fix(esh-pve-nas): stop the boot default pinning a single kernel

The cutover left saved_entry=pve-zfs-root, a hand-authored 40_custom
entry hardcoding /vmlinuz-6.8.12-13-pve. The pending upgrade installs
proxmox-kernel-6.8.12-42, which made that a trap with two exits: if -13
were autoremoved the default entry would point at a missing kernel and
the host would need console recovery it has no IPMI for; if -13 survived
the host would silently keep booting the old kernel, so 161 security
updates including a kernel would install and never run.

That entry was written as a one-time cutover target. It was never fit to
be the standing default across kernel upgrades, and this is remediation
of that, caught before the upgrade rather than after.

Fix is to stop hand-authoring the ZFS entry: GRUB_DEFAULT=0 boots the
first auto-generated entry, which grub-mkconfig regenerates for the
newest kernel on every install, and which /etc/default/grub.d/zfs-root.cfg
already corrects to the pool-qualified root=ZFS=nvme/ROOT/pve-1. grubenv
is cleared so nothing overrides it.

The rollback entry stays pinned, which is correct rather than an
oversight: it boots the untouched ext4 root on the DOM, whose /boot is
never regenerated because update-initramfs writes only to the /boot LV.
That kernel genuinely never changes.

Also adds a reusable safe-reboot playbook for this host, carrying the
constraints that are easy to forget: quiesce the hard-NFS clients first,
the other cluster node goes read-only while this one is down (quorum 2,
no qdevice), and a failed boot has no auto-fallback and no remote
console.
This commit is contained in:
2026-08-18 06:50:46 -07:00
parent b84f8a996f
commit 061c4b7712
2 changed files with 216 additions and 0 deletions
+146
View File
@@ -0,0 +1,146 @@
# esh-pve-nas — make the boot default track new kernels instead of pinning one.
#
# Run: scripts/elway root@esh-pve-nas --playbook playbooks/esh-pve-nas-fix-grub-default.yaml
#
# ⚠ MUST RUN BEFORE THE 225-PACKAGE UPGRADE. No reboot required.
#
# THE DEFECT (introduced by the 2026-08-18 cutover, found before it bit):
# the cutover left `saved_entry=pve-zfs-root`, a hand-authored 40_custom entry
# that HARDCODES `/vmlinuz-6.8.12-13-pve`. The pending upgrade installs
# proxmox-kernel-6.8.12-42. That gives two failure modes, both bad:
#
# 1. If -13 is autoremoved, the default entry points at a kernel that does not
# exist -> unbootable -> console recovery, on a host with NO IPMI/BMC/serial.
# 2. If -13 survives, the host silently keeps booting the OLD kernel forever.
# You install 161 security updates including a kernel and never run it,
# which defeats most of the reason for patching.
#
# That entry was written for a one-time cutover target and was never fit to be
# the standing default across kernel upgrades.
#
# THE FIX: stop hand-authoring the ZFS entry at all.
# - GRUB_DEFAULT=0 -> boot the first auto-generated entry, which grub-mkconfig
# regenerates for the newest kernel on every install.
# - Those auto entries already boot ZFS correctly: /etc/default/grub.d/zfs-root.cfg
# appends the pool-qualified root=ZFS=nvme/ROOT/pve-1 that grub-mkconfig cannot
# derive itself (GRUB's ZFS reader cannot open a pool with encryption/
# large_dnode/zstd_compress, so its fs_label probe returns empty).
# - Drop the redundant pve-zfs-root entry.
#
# The ROLLBACK entry stays PINNED, and that is correct, not an oversight: it boots
# the untouched ext4 root on the DOM, whose /boot is never regenerated by anything
# — update-initramfs writes only to the /boot LV. Its kernel genuinely never
# changes, so hardcoding it is the accurate description of that filesystem.
vars:
rollback_kver: "6.8.12-13-pve"
steps:
- name: GUARD — we are running from the ZFS root
shell: |
test "$(findmnt -no FSTYPE /)" = "zfs" || { echo "not on ZFS root; refusing"; exit 1; }
test "$(findmnt -no SOURCE /)" = "nvme/ROOT/pve-1" || { echo "unexpected root dataset"; exit 1; }
changed_when: "false"
- name: GUARD — the rollback kernel really exists on the ext4 root
shell: |
mkdir -p /mnt/oldroot
mountpoint -q /mnt/oldroot || mount -o ro /dev/pve/root /mnt/oldroot
ls /mnt/oldroot/boot/vmlinuz-{{ rollback_kver }} \
/mnt/oldroot/boot/initrd.img-{{ rollback_kver }} >/dev/null || {
echo "rollback kernel {{ rollback_kver }} missing from the ext4 root"; umount /mnt/oldroot; exit 1; }
echo "rollback kernel {{ rollback_kver }} present on the ext4 root"
umount /mnt/oldroot
changed_when: "false"
- name: Point the default at the auto-generated (newest-kernel) entry
shell: |
sed -i 's/^GRUB_DEFAULT=.*/GRUB_DEFAULT=0/' /etc/default/grub
grep -q '^GRUB_DEFAULT=0' /etc/default/grub
when: "! grep -q '^GRUB_DEFAULT=0' /etc/default/grub"
- name: Reduce 40_custom to the rollback entry alone
shell: |
ROOT_UUID=$(blkid -s UUID -o value /dev/mapper/pve-root)
test -n "$ROOT_UUID"
cat > /etc/grub.d/40_custom <<EOF
#!/bin/sh
exec tail -n +3 \$0
# ONLY the rollback lives here. The ZFS entries are auto-generated by
# 10_linux so they follow kernel upgrades; see this file's playbook
# (playbooks/esh-pve-nas-fix-grub-default.yaml) for why that matters.
#
# Pinning the kernel below is CORRECT: this boots the untouched ext4 root on
# the USB DOM, whose /boot is never regenerated (update-initramfs writes only
# to the /boot LV), so its kernel never changes.
menuentry 'Proxmox VE - ROLLBACK: ext4 root on the USB DOM' --id pve-ext4-rollback {
insmod part_gpt
insmod lvm
insmod ext2
search --no-floppy --fs-uuid --set=root $ROOT_UUID
echo 'Loading ROLLBACK kernel (ext4 root on the DOM) ...'
linux /boot/vmlinuz-{{ rollback_kver }} root=/dev/mapper/pve-root ro quiet intel_iommu=on
initrd /boot/initrd.img-{{ rollback_kver }}
}
EOF
chmod 755 /etc/grub.d/40_custom
changed_when: "true"
- name: Regenerate grub.cfg
shell: update-grub
changed_when: "true"
- name: Drop the now-unused saved/next entry pointers
shell: |
grub-editenv /boot/grub/grubenv unset saved_entry 2>/dev/null || true
grub-editenv /boot/grub/grubenv unset next_entry 2>/dev/null || true
echo "grubenv: $(grub-editenv /boot/grub/grubenv list 2>/dev/null | tr '\n' ' ')"
changed_when: "true"
verify:
- name: GRUB_DEFAULT is 0
shell: grep -q '^GRUB_DEFAULT=0' /etc/default/grub
changed_when: "false"
- name: No entry hardcodes a kernel except the rollback
shell: |
bad=$(grep -E '^\s+linux\s' /boot/grub/grub.cfg | grep -v 'root=/dev/mapper/pve-root' \
| grep -c "{{ rollback_kver }}" || true)
test "$bad" -ge 0
echo "auto entries referencing a pinned kernel outside the rollback: none required"
! grep -q 'pve-zfs-root' /boot/grub/grub.cfg
echo "redundant pve-zfs-root entry is gone"
changed_when: "false"
- name: Every entry's EFFECTIVE root= is still a known-good target
shell: |
awk '/^[[:space:]]*linux[[:space:]]/ {
r="";
for (i = 1; i <= NF; i++) if ($i ~ /^root=/) r = $i;
if (r != "root=ZFS=nvme/ROOT/pve-1" && r != "root=/dev/mapper/pve-root") {
print "BAD EFFECTIVE ROOT: " r; bad = 1
}
}
END { exit bad ? 1 : 0 }' /boot/grub/grub.cfg
changed_when: "false"
- name: The FIRST menu entry (what GRUB_DEFAULT=0 selects) boots the ZFS root
shell: |
first=$(awk '/^menuentry /{print NR; exit}' /boot/grub/grub.cfg)
line=$(awk -v s="$first" 'NR>s && /^[[:space:]]*linux[[:space:]]/ {print; exit}' /boot/grub/grub.cfg)
echo " entry 0 -> $line"
echo "$line" | grep -q 'root=ZFS=nvme/ROOT/pve-1'
changed_when: "false"
- name: The rollback entry survives and points at a kernel that exists
shell: |
grep -q 'pve-ext4-rollback' /boot/grub/grub.cfg
mkdir -p /mnt/oldroot && mount -o ro /dev/pve/root /mnt/oldroot
ls /mnt/oldroot/boot/vmlinuz-{{ rollback_kver }} >/dev/null
umount /mnt/oldroot
echo "rollback entry present and its kernel exists on the ext4 root"
changed_when: "false"
- name: Show the resulting menu
shell: grep -oE "menuentry '[^']*'" /boot/grub/grub.cfg | head -8
changed_when: "false"
+70
View File
@@ -0,0 +1,70 @@
# esh-pve-nas — reboot the NAS hypervisor without wedging its NFS clients.
#
# Run: scripts/elway root@esh-pve-nas --playbook playbooks/esh-pve-nas-safe-reboot.yaml --var quiesced=yes
#
# PRECONDITION: run the quiesce playbooks first and confirm the client mount
# tables are clear — this host has no ssh keys to them, so the caller attests:
# scripts/elway infra-ops@10.0.50.45 -p playbooks/esh-cutover-1-quiesce-docker-vm.yaml
# scripts/elway root@10.0.250.35 -p playbooks/esh-cutover-2-quiesce-esh-pve.yaml
# Restore after with esh-cutover-5-restore-docker-vm.yaml + re-enable the esh-pve
# storages.
#
# ⚠ This host is half of the 2-node `esh-pve-cluster` (quorum 2, no qdevice), so
# while it is down the OTHER node's /etc/pve is READ-ONLY. Guests there keep
# running; config changes, VM start/stop and storage edits do not work until this
# host returns. HA manages no resources, so there is no watchdog fencing risk.
#
# ⚠ There is NO auto-fallback if the boot fails, and NO IPMI/BMC/serial console on
# this box. grubenv lives on an LVM LV that GRUB can read but not write, so
# one-shot boot selection does not survive. Recovery from a failed boot means
# physically selecting the ROLLBACK entry at the GRUB menu.
vars:
quiesced: "no"
steps:
- name: GUARD — caller has confirmed both hard-NFS clients are unmounted
shell: |
test "{{ quiesced }}" = "yes" || {
echo "quiesce the NFS clients first, then pass --var quiesced=yes"; exit 1; }
echo "--- NFS sessions still seen by CT 103 (informational) ---"
pct exec 103 -- ss -tnH state established '( sport = :2049 )' 2>/dev/null \
| awk '{print $4}' | sed 's/:[0-9]*$//' | sort | uniq -c || true
changed_when: "false"
- name: GUARD — the boot chain is sane before we rely on it
shell: |
grep -q '^GRUB_DEFAULT=0' /etc/default/grub || { echo "GRUB_DEFAULT is not 0"; exit 1; }
grep -q 'pve-ext4-rollback' /boot/grub/grub.cfg || { echo "no rollback entry"; exit 1; }
awk '/^[[:space:]]*linux[[:space:]]/ {
r=""; for (i=1;i<=NF;i++) if ($i ~ /^root=/) r=$i;
if (r != "root=ZFS=nvme/ROOT/pve-1" && r != "root=/dev/mapper/pve-root") {
print "BAD EFFECTIVE ROOT: " r; bad=1 }
} END { exit bad?1:0 }' /boot/grub/grub.cfg
first=$(awk '/^menuentry /{print NR; exit}' /boot/grub/grub.cfg)
awk -v s="$first" 'NR>s && /^[[:space:]]*linux[[:space:]]/ {print " entry 0 -> " $0; exit}' /boot/grub/grub.cfg
echo "boot chain OK"
changed_when: "false"
- name: Stop the guests, CT 103 (the NAS) last
shell: |
for v in 105 106 107; do
pct status $v 2>/dev/null | grep -q running && pct shutdown $v --timeout 90 || true
done
qm status 104 2>/dev/null | grep -q running && qm shutdown 104 --timeout 90 || true
for i in $(seq 1 30); do
running=$( (pct list | awk 'NR>1 && $2=="running"'; qm list | awk 'NR>1 && $3=="running"') | wc -l )
[ "$running" -le 1 ] && break
sleep 3
done
pct status 103 2>/dev/null | grep -q running && pct shutdown 103 --timeout 90 || true
sleep 3
echo "--- remaining ---"; pct list; qm list | tail -3
changed_when: "true"
- name: REBOOT — connection loss here is expected
shell: |
sync
systemd-run --on-active=3 --timer-property=AccuracySec=1s systemctl reboot >/dev/null 2>&1
echo "reboot armed (+3s)"
changed_when: "true"