[0])" "$VIP_OVERLAY" 2>/dev/null || true)
  KEYSTONE_VIP=$(python3 -c "import yaml,sys;print(yaml.safe_load(open(sys.argv[1]))['applications']['keystone']['options']['vip'].split()[0])" "$VIP_OVERLAY" 2>/dev/null || true)
  VIP_PREFIX="${KEYSTONE_VIP%.*}"      # this DC's provider-public /24-ish prefix, measured
  echo "measured octavia provider VIP: ${OCTAVIA_VIP:-<none>} (DC=$DC, expected prefix ${VIP_PREFIX:-<none>}.)"
  [ -n "$VIP_PREFIX" ] || { echo "FAIL: could not derive this DC's provider prefix from $VIP_OVERLAY -- stop"; exit 1; }
  echo "$OCTAVIA_VIP" | grep -qE "^$(printf '%s' "$VIP_PREFIX" | sed 's/\./\\./g')\.[0-9]{1,3}$" || { echo "FAIL: implausible VIP -- stop"; exit 1; }
  # D-109 RULING NOTE (2026-07-29, operator: "Yes, add the v6 IP sans"): the cert gains
  # this DC's provider v6 leg alongside its v4 one. Both are DERIVED from the same
  # per-DC overlay; the admin and internal legs stay excluded exactly as before, because
  # DOCFIX-067's design has always been provider-leg-only. A v4-only (3-address) vip
  # yields IP.1 alone and no v6 SAN, so this stays correct on a pre-dual-stack tree.
  OCTAVIA_VIP6=$(python3 -c "import yaml,sys,ipaddress;v=yaml.safe_load(open(sys.argv[1]))['applications']['octavia']['options']['vip'].split();print(next((a for a in v if ipaddress.ip_address(a).version==6),''))" "$VIP_OVERLAY" 2>/dev/null || true)
  printf 'IP.1 = %s\n' "$OCTAVIA_VIP" >> controller.cnf
  if [ -n "${OCTAVIA_VIP6:-}" ]; then
    echo "measured octavia provider v6 VIP: $OCTAVIA_VIP6"
    printf 'IP.2 = %s\n' "$OCTAVIA_VIP6" >> controller.cnf
  else
    echo "NOTE: octavia vip is v4-only (3 addresses) -- no IPv6 IP SAN emitted (expected before the dual-stack ADD)"
  fi
  openssl req -new -sha256 -key controller.key -config controller.cnf -out controller.csr
  openssl x509 -req -sha256 -in controller.csr \
    -CA ../controller-ca/controller-ca.cert.pem \
    -CAkey ../controller-ca/controller-ca.key.enc \
    -passin file:../controller-ca/passphrase.txt \
    -CAcreateserial -days 730 \
    -extfile controller.cnf -extensions v3_req \
    -out controller.cert.pem
  openssl verify -CAfile ../controller-ca/controller-ca.cert.pem controller.cert.pem  # expect: OK
  openssl x509 -in controller.cert.pem -noout -ext subjectAltName     # DNS x2 + IP present
  openssl x509 -in controller.cert.pem -noout -dates
  cat controller.cert.pem controller.key > controller.bundle.pem
  chmod 600 controller.bundle.pem
} )

1.0-GEN.d -- Write overlays/${DC}-octavia-pki.yaml (base64 blobs + plaintext passphrase)

Four values are base64(PEM); the issuing-CA passphrase is a PLAIN string. The file is gitignored. $REPO and $OCTAVIA_PKI_OVERLAY are exported in 1.0-GEN.0.

THE GITIGNORE CHECK BELOW IS LOAD-BEARING, NOT HYGIENE (F4, 2026-07-29). This file carries CA private-key blobs PLUS the issuing-CA passphrase in PLAINTEXT, and SEC-004 records this repo as currently PUBLIC -- the gitignore is the only thing between a CA private key and publication. .gitignore used to pin the single exact path overlays/octavia-pki.yaml, so the moment F1 made the name per-DC the new name stopped matching and the file would have been committable. The pattern is now a glob, and this block ASSERTS that fact at generation time rather than trusting anyone to remember it.

RUN -- jumphost

( {
  WORKDIR="$HOME/octavia-pki/${DC:?DC not set -- run 1.0-GEN.0 first}"
  cd "$WORKDIR" || exit 1   # dir from 1.0-GEN.0
  REPO="${REPO:?REPO not set -- export REPO=<path to your clone> (see runbooks/README.md Conventions)}"
  OUT="${OCTAVIA_PKI_OVERLAY:?OCTAVIA_PKI_OVERLAY not set -- run 1.0-GEN.0 first}"
  mkdir -p "$REPO/overlays"
  # F4 GATE: refuse to write key material into a path git would track. Asserts on the
  # ACTUAL ignore decision for THIS path, not on the presence of a pattern.
  git -C "$REPO" check-ignore -q "$OUT" || {
    echo "ABORT: $OUT is NOT gitignored -- writing CA key material here would make it"
    echo "       committable in a repo SEC-004 records as PUBLIC. Fix .gitignore first."
    exit 1; }
  ISS_CERT=$(base64 -w0 issuing-ca/issuing-ca.cert.pem)
  ISS_KEY=$(base64 -w0 issuing-ca/issuing-ca.key.enc)
  ISS_PASS=$(cat issuing-ca/passphrase.txt)
  CON_CACERT=$(base64 -w0 controller-ca/controller-ca.cert.pem)
  CON_CERT=$(base64 -w0 controller/controller.bundle.pem)
  cat > "$OUT" <<OVL
applications:
  octavia:
    options:
      lb-mgmt-issuing-cacert: "$ISS_CERT"
      lb-mgmt-issuing-ca-private-key: "$ISS_KEY"
      lb-mgmt-issuing-ca-key-passphrase: "$ISS_PASS"
      lb-mgmt-controller-cacert: "$CON_CACERT"
      lb-mgmt-controller-cert: "$CON_CERT"
OVL
  chmod 600 "$OUT"
  echo "wrote $OUT"
} )

Then run the Step 1.0 sanity block (5 keys present; ASCII clean) before deploying. Keep $HOME/octavia-pki/ (the CA keys + passphrases) OFF the repo and backed up securely; the 10y CAs are reused across rebuilds -- regenerate only on key compromise or CA expiry.

Per-DC layout produced (F1). Each DC is a wholly separate trust domain on disk, so a second DC's generation cannot reach the first DC's material:

$HOME/octavia-pki/vr1-dc0/{issuing-ca,controller-ca,controller}/
$HOME/octavia-pki/vr1-dc1/{issuing-ca,controller-ca,controller}/
$REPO/overlays/vr1-dc0-octavia-pki.yaml
$REPO/overlays/vr1-dc1-octavia-pki.yaml

Both the workspace tree and the overlay MUST be in the per-DC backup set: losing the controller-CA key makes that DC's controller-cert rotation impossible.

Phase 01 -- Bundle Deploy

Prerequisites (must be true entering phase-01)

Constants and env-literals

Run-location legend

Command-label convention

Step 1.0 -- Octavia PKI overlay (secret-handling prereq) DISCRETE

Step 1.1 -- Pre-deploy verify (read-only; 4 checks)

Step 1.2 -- Dry-run (guarded)

Step 1.3 -- Deploy (VIP-guarded)

Step 1.4 -- DNS gate during deploy (as machines come up)

EXIT GATE (phase-01 complete)

As-built reference (2026-06-03 second redeploy -- audit trail)

Balance / stability observations (Roosevelt rebalancing inputs -- post-deploy item 6)

PATTERN-1 (standing convention) -- dynamic lookup vs. pinned identifiers

Next

Step 1.0-GEN -- Octavia management-PKI generation (regeneration path) DISCRETE / SECRET

1.0-GEN.0 -- select the DC, then the workspace (openssl 3.x; $HOME only -- snap home-confinement, never /tmp)

1.0-GEN.a -- Issuing CA (EC P-384, AES-256 encrypted key, self-signed 10y)

1.0-GEN.b -- Controller CA (EC P-384, AES-256 encrypted key, self-signed 10y; own passphrase)

1.0-GEN.c -- Controller cert (EC P-256 UNENCRYPTED, SAN, signed by Controller CA, 2y)

1.0-GEN.d -- Write overlays/${DC}-octavia-pki.yaml (base64 blobs + plaintext passphrase)