Red Hat released RHEL 10.3 and RHEL 9.9 on 11 November. Most of the launch talk is about sealed images for image mode and the RHEL MCP server going GA. The change more people will feel next quarter got less attention. The announcement lists validated Leapp in-place upgrade paths from RHEL 9.9 to 10.3 and from RHEL 8.10 to 9.9.
If you’ve been holding RHEL 8 or 9 hosts back until there was a clear target, you now have one. The hard part is rarely a single upgrade. It’s running the pre-upgrade check on a few hundred hosts, getting a few hundred text reports back, and working out what actually stands in the way.
Here’s one way to do it with Ansible. Check the hardware first, run the Leapp analysis everywhere, roll the results into one CSV, fix the blockers that hit the most hosts, and only then upgrade, in waves.
What the new paths mean in practice
- Leapp moves one major version at a time. A RHEL 8 host headed for RHEL 10 needs two upgrades: 8.10 to 9.9, then 9.9 to 10.3. Red Hat’s supported upgrade paths page spells this out.
- The source has to be on the right minor release first. For a 10.3 target, that’s 9.9. EUS subscribers get other pairs (9.6 to 10.0 and 9.8 to 10.2), listed on the same page.
- On Intel and AMD, the RHEL 10 in-place upgrade path is listed for x86-64-v3 hardware. Older CPUs may not qualify, and neither may VMs whose hypervisor exposes a conservative CPU model.
- The pre-upgrade report doesn’t change system configuration, but Red Hat’s RHEL 9 to 10 guide warns it can use up to about 4 GB under
/var/lib/leapp. A cramped/vargets you an incomplete report.
Step 0: check the CPU before anything else
This check is cheap, and you want the answer in week one rather than week six. On RHEL 9 the dynamic loader can tell you which x86-64 levels the CPU supports. Its help output includes a line like “x86-64-v3 (supported, searched)” when the level is available. Give it a play of its own and let it fail loudly:
- name: Check that x86_64 hosts can run RHEL 10 at all
hosts: rhel9
gather_facts: true
tasks:
- name: Ask the dynamic loader which x86-64 levels this CPU supports
ansible.builtin.command: /lib64/ld-linux-x86-64.so.2 --help
register: hwcaps
changed_when: false
when: ansible_facts['architecture'] == 'x86_64'
- name: Flag hosts without x86-64-v3
ansible.builtin.assert:
that: "'x86-64-v3 (supported' in hwcaps.stdout"
fail_msg: "{{ inventory_hostname }} does not report x86-64-v3"
when: ansible_facts['architecture'] == 'x86_64'
Hosts that fail here go on a separate list. They need new hardware, a different VM CPU model, or a rebuild elsewhere. No amount of remediation will fix it.
Step 1: run the analysis everywhere
Red Hat ships Ansible roles for exactly this job. The upstream collection is infra.leapp on GitHub and Galaxy. For the 9 to 10 jump, the supported package is ansible-collection-redhat-leapp, and Red Hat’s guide for large deployments says to install and run it from a RHEL 10 control node. There are three roles: analysis, remediate and upgrade.
Analysis is the gentle one. It installs the Leapp packages if they’re missing, runs leapp preupgrade, and copies the reports back to the controller. It doesn’t fail when it finds inhibitors. Instead it sets two facts: upgrade_inhibited (true or false) and leapp_inhibitors (the matching report entries). Those two facts are what make a fleet view possible.
- name: Leapp pre-upgrade analysis, 9.9 to 10.3
hosts: rhel9
become: true
vars:
leapp_upgrade_type: cdn # or satellite, rhui, custom
leapp_preupg_opts: --target 10.3
leapp_system_roles_collection: redhat.rhel_system_roles
tasks:
- name: Run the analysis role
ansible.builtin.import_role:
name: infra.leapp.analysis
If you ever run Leapp by hand through sudo, note the SELinux detail in the guide: each leapp command needs -r unconfined_r -t unconfined_t.
Step 2: one CSV instead of a pile of reports
Each entry in a host’s leapp_inhibitors list carries a key, a title and a severity. Add a play at the end of the same run that writes them into one file and counts which titles show up most often:
- name: Turn per-host Leapp results into one fleet view
hosts: localhost
gather_facts: false
vars:
fleet: "{{ groups['rhel9'] }}"
tasks:
- name: Write one row per host and inhibitor
ansible.builtin.copy:
dest: "{{ playbook_dir }}/leapp-inhibitors.csv"
mode: "0644"
content: |
host,status,key,title
{% for h in fleet %}
{% if hostvars[h].leapp_inhibitors is not defined %}
{{ h }},no-analysis,,
{% elif hostvars[h].leapp_inhibitors | length == 0 %}
{{ h }},ready,,
{% else %}
{% for i in hostvars[h].leapp_inhibitors %}
{{ h }},blocked,{{ i.key }},"{{ i.title | replace('"', "'") }}"
{% endfor %}
{% endif %}
{% endfor %}
- name: Show which inhibitors block the most hosts
ansible.builtin.debug:
msg: >-
{{ fleet
| map('extract', hostvars, 'leapp_inhibitors')
| select('defined') | flatten
| map(attribute='title')
| community.general.counter
| dict2items | sort(attribute='value', reverse=true) }}
Two details matter. Hosts where the analysis never ran (unreachable, out of disk, broken repos) show up as no-analysis instead of vanishing from the list. Silent gaps bite later. The counter turns the reports into a ranked to-do list. If one missing answer-file entry blocks half the fleet, that’s your first fix, not the odd kernel driver on a single database server.
The counter filter comes from community.general. If you’d rather avoid the dependency, drop that task. The CSV opens fine in any spreadsheet.
Step 3: fix the big blockers, then run it again
The analysis role also writes a host_vars file per host with the known remediations for that host, plus a remediate.yml playbook that applies them. Read it before you run it. The collection’s README is clear that the remediate role can’t fix every inhibitor: some need manual work, and some can’t be fixed at all.
A few habits help:
- Put answers in
leapp_answerfileinstead of editing hosts by hand, so the answer lives in version control. - If you’ve decided to accept a specific finding, list its key in
leapp_known_inhibitors. That’s a deliberate decision, so write the reason next to it. - Re-run the analysis after every round of fixes. By default, the upgrade role checks the latest report before it starts and stops if it still lists inhibitors.
- Read the whole report, not just the inhibitors. Red Hat’s guide points out that it also lists recommended actions for the upgraded system to work properly, and that it can’t catch everything. Broken package downloads are one example it misses.
Step 4: upgrade in waves
Once a host comes back clean, move it into a group like rhel9_ready and hand it to the upgrade role. By default the role refuses to start if the last analysis found inhibitors. It reboots the host more than once, runs some basic checks on the result, and records the non-Red Hat packages that were removed during the upgrade in /etc/ansible/facts.d/missing_packages.fact. Read that file. It’s where you find out that a monitoring agent or backup client didn’t make it across.
- name: Upgrade RHEL 9.9 to 10.3 in waves
hosts: rhel9_ready
become: true
serial:
- 1
- "10%"
- "30%"
max_fail_percentage: 0
vars:
leapp_upgrade_type: cdn
leapp_upgrade_opts: --target 10.3
leapp_system_roles_collection: redhat.rhel_system_roles
tasks:
- name: Run the upgrade role
ansible.builtin.import_role:
name: infra.leapp.upgrade
- name: Refresh facts on the upgraded host
ansible.builtin.setup:
- name: Confirm the host landed on 10.3
ansible.builtin.assert:
that: ansible_facts['distribution_version'] == '10.3'
The serial list starts with one host, then 10 percent, then 30 percent, and max_fail_percentage: 0 stops the run at the first broken host. If you read our post on wave reboots for CVE-2026-98432, it’s the same idea, with a longer and much less reversible step in the middle.
Don’t underestimate that middle step. Leapp replaces the OS in place, so take a VM snapshot, or a backup you’ve actually tested restoring, before each wave, and drain the host from whatever load balancer sits in front of it. The upgrade role doesn’t touch third-party products, EPEL packages or clustered software stacks either, so plan that work on its own.
A rough plan for the next few weeks
- Week 1: CPU check and analysis on every RHEL 8 and 9 host. No changes, just data.
- Week 2: fix the top three or four inhibitors from the CSV, then re-run the analysis.
- Week 3: a canary wave on test and low-risk hosts. Check the applications, the agents and
missing_packages.fact. - After that: production in waves, with RHEL 8 hosts doing their 8.10 to 9.9 hop first.
A fleet-wide CSV of what blocks the upgrade turns “we should get to RHEL 10 at some point” into a list with owners and numbers. That list is usually what gets the project moving.
Sources
- Red Hat: Introducing Red Hat Enterprise Linux 10.3 and 9.9 (11 Nov 2026)
- Red Hat docs: Supported upgrade paths, RHEL 9 to RHEL 10
- Red Hat docs: RHEL 10.3 in-place upgrade notes
- Red Hat docs: Reviewing the pre-upgrade report
- Red Hat docs: Upgrading large deployments by using Ansible roles
- redhat-cop/infra.leapp on GitHub



Leave a Reply