Virtualization

Proxmox VE 8 Is End of Life: How to Upgrade Nodes and Clusters to Proxmox VE 9

The Back Room Tech Staff · Edited by Aseem Kishore
22 min read

The Back Room Tech is reader-supported. We may earn a commission when you buy through links on our site. Learn more.

Proxmox VE 8 runs on Debian 12 “Bookworm”, and it doesn’t get security or bug-fix updates anymore. This runbook moves standalone nodes and clusters to Proxmox VE 9 on Debian 13 “Trixie”. It covers the pve8to9 pre-flight check, switching repositories to Trixie (including the newer deb822 format), apt dist-upgrade, the rolling order that keeps a cluster quorate, and the breaking changes that can stop a node or container from booting.

The short answer: get every node to the latest 8.4 first. If you use Ceph, upgrading it to Squid (19.2) before the PVE packages is recommended; check the upgrade wiki for the exact sequencing for your version.

Run pve8to9 --full and review its output until no issues remain, then switch your APT sources to Trixie and run apt dist-upgrade. In a cluster, do one node at a time, migrating its guests off and confirming quorum first.

Is Proxmox VE 8 End of Life? Dates and Current Release

Proxmox VE 8 is end of life. The support table in the official Proxmox FAQ lists two dates for PVE 8: Debian 12 (Bookworm), its base, reached end of life in July 2026, and Proxmox’s own support for PVE 8 ended in August 2026.

The Upgrade from 8 to 9 wiki page gives an estimated EOL of July 2026, which matches the Debian date. Treat the FAQ table as the source of truth, and check it for the exact day if your compliance paperwork needs one. Either way, both dates have passed.

The Proxmox VE Roadmap lists Proxmox VE 9.2 for x86-64 (released May 21, 2026) as the newest x86-64 point release when we checked on November 1, 2026. At release it was based on Debian Trixie 13.5 and shipped Ceph Squid 19.2.3, with Ceph Tentacle 20.2.1 also available. Updates since then move the Debian point level and package versions forward, so your node will report newer numbers after apt update. Point releases ship often, so check the roadmap before your change window.

Quick Reference

ScenarioWhat to Do
Standalone homelab nodeUpdate to latest 8.4 → pve8to9 --full → switch repos → apt dist-upgrade → reboot
Multi-node clusterSame per node, one node at a time; migrate guests off first; check pvecm status after each node
Hyper-converged CephUpgrading Ceph to Squid (19.2) before the PVE package upgrade is recommended; verify current sequencing in the wiki
Co-installed Proxmox Backup ServerFollow the separate PBS 3-to-4 upgrade procedure
Older LXC containersCheck container compatibility first; PVE 9 changes cgroup handling and some older containers may need upgrading
NVIDIA vGPU/passthrough hostConfirm the driver supports the PVE 9 kernel before rebooting
Heavily customized nodeConsider a fresh PVE 9 ISO install + restore from backup

What Is Proxmox VE?

Proxmox VE is an open-source, Debian-based hypervisor. It runs QEMU/KVM virtual machines and LXC containers on bare metal, as a single node or as a highly available cluster. Coming from ESXi instead of an existing PVE 8 host? See our VMware-to-Proxmox migration guide. This article assumes you already run PVE 8.

Prerequisites

This applies to Proxmox VE 8.4.x on x86_64 hardware. Per the FAQ, aarch64 is supported only on NVIDIA Grace Hopper/Vera platforms. Standard ARM boards such as a Raspberry Pi aren’t supported.

  • Every node on the latest 8.4 packages, with pve-manager at 8.4.1 or later
  • At least 5 GB free on the root filesystem (10 GB or more is recommended)
  • A current, tested backup of every VM and container. Only a successful test restore counts here. A green backup job proves very little.
  • Out-of-band console access: physical keyboard/monitor, IPMI/iKVM, or a KVM-over-IP device like PiKVM. At minimum, an SSH session inside tmux or screen.
  • Hyper-converged Ceph clusters upgraded to Ceph Squid (19.2) before starting the PVE upgrade, per current wiki guidance
  • Co-installed Proxmox Backup Server upgraded through the PBS 3-to-4 procedure
  • A UPS on the hosts. A power blip in the middle of dpkg is a bad afternoon.
RequirementDetails
Hardwarex86_64 (AMD64/Intel 64)
Starting versionProxmox VE 8.4, pve-manager >= 8.4.1
Target versionProxmox VE 9.x (Debian 13 Trixie)
Ceph (if used)Upgrade to Squid (19.2) first; confirm exact sequencing in the wiki
AccessRoot shell + Proxmox web UI on port 8006

Step-by-Step Guide: Proxmox VE 8 to 9 Upgrade

Step 1: Bring the Node to the Latest Proxmox VE 8.4

pve8to9 and the upgrade path assume you start from current 8.4 packages. Run this on every node:

apt update && apt dist-upgrade
pveversion

Expected output (your build hash and kernel will differ):

pve-manager/8.4.x/xxxxxxxxxxxxxxxx (running kernel: 6.x.x-x-pve)

The pve-manager version must be 8.4.1 or higher on every node, apt update must finish without repository errors, and apt dist-upgrade should report nothing left to upgrade (no held-back packages) before you continue. You can also check it in the web UI. Select the node in the left tree and open Summary.

Proxmox web UI with a node selected in the left resource tree and the Summary tab open, with the PVE Manager Version field showing pve-manager/8.4.x highlighted

Step 2: Open a Persistent Session and Check Disk Space

If SSH drops in the middle of apt, the node can end up half-upgraded. Run everything inside tmux:

apt install tmux
tmux new -s pve9upgrade   # reattach later with: tmux attach -t pve9upgrade
df -h /

Expected output: the Avail column for / shows at least 5G. If it shows less, clear space first. Old kernels, rotated logs, and stale ISOs and backups on local storage are the usual suspects.

Warning: Don’t run the upgrade from a shell inside the Proxmox web UI on the same node. That shell depends on services the upgrade restarts. Use SSH in tmux or a real console.

Step 3: Upgrade Ceph to Squid First (Hyper-Converged Clusters Only)

Upgrading Ceph to Squid (19.2) before the PVE package upgrade is the generally recommended approach; check the upgrade wiki for exact sequencing requirements for your version. Check every daemon:

ceph versions

Every entry should report 19.2.x ... squid. If any daemon still shows Reef (18.2.x), stop here. Follow the Ceph Reef-to-Squid guide on the Proxmox wiki first, then come back.

Step 4: Run pve8to9 –full

pve8to9 is designed to check your system for known upgrade blockers, such as repository config, Ceph versions, storage, and cluster health.

Even so, keep a recent tested backup before you run it, and don’t treat any single tool as a substitute for checking the official upgrade wiki for your specific setup.

pve8to9 --full   # --full adds the slower, more thorough checks

The output lists each check along with its result and flags anything that needs attention, ending with a summary. Fix anything reported as a failure, review anything flagged as a warning, and re-run the check until it comes back clean.

Root shell on a Proxmox VE 8.4 node showing real pve8to9 --full output with PASS/WARN/FAIL lines and the final SUMMARY block with counts, one WARN line and the summary highlighted

Community reports on the Proxmox forum mention microcode and systemd-boot/GRUB warnings. Those are anecdotal. Act on what your run reports.

Step 5: Clear the Breaking Changes Before You Reboot

These are the PVE 9 changes most likely to bite. Handle them on PVE 8, while you still have a working system.

cgroup v1 support is gone. PVE 9 runs only the unified cgroup v2 hierarchy; PVE 8 still allowed containers to fall back to legacy cgroup v1. The upgrade wiki gives the concrete threshold: containers running systemd version 230 or older (released in 2016) are not supported on PVE 9. That covers distributions such as CentOS 7 (systemd 219) and Ubuntu 16.04 (systemd 229). Migrate those containers to a supported OS while you’re still on PVE 8.

There’s no single command that proves compatibility for every container: not every container runs systemd, and a systemctl --version check only reports the version packaged inside the guest, not whether the host-kernel combination will actually work. Instead, check the container’s OS against the compatibility list in the upgrade wiki, and test-boot any questionable container on a spare PVE 9 host before you migrate it into production.

Network interface names can change. Moving to Debian Trixie’s newer udev and kernel can rename network interfaces — for example, a NIC that used an older predictable-naming scheme may come up under a different name after the reboot. That breaks any bridge or VLAN in /etc/network/interfaces that references the old interface name directly.

Before you reboot, record the current mapping with ip -br link show and compare it against what /etc/network/interfaces expects. If a bridge is tied to a specific NIC name, pin the names. The upgrade wiki points to the pve-network-interface-pinning tool shipped with PVE 9, which pins interfaces to nicX names; a systemd .link file also works. Have IPMI/iKVM or physical console access ready in case the node comes up unreachable.

LVM autoactivation defaults changed. If pve8to9 flags LVM volumes, follow the fix it prints and the remediation in the upgrade wiki. Pay extra attention on shared LVM storage. To see the current state:

lvs -o lv_name,vg_name,autoactivation

An enabled value means the LV is still set to autoactivate; that’s expected on PVE 8 and isn’t an error on its own. Treat pve8to9‘s warning as the authority: the wiki provides a migration script that disables autoactivation for guest volumes, strongly recommended on shared LVM and optional on local-only storage.

GRUB/LVM “disk not found” on boot. On nodes that boot in UEFI mode with root on LVM, GRUB can fail on the first reboot with `disk lvmid/… not found`. ZFS-root and legacy-BIOS nodes aren’t affected. The upgrade wiki’s fix is to install the correct GRUB meta-package before you reboot:

[ -d /sys/firmware/efi ] && apt install grub-efi-amd64

The [ -d /sys/firmware/efi ] test only installs the package on UEFI-booted nodes. Confirm root is on LVM first (findmnt -no SOURCE / shows a /dev/mapper/... device), and if the package was already installed, the command changes nothing.

Fixing it afterward means recovery media and a USB flash drive.

/etc/sysctl.conf is no longer read. Per the upgrade wiki, the systemd-sysctl version in PVE 9 no longer honors /etc/sysctl.conf, so anything you set there silently stops applying after the reboot. This mostly affects hosts with hand-tuned kernel parameters — networking, memory, or inotify limits are common examples.

Before you upgrade, check the file for active settings:

grep -v '^\s*#' /etc/sysctl.conf | grep -v '^\s*$'

Move anything it prints into a numbered drop-in such as /etc/sysctl.d/99-local.conf. After the upgrade, confirm the values are in effect, for example sysctl net.ipv4.ip_forward on a router-style host.

FRR/SDN customizations. The upgrade wiki warns that after the first SDN configuration apply on PVE 9, custom OpenFabric or OSPF fabric configurations in /etc/frr/frr.conf.local no longer work, and applying SDN disables the daemons enabled in /etc/frr/daemons. Back the file up now:

cp -a /etc/frr/frr.conf.local /root/frr.conf.local.pve8   # -a preserves ownership and timestamps

Passthrough and vGPU drivers. The default kernel changes across PVE 9 point releases: 9.0 shipped a 6.14 series, 9.1 moved to 6.17, and later releases can move again. Check the kernel for the exact release you’re installing on the Proxmox VE roadmap before you match drivers to it. For NVIDIA vGPU, the upgrade wiki says to be on at least GRID 18.3 on the host before upgrading.

It also notes user reports of VMs with PCI passthrough failing to start on kernel 6.14; the workaround is to boot and pin an older kernel until a fixed one lands. Older NVIDIA vGPU/GRID host driver branches are validated by NVIDIA against specific kernel versions, and a driver that worked on PVE 8’s kernel may not support whichever kernel your target PVE 9 release ships.

This mainly affects licensed vGPU/GRID deployments, where NVIDIA publishes a driver-to-kernel support matrix in its vGPU software documentation. Plain PCIe passthrough with the standard proprietary driver is less version-sensitive, but still worth checking. Confirm your driver branch against NVIDIA’s current compatibility documentation and the kernel version listed for your target release on the roadmap before you reboot.

Check what you have:

nvidia-smi --query-gpu=driver_version --format=csv,noheader

If your driver branch isn’t on NVIDIA’s documented support list for your target kernel, update it before upgrading. For IOMMU and VFIO setup, see our GPU passthrough on Proxmox VE guide.

Third-party packages. Forum users report Docker’s docker-compose-plugin and docker-buildx-plugin conflicting with Trixie. Treat that as a community report. If you run Docker on the host, consider removing those packages before the upgrade and reinstalling them afterward.

Step 6: Evacuate the Node (Clusters)

In a cluster, upgrade one node at a time. First confirm the cluster is healthy:

pvecm status

Expected output includes:

Quorate:          Yes

This one-node-at-a-time approach, evacuating guests and confirming quorum before and after, is the generally recommended path. Check a few gating conditions before you start each node: a two-node cluster loses quorum when one node reboots unless you have a QDevice, so confirm the QDevice is reachable; HA-managed guests should be moved by the HA stack (or HA put into maintenance) rather than migrated by hand; and on Ceph, ceph -s should be HEALTH_OK before you touch the next node.

Next, migrate or shut down every guest on the node you’re upgrading. In the web UI, use the node’s Bulk Migrate action if your version offers it (its location in the UI varies by release). It’s a convenient way to move every guest off, and the UI only offers options valid for your installed version.

For a single running VM, the CLI tool is qm migrate, which takes your VM ID and target node name; online migration of a running VM needs the --online flag. Don’t paste an example ID or node name — check the admin guide for the exact flags on your version.

Container migration is more version-dependent: available modes and flags for pct migrate differ between releases, so look up the current syntax in the admin guide rather than copying a command from here, and prefer the Bulk Migrate UI for containers if you’re unsure.

The rolling upgrade relies on migrating from older nodes to newer ones. Don’t plan on migrating guests from a PVE 9 node back to a PVE 8 node. Treat each node as one-way, and base your rollback plan on backups. Verify the exact migration-compatibility notes on the upgrade wiki for your versions.

Setting noout is cluster-wide: it stops Ceph from marking any OSD out while it’s set, not just the ones on the node you’re rebooting. Left on too long, it can mask a real, prolonged outage elsewhere in the cluster, so treat it as a short maintenance window rather than a set-and-forget flag.

Check cluster health before you touch anything, then set the flag:

ceph -s

Note whether noout already appears in the flags line. If it doesn’t, set it:

ceph osd set noout

Confirm noout now shows in the flags line of ceph -s. Record whether you set it or it was already set, so you restore that same state afterward.

Step 7: Switch the Repositories to Trixie (deb822)

PVE 9’s documentation points toward the deb822 .sources format in /etc/apt/sources.list.d/ for Debian and Proxmox repositories (Package Repositories).

Legacy one-line .list files may still work during the transition, but check the wiki for the current recommendation before you decide which format to use. Before changing anything, inventory every repository file on the node — Proxmox, Debian, Ceph, any vendor or subscription repo, and anything a previous admin dropped in by hand:

cat /etc/apt/sources.list
ls -la /etc/apt/sources.list.d/
grep -rn . /etc/apt/sources.list.d/ /etc/apt/sources.list
Terminal showing cat /etc/apt/sources.list and ls /etc/apt/sources.list.d/ on PVE 8 with bookworm entries visible, before the repository switch

Back up the whole directory, not just the one file you’re about to edit — it’s easy to miss a stray Ceph or vendor repo otherwise:

cp -a /etc/apt/sources.list /root/sources.list.pve8
cp -a /etc/apt/sources.list.d /root/sources.list.d.pve8

Then replace bookworm with trixie in the legacy file, and remove any bookworm-backports lines — they don’t apply to Trixie. This only covers /etc/apt/sources.list; you’ll handle every file in /etc/apt/sources.list.d/ below and verify with a final grep before apt update:

sed -i 's/bookworm/trixie/g' /etc/apt/sources.list   # -i edits in place; review the result

Better still, move the Debian repos to deb822 format. The suites shown below are a typical example; confirm the exact suite names (and whether to include things like trixie-updates and trixie-security) and keyring paths on the Package Repositories wiki before you save, since specifics can change between point releases. Create /etc/apt/sources.list.d/debian.sources:

# /etc/apt/sources.list.d/debian.sources
Types: deb
URIs: http://deb.debian.org/debian/
Suites: trixie trixie-updates
Components: main contrib non-free-firmware
Signed-By: /usr/share/keyrings/debian-archive-keyring.gpg

Types: deb
URIs: http://security.debian.org/debian-security/
Suites: trixie-security
Components: main contrib non-free-firmware
Signed-By: /usr/share/keyrings/debian-archive-keyring.gpg

Then add the Proxmox repo. No-subscription example:

# /etc/apt/sources.list.d/proxmox.sources
Types: deb
URIs: http://download.proxmox.com/debian/pve
Suites: trixie
Components: pve-no-subscription
Signed-By: /usr/share/keyrings/proxmox-archive-keyring.gpg

With a subscription, use URIs: https://enterprise.proxmox.com/debian/pve and Components: pve-enterprise instead. The enterprise repository is the one Proxmox recommends for production, and it needs a valid subscription key. On a node without one, disable the enterprise repository and use the no-subscription repository instead; leaving the enterprise source enabled makes apt update report repository errors.

On a hyper-converged cluster running Ceph Squid, replace the old Ceph entry with /etc/apt/sources.list.d/ceph.sources:

# /etc/apt/sources.list.d/ceph.sources
Types: deb
URIs: http://download.proxmox.com/debian/ceph-squid
Suites: trixie
Components: no-subscription
Signed-By: /usr/share/keyrings/proxmox-archive-keyring.gpg

With a subscription, use https://enterprise.proxmox.com/debian/ceph-squid and Components: enterprise. The Package Repositories page lists ceph-squid as available for Trixie and shows this same layout for ceph-tentacle. If you only connect to an external Ceph cluster (no Ceph daemons on the PVE nodes), you don’t need to switch Ceph releases: the client packages come with normal PVE updates.

About Signed-By: every deb822 stanza must point at a valid keyring, or apt update fails. Confirm the exact file names and keyring paths on the Package Repositories page before you save.

Before you add the new deb822 files, disable the old one-line entries they replace rather than leaving both active — rename or comment out pve-enterprise.list and any other legacy .list file pointing at bookworm, so you don’t end up with the same repository listed twice:

mv /etc/apt/sources.list.d/pve-enterprise.list /etc/apt/sources.list.d/pve-enterprise.list.disabled

Trixie’s APT also has apt modernize-sources, which converts one-line entries to deb822 and keeps the old files with a .bak suffix. The upgrade wiki lists it as an optional step after the upgrade. Answer n at its prompt first to preview the changes, and remember it only rewrites the format: it doesn’t check whether a Proxmox, Ceph, subscription, or third-party repository points at the right suite.

Terminal output of cat /etc/apt/sources.list.d/debian.sources and proxmox.sources in deb822 format, with the Suites: trixie line and the Signed-By keyring path highlighted

Before running apt update, grep everything one more time for leftover Bookworm references, duplicate Proxmox suites, and any third-party or Ceph repo you haven’t touched yet:

grep -rn bookworm /etc/apt/sources.list /etc/apt/sources.list.d/
grep -rln pve /etc/apt/sources.list.d/

Fix anything that still points at bookworm or duplicates a suite before you move on to Step 8.

Step 8: Run the Distribution Upgrade

apt update

Expected result: no errors, and every line references trixie. Then run:

apt dist-upgrade

Read the summary before you answer Y. If apt proposes removing proxmox-ve, answer no and see Troubleshooting below.

During the run, dpkg normally stops on config files you changed locally. With the default interactive frontend, press D at that prompt to view the diff; other debconf frontends or a noninteractive run behave differently, so run the upgrade interactively in tmux. If you never touched the file, take the maintainer’s version. If you did, keep yours and merge the new defaults by hand afterward.

Step 9: Reboot and Verify

Warning: This reboots the host. Make sure you finished the GRUB/LVM remediation from Step 5 and that you have console access ready.

reboot

After the node comes back:

pveversion
pve8to9 --full

Expected output:

pve-manager/9.x.x/xxxxxxxxxxxxxxxx (running kernel: x.x.x-x-pve)

pve8to9 is a checklist program shipped with the PVE 8.4 packages and is meant for the pre-upgrade check. If it’s still installed after the upgrade, a second run can flag leftovers, but pveversion and the web UI are the real confirmation. Clear your browser cache (Ctrl+Shift+R) before you open the web UI, or you may get a stale interface. Then check the Summary, Updates, and Repositories panels under the node.

Proxmox VE 9 web UI with the node Repositories panel open showing trixie repositories with active status indicators highlighted
Proxmox VE 9 web UI node Updates panel showing the package list and pending update count highlighted

Step 10: Confirm Quorum, Then Repeat on the Next Node

pvecm status   # expect: Quorate: Yes, and all nodes listed

Migrate guests back to the upgraded node as needed. Then repeat Steps 4 through 10 on the next node. If quorum ever drops, stop and investigate.

When the last Ceph node is done, check cluster health before changing anything:

ceph -s

Resolve any degraded PGs or down OSDs first. Then restore the flag to whatever state it was in before you started. Only if you set noout yourself in Step 6, unset it now:

ceph osd unset noout

If it was already set before you began, leave it set and hand it back to whoever owns that maintenance.

Configuration

Repository Settings

SettingPVE 8 ValuePVE 9 Value
Debian suitebookworm, bookworm-updatestrixie, trixie-updates
Security suitebookworm-securitytrixie-security
Proxmox suitebookwormtrixie
Backportsbookworm-backportsRemove
FormatOne-line .listdeb822 .sources with Signed-By

Exact suite names to include can vary by point release — confirm current guidance on the Package Repositories wiki before you finalize your config. Don’t forget to apply the same check to Ceph and any third-party repositories; it’s easy to update the Proxmox and Debian lines and forget a vendor repo still pointing at Bookworm.

/tmp on tmpfs

The upgrade wiki notes that on Trixie /tmp is now a tmpfs (RAM-backed) file system using up to 50% of memory, and that files in /tmp and /var/tmp are cleaned regularly while the system runs. Upgraded nodes can differ, so check what yours does:

findmnt /tmp

If the output shows tmpfs, any script that writes large temp files now eats RAM. On a memory-tight host, you can keep /tmp on disk:

systemctl mask tmp.mount   # takes effect at next boot

In-Place Upgrade vs. Fresh Install

PathGood ForTrade-off
In-place apt dist-upgradeMost nodes; keeps config, storage, and cluster membershipInherits years of local tweaks and cruft
Fresh PVE 9 ISO + restoreHeavily customized or messy nodesFull backup/restore cycle; cluster nodes must be removed and rejoined

For a fresh install, back up all guests to a Proxmox Backup Server or NAS drive. Then install PVE 9 from a USB stick and restore. It’s cleaner, but it eats far more of your change window.

Tips and Troubleshooting

apt wants to remove proxmox-ve

Why it happens: The upgrade wiki lists two causes. If apt prints W: (pve-apt-hook) You are attempting to remove the meta-package 'proxmox-ve'!, some packages can’t be upgraded because the Trixie Proxmox repository isn’t configured correctly. On nodes installed on top of plain Debian, the linux-image-amd64 package also conflicts with PVE 9.

Fix: Answer no. Re-check every file in /etc/apt/sources.list.d/ for the trixie Proxmox suite and run apt update until it’s error-free. If the node was installed on plain Debian, first confirm the package is present with dpkg -l linux-image-amd64 and that a Proxmox kernel is installed (dpkg -l 'proxmox-kernel-*'), then run apt remove linux-image-amd64. Then re-run apt dist-upgrade and confirm the summary no longer removes proxmox-ve.

Old container won’t start after the upgrade

Why it happens: PVE 9 only supports the unified cgroup v2 hierarchy; containers built on very old distributions may not start without it.

Fix: Upgrade the container’s distribution to a release that supports cgroup v2, or rebuild it. Next time, check the container’s OS against the compatibility list in the upgrade wiki before you upgrade the host, and test-boot questionable containers on a spare PVE 9 host first.

GPU passthrough broke after the kernel update

Why it happens: The host driver branch doesn’t support the newer kernel shipped by your target PVE 9 release.

Fix: Install a driver branch that NVIDIA lists as compatible with that kernel — check NVIDIA’s vGPU software documentation (for licensed vGPU/GRID setups) or your driver vendor’s release notes (for plain passthrough), and cross-reference the kernel version shown on the roadmap for your target release. For VFIO setup, see our GPU passthrough on Proxmox VE guide.

Node won’t boot: GRUB can’t find the disk (LVM)

Why it happens: A known GRUB issue on LVM installs.

Fix: Boot from recovery media and apply the remediation from the upgrade wiki. This is why the wiki tells you to do it before the reboot.

Custom FRR routing stopped working

Why it happens: After the first SDN apply, PVE 9 no longer uses custom OpenFabric/OSPF fabric configs from /etc/frr/frr.conf.local and disables the daemons listed in /etc/frr/daemons.

Fix: The upgrade wiki’s workaround is to create /etc/default/frr with ospfd=yes (OSPF) or fabricd=yes (OpenFabric), or move the setup to PVE 9’s SDN Fabrics, in which case remove that file again. Compare your backup from Step 5 and test before production traffic depends on it.

apt fails on disk space or SSH dropped mid-upgrade

Fix: Free at least 5 GB on /. If SSH dropped, reattach with tmux attach -t pve9upgrade. If the session is gone, run dpkg --configure -a followed by apt dist-upgrade to finish the upgrade.

Wrapping Up

Most of the real work happens before the upgrade: a clean pve8to9 run, Ceph on Squid, and the GRUB and container fixes done on PVE 8. Get those right and the dist-upgrade itself is usually quiet. Upgrade one node at a time and check quorum after each, and the cluster stays up the whole way.

If a node has years of hand-edited config, I’d lean toward a fresh install and restore over an in-place upgrade. It costs more time up front, but you stop carrying old cruft into PVE 9.

StepActionApplies To
1–2Latest 8.4, tmux, ≥5 GB freeAll nodes
3Ceph → 19.2 SquidHyper-converged Ceph
4–5pve8to9 --full, fix breaking changesAll nodes
6Evacuate guests, nooutClusters / Ceph
7–9Trixie deb822 repos, dist-upgrade, reboot, verifyAll nodes
10Confirm quorum, repeatClusters

Resources

  • Upgrade from 8 to 9 (Proxmox VE wiki)
  • Proxmox VE Roadmap and release notes
  • Package Repositories (Proxmox VE wiki)
  • Proxmox VE FAQ
  • Proxmox VE support lifecycle (forum)
  • Proxmox VE Administration Guide (PDF)

Sources