Root cause: the k3s nodes shipped with a SINGLE upstream resolver, PiHole
(192.168.1.18), set by cloud-init/netplan. PiHole answers the local jackson.lan
zone instantly but its external/recursive resolution flaps — verified 0/25 for
public names (registry-1.docker.io, api.cloudflare.com, quay.io, ghcr.io,
google.com) while gitea.jackson.lan resolved 5/5. A single upstream is a hard
SPOF: both containerd image pulls ("lookup registry-1.docker.io: Try again")
and CoreDNS ("forward . /etc/resolv.conf" -> pods' external DNS, e.g.
external-dns -> Cloudflare) fail whenever PiHole recursion is down. CoreDNS
logs were flooded with "read udp ...->192.168.1.18:53: i/o timeout".
Fix (additive, no policy bypass): keep PiHole PRIMARY (preserves jackson.lan +
network ad-blocking) and ADD the pfSense gateway (192.168.1.1) as a FALLBACK,
which resolves the public zones reliably (verified). The resolver fails over to
the fallback when PiHole times out.
- node_baseline: netplan drop-in (99-k3s-dns.yaml) with both nameservers +
"Apply netplan" handler; new node_baseline_dns_* defaults; tagged `dns`.
Applied to all 3 nodes via playbooks/node-baseline.yml --tags dns.
- tofu: initialization.dns.servers now lists [dns_server, network_gateway] so a
fresh cloud-init/netplan build stays in parity with the drop-in.
Verified: all external names resolve cluster-wide (10/10, zero timeouts) from
multiple pods after a CoreDNS restart; CoreDNS i/o-timeout floods stopped;
split-horizon (keycloak.home.keithjackson.dev -> Gateway VIP) intact.
NOTE: a SEPARATE, pre-existing WAN-egress outage (ICMP + TCP 443 to the internet
time out from every node while the LAN/pfSense resolver work) still blocks image
pulls, external-dns->Cloudflare, and Flux remote bases. That is a pfSense
NAT/firewall/WAN problem outside this repo, flagged for PM follow-up.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
bpg/proxmox deprecates proxmox_virtual_environment_download_file ("Use
proxmox_download_file instead. This resource will be removed in v1.0."),
the short-name rename from ADR-007. Both are the same dual-registered
Framework resource, so rename the resource type + import_from reference
and add a `moved` block. OpenTofu 1.9 drives the provider's MoveState
(migration.PrefixMoveState) to transfer state in place: same id, no
re-download of the Debian image, no change to the VMs.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
Replace the dmacvicar/libvirt provider + resources with the modern
bpg/proxmox provider. The dev cluster moves off the memory-constrained
local libvirt host onto the Proxmox VE box.
- versions.tf: pin bpg/proxmox ~> 0.66; provider reads auth from env
(PROXMOX_VE_*, token never a tofu var) and normalizes a schemeless
PROXMOX_VE_ENDPOINT (prepend https://) to avoid the :8006 HTTP 301.
- variables.tf: Proxmox placement vars (node=proxmox, datastore=Storage,
image_datastore=local, bridge=vmbr0); LAN 192.168.1.0/24 gw .1 DNS .18;
node_ips .50/.51/.52; sizing 6 vCPU / 16384 MiB / 40+100 GiB; cpu_type
host; Cilium VIP range .53-.60.
- main.tf: proxmox_virtual_environment_download_file (cloud image -> local),
per-node user-data snippet (proxmox_virtual_environment_file), and
proxmox_virtual_environment_vm per node via name-keyed for_each (imported
OS disk virtio0, blank data disk virtio1, native initialization ip_config
+ dns + user_data_file_id).
- remove moved.tf (libvirt-specific count->for_each state migration).
Authoring only; no apply. Traces R-ENV-01, R-PLAT-01.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
Implements T-1.5 findings F1/F2 (author-first, no cluster change).
F1 (tofu): refactor libvirt node resources count -> for_each keyed by
node name so single-node -replace/destroy cannot cascade to siblings.
Add ordered-list/primary helper locals; update outputs.tf off the map;
add moved.tf (12 moved blocks). tofu plan proves a pure state-address
migration: 0 to add, 0 to change, 0 to destroy (all 12 'has moved to').
State-persist via a 0-change apply is operator-permission-gated.
F2 (ansible): k3s_server secondary fetches the join token via
delegate_to the primary (published fact as fast path), so a single-node
rejoin with --limit k3s-dev-N is self-contained and no longer depends on
a fact an earlier play had to set. Syntax-check + CI gate green.
Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
jakman85andClaude Opus 4.8