Nixos updates 11-10-2026
This commit is contained in:
Alexander
@@ -39,7 +39,7 @@ in {
|
||||
|
||||
## Module dependencies
|
||||
# :checkers spell
|
||||
(aspellWithDicts (ds: with ds; [ en en-computers en-science ]))
|
||||
(aspellWithDicts (ds: with ds; [ en ]))
|
||||
# :tools editorconfig
|
||||
editorconfig-core-c # per-project style config
|
||||
# :tools lookup & :lang org +roam
|
||||
|
||||
+101
-61
@@ -8,8 +8,8 @@
|
||||
#
|
||||
# Stage Step command Effect
|
||||
# ------ ------------------------- --------------------------------------
|
||||
# 1 check-image Read-only: is a NixOS LXC ostemplate on PVE?
|
||||
# deploy-image Idempotent: build+upload if missing, no-op if present.
|
||||
# 1 check-image Read-only: is the ostemplate matching the current flake on PVE?
|
||||
# deploy-image Idempotent: build+upload if missing/stale, no-op if current.
|
||||
# update-image Force: rebuild locally + replace on PVE.
|
||||
# 2 start-builder deploy-image + clone a fresh CT under a free
|
||||
# VMID + start it + wait for DHCP.
|
||||
@@ -90,6 +90,7 @@ PVE_BRIDGE="${PVE_BRIDGE:-vmbr0}" # network bridge
|
||||
VMID_START="${VMID_START:-9999}" # scan downward from here for a free VMID
|
||||
VMID_FLOOR="${VMID_FLOOR:-100}" # lowest VMID to consider (PVE minimum)
|
||||
TEMPLATE_OUTPUT="${TEMPLATE_OUTPUT:-nixos-lxc}" # flake output to build
|
||||
TEMPLATE_TOPLEVEL_OUTPUT="${TEMPLATE_TOPLEVEL_OUTPUT:-nixos-lxc-toplevel}" # flake output a disposable must be running
|
||||
CT_HOSTNAME_PREFIX="${CT_HOSTNAME_PREFIX:-nixos-builder}" # disposables' hostname prefix
|
||||
BUILD_SSH_USER="${BUILD_SSH_USER:-builder}" # user nixos-rebuild SSHes as on the CT
|
||||
CT_BOOT_TIMEOUT="${CT_BOOT_TIMEOUT:-120}" # seconds to wait for DHCP lease
|
||||
@@ -244,9 +245,9 @@ Composite commands (chain all stages):
|
||||
+ destroy-builder. Disposable destroyed on success.
|
||||
|
||||
Step commands (run one stage — VMID auto-discovered from PVE if omitted):
|
||||
check-image Read-only: exit 0 if a NixOS LXC ostemplate is on PVE, 1 if not.
|
||||
deploy-image Idempotent: build+upload if missing, no-op if present.
|
||||
update-image Force: rebuild .$TEMPLATE_OUTPUT locally + replace on PVE.
|
||||
check-image Read-only: exit 0 if the ostemplate matching the current flake is on PVE.
|
||||
deploy-image Idempotent: build+upload if missing/stale, no-op if current.
|
||||
update-image Force: rebuild .$TEMPLATE_OUTPUT + upload. Old templates are kept.
|
||||
start-builder deploy-image + clone a fresh CT + start.
|
||||
build-on [VMID] Build closure on an existing disposable.
|
||||
activate <test|switch|boot>
|
||||
@@ -279,7 +280,8 @@ Options:
|
||||
Environment:
|
||||
PVE_HOST, PVE_STORAGE, PVE_ROOTFS_STORAGE, PVE_BRIDGE,
|
||||
VMID_START, VMID_FLOOR, BUILD_SSH_USER,
|
||||
CT_BOOT_TIMEOUT, CT_ROOTFS_GIB, TEMPLATE_OUTPUT, CT_HOSTNAME_PREFIX
|
||||
CT_BOOT_TIMEOUT, CT_ROOTFS_GIB, TEMPLATE_OUTPUT, TEMPLATE_TOPLEVEL_OUTPUT,
|
||||
CT_HOSTNAME_PREFIX
|
||||
|
||||
Examples:
|
||||
# Full happy path — build + activate susano permanently, auto-destroy CT.
|
||||
@@ -481,17 +483,33 @@ resolve_vmid() {
|
||||
## ---------------------------------------------------------------------------
|
||||
# Stage 1: image management (check / deploy / update).
|
||||
# All three operate on the NixOS LXC ostemplate stored in PVE's
|
||||
# /var/lib/vz/template/cache/. The "latest" template is whichever
|
||||
# nixos-lxc-*.tar.xz sorts last alphabetically (matches the helpers below).
|
||||
# /var/lib/vz/template/cache/.
|
||||
#
|
||||
# Template identity: .$TEMPLATE_OUTPUT's store path is evaluated (no build);
|
||||
# its basename embeds the full derivation hash — nixpkgs pin AND NixOS
|
||||
# config — and becomes the uploaded filename. Matching against PVE is exact:
|
||||
# a template matches the current flake iff its name equals the evaluated
|
||||
# one. This catches both "missing" and "stale" images. (The previous
|
||||
# "alphabetically last nixos*lxc*.tar.xz" pick was wrong: unstable version
|
||||
# strings like 26.05pre-git sort above dated stable ones like 25.05.2026…,
|
||||
# so a stale image could win over the current one.)
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
# Echo the storage-relative path of the newest nixos-lxc template on PVE, or
|
||||
# empty if none. Read-only.
|
||||
find_template_on_pve() {
|
||||
debug "find_template_on_pve: pveam list $PVE_STORAGE"
|
||||
pve pveam list "$PVE_STORAGE" 2>/dev/null \
|
||||
| awk '/nixos.*lxc.*\.tar\.xz/ {print $1}' \
|
||||
| sort | tail -n1 || true
|
||||
# Evaluate the image derivation's store path WITHOUT building it, and derive
|
||||
# the canonical template filename from it. Sets TEMPLATE_FILENAME and
|
||||
# OSTEMPLATE. Because the store-path hash covers the flake lock and the
|
||||
# whole NixOS config, any edit to machines/builder/ or nixpkgs pin bump
|
||||
# changes the filename — "same name, older build" becomes impossible.
|
||||
resolve_template_identity() {
|
||||
step "Evaluating .$TEMPLATE_OUTPUT image identity (nix eval, no build) ..."
|
||||
local out_path
|
||||
out_path=$(cd "$FLAKE_ROOT" && nix eval --raw ".#$TEMPLATE_OUTPUT.outPath") \
|
||||
|| die "Failed to eval .$TEMPLATE_OUTPUT.outPath — is the flake evaluable?"
|
||||
TEMPLATE_FILENAME="$(basename "$out_path").tar.xz"
|
||||
OSTEMPLATE="$PVE_STORAGE:vztmpl/$TEMPLATE_FILENAME"
|
||||
debug "image outPath: $out_path"
|
||||
debug "template on PVE: $OSTEMPLATE"
|
||||
ok "Current flake image: $TEMPLATE_FILENAME"
|
||||
}
|
||||
|
||||
# Build .$TEMPLATE_OUTPUT locally and scp the tarball to PVE's template cache.
|
||||
@@ -505,80 +523,60 @@ build_and_upload_image() {
|
||||
|| die "nix build .$TEMPLATE_OUTPUT produced no output path."
|
||||
log " nix output: $result_dir"
|
||||
|
||||
# nixos-generators proxmox-lxc format lays the file under result/tarball/*.tar.xz
|
||||
# Identity first, so the upload target is the canonical hash-embedded
|
||||
# filename even when called without a prior check_image (update-image).
|
||||
resolve_template_identity
|
||||
|
||||
# nixos-generators / proxmox-lxc image lays the file under result/tarball/*.tar.xz
|
||||
local tarball
|
||||
tarball=$(find "$result_dir/tarball" -maxdepth 1 -type f -name '*.tar.xz' -print -quit | head -n1)
|
||||
[[ -n "$tarball" ]] \
|
||||
|| die "No .tar.xz found under $result_dir/tarball/. Did nixos-generators output layout change?"
|
||||
local tname; tname=$(basename "$tarball")
|
||||
|| die "No .tar.xz found under $result_dir/tarball/. Did the image output layout change?"
|
||||
debug "tarball path: $tarball"
|
||||
debug "tarball name: $tname"
|
||||
ok "Built image: $tname"
|
||||
ok "Built image: $(basename "$tarball")"
|
||||
|
||||
step "Uploading to $PVE_HOST:/var/lib/vz/template/cache/ ..."
|
||||
local size_human
|
||||
size_human=$(du -h "$tarball" | cut -f1)
|
||||
log " scp $tarball ($size_human) -> $PVE_HOST:/var/lib/vz/template/cache/$tname"
|
||||
pve_scp "$tarball" "$PVE_HOST:/var/lib/vz/template/cache/$tname"
|
||||
|
||||
OSTEMPLATE="$PVE_STORAGE:vztmpl/$tname"
|
||||
log " scp $tarball ($size_human) -> $PVE_HOST:/var/lib/vz/template/cache/$TEMPLATE_FILENAME"
|
||||
pve_scp "$tarball" "$PVE_HOST:/var/lib/vz/template/cache/$TEMPLATE_FILENAME"
|
||||
}
|
||||
|
||||
# Remove old nixos-lxc templates from PVE, keeping $1. Must be called after
|
||||
# the new image is uploaded so a build/upload failure leaves the previous
|
||||
# template intact.
|
||||
purge_old_templates_on_pve() {
|
||||
local keep="${1:-}"
|
||||
step "Purging old nixos-lxc templates from $PVE_HOST (keeping ${keep:-none}) ..."
|
||||
local all
|
||||
all=$(pve pveam list "$PVE_STORAGE" 2>/dev/null \
|
||||
| awk '/nixos.*lxc.*\.tar\.xz/ {print $1}' || true)
|
||||
local removed=0
|
||||
while IFS= read -r t; do
|
||||
[[ -n "$t" ]] || continue
|
||||
if [[ -n "$keep" && "$t" == "$keep" ]]; then
|
||||
debug "keeping: $t"
|
||||
continue
|
||||
fi
|
||||
local filename="${t##*/}"
|
||||
log " removing: $t"
|
||||
pve rm -f "/var/lib/vz/template/cache/$filename" 2>/dev/null || true
|
||||
removed=$((removed + 1))
|
||||
done <<<"$all"
|
||||
ok "Purged $removed old template(s)"
|
||||
}
|
||||
|
||||
# Read-only check. Sets OSTEMPLATE and returns 0 if a template exists;
|
||||
# clears OSTEMPLATE and returns 1 if missing.
|
||||
# Read-only check. Sets OSTEMPLATE and returns 0 if the template matching
|
||||
# the CURRENT flake build is on PVE; clears OSTEMPLATE and returns 1 if it
|
||||
# is missing or stale (a differently-named nixos-lxc template may exist).
|
||||
check_image() {
|
||||
step "Checking for nixos-lxc ostemplate on $PVE_HOST (storage: $PVE_STORAGE) ..."
|
||||
resolve_template_identity
|
||||
step "Checking for $OSTEMPLATE on $PVE_HOST (storage: $PVE_STORAGE) ..."
|
||||
local existing
|
||||
existing=$(find_template_on_pve)
|
||||
existing=$(pve pveam list "$PVE_STORAGE" 2>/dev/null | awk -v t="$OSTEMPLATE" '$1 == t {print $1}')
|
||||
if [[ -n "$existing" ]]; then
|
||||
OSTEMPLATE="$existing"
|
||||
ok "Image present: $OSTEMPLATE"
|
||||
ok "Image present (matches current flake build): $OSTEMPLATE"
|
||||
return 0
|
||||
fi
|
||||
OSTEMPLATE=""
|
||||
warn "No nixos-lxc ostemplate found."
|
||||
warn "No template matching the current flake build on $PVE_HOST."
|
||||
return 1
|
||||
}
|
||||
|
||||
# Idempotent deploy: build + upload only if no template exists.
|
||||
# Idempotent deploy: ensure PVE holds the template matching the CURRENT
|
||||
# flake build. No-op when it matches; build + upload otherwise. Historical
|
||||
# templates are deliberately left in place — other tooling (e.g. the
|
||||
# terraform handler) may pin containers to them.
|
||||
deploy_image() {
|
||||
if check_image; then
|
||||
log "Nothing to do — image already on PVE."
|
||||
log "Nothing to do — image on PVE matches current flake build."
|
||||
return
|
||||
fi
|
||||
build_and_upload_image
|
||||
ok "Image deployed: $OSTEMPLATE"
|
||||
}
|
||||
|
||||
# Force update: build + upload first, then purge old. If the build or upload
|
||||
# fails, the previous template remains on PVE.
|
||||
# Force update: rebuild + upload even if the current template is present.
|
||||
# Old templates are never deleted (terraform may pin them); prune manually
|
||||
# with `pveam remove` if really desired.
|
||||
update_image() {
|
||||
build_and_upload_image
|
||||
purge_old_templates_on_pve "$OSTEMPLATE"
|
||||
ok "Image updated: $OSTEMPLATE"
|
||||
}
|
||||
|
||||
@@ -640,11 +638,27 @@ create_and_start_ct() {
|
||||
# Echo the current IPv4 of a running container's eth0, or empty if unavailable.
|
||||
get_ct_ip() {
|
||||
local vmid="$1"
|
||||
pve pct exec "$vmid" -- ip -j -4 addr show eth0 2>/dev/null \
|
||||
# NixOS containers have no standard /bin|/sbin — lxc-attach needs the
|
||||
# full path or exec fails with "No such file or directory".
|
||||
pve pct exec "$vmid" -- /run/current-system/sw/bin/ip -j -4 addr show eth0 2>/dev/null \
|
||||
| jq -r '.[0].addr_info[]? | select(.scope=="global") | .local' 2>/dev/null \
|
||||
| head -n1 || true
|
||||
}
|
||||
|
||||
# Echo the NixOS toplevel a running disposable actually booted, or empty.
|
||||
# pct config drops the ostemplate field after creation, so the container's
|
||||
# own /run/current-system is the only reliable record of what it runs.
|
||||
ct_toplevel() {
|
||||
local vmid="$1"
|
||||
pve pct exec "$vmid" -- /run/current-system/sw/bin/readlink /run/current-system 2>/dev/null || true
|
||||
}
|
||||
|
||||
# Echo the toplevel the current flake's builder image boots, or empty on
|
||||
# eval failure (callers treat empty as "cannot tell, skip the check").
|
||||
expected_toplevel() {
|
||||
(cd "$FLAKE_ROOT" && nix eval --raw ".#$TEMPLATE_TOPLEVEL_OUTPUT.outPath" 2>/dev/null) || true
|
||||
}
|
||||
|
||||
# Wait for a disposable to become SSH-reachable via its hostname. Echos the
|
||||
# hostname on success (so the caller can pass it to do_build / probe_ssh).
|
||||
# We deliberately do NOT use the IP here — the script's contract is that
|
||||
@@ -832,6 +846,21 @@ cmd_composite() {
|
||||
step "Container $vmid is stopped — starting ..."
|
||||
pve pct start "$vmid"
|
||||
fi
|
||||
|
||||
local running_top want_top
|
||||
running_top=$(ct_toplevel "$vmid")
|
||||
want_top=$(expected_toplevel)
|
||||
if [[ -n "$running_top" && -n "$want_top" && "$running_top" != "$want_top" ]]; then
|
||||
warn "Disposable $vmid runs a stale image:"
|
||||
warn " running: $running_top"
|
||||
warn " current: $want_top"
|
||||
step "Destroying $vmid and recreating from current image ..."
|
||||
pve pct destroy "$vmid" --purge --force
|
||||
vmid=$(find_free_vmid)
|
||||
hostname="$CT_HOSTNAME_PREFIX-$vmid"
|
||||
log "Created fresh disposable: $hostname"
|
||||
create_and_start_ct "$vmid"
|
||||
fi
|
||||
fi
|
||||
|
||||
hostname=$(wait_for_ready "$vmid")
|
||||
@@ -867,6 +896,7 @@ cmd_composite() {
|
||||
# ---------------------------------------------------------------------------
|
||||
cmd_check_image() {
|
||||
require ssh "drive PVE"
|
||||
require nix "evaluate the image identity"
|
||||
require_pve
|
||||
phase "check-image — read-only probe of ostemplate on PVE"
|
||||
if check_image; then
|
||||
@@ -970,6 +1000,16 @@ cmd_build_on() {
|
||||
die "Start it on PVE first ('pct start $vmid') or pick another VMID."
|
||||
fi
|
||||
|
||||
# Guard: a disposable cloned from a stale template boots without
|
||||
# sshd/networkd config and hangs the build — refuse instead of hanging.
|
||||
local running_top want_top
|
||||
running_top=$(ct_toplevel "$vmid")
|
||||
want_top=$(expected_toplevel)
|
||||
if [[ -n "$running_top" && -n "$want_top" && "$running_top" != "$want_top" ]]; then
|
||||
fail "Disposable $vmid runs a stale image: $running_top"
|
||||
die "Current flake boots: $want_top — recreate with: $0 destroy-builder $vmid && $0 start-builder"
|
||||
fi
|
||||
|
||||
probe_ssh "$hostname" || { print_debug_info "$vmid" "$hostname"; die "Aborting build-on."; }
|
||||
|
||||
do_build "$vmid" "$hostname" && rc=0 || rc=$?
|
||||
|
||||
Reference in New Issue
Block a user