RHEL 9 support

[Obihoernchen]

This issue tracks the current status of Enterprise Linux 9 support in xCAT. It covers RHEL9 and all EL9 distros like alma, rocky etc.

Known issues

• Use 'nmcli con modify' for nicextraparams on RHEL9 #7444
  • Fix configib for EL9 #7453
• Add support for static IPv6 addresses for EL9 #7447
• initscripts dependency missing. See: /etc/init.d/xcatd missing the right path for RH9 (xCAT 2.16.5) #7343
  • RHEL 9.0 statelite support requires /etc/init.d directory to exist in the rootimg to work #7341
• Add osimage symlinks for EL9 distros #7419
• Add more disk infos for copycds #7420

Feel free to report additional issues.

[lbgracioso]

Hi @Obihoernchen

I know this is about RHEL9 but I would like to add that EL9 based distros are also having issues.
genimage doesn't work in Rocky 9 (and I assume it doesn't work for OL and Alma either)

What I did to fix it:
I had to create a symbolic link for the following files and change their names to follow the naming pattern of each distribution in their respective folders (/opt/xcat/share/xcat/netboot/<distro>, /opt/xcat/share/xcat/install/<distro>).

/opt/xcat/share/xcat/netboot/rh/compute.rhels9.x86_64.pkglist
/opt/xcat/share/xcat/netboot/rh/compute.rhels9.x86_64.postinstall
/opt/xcat/share/xcat/netboot/rh/service.rhels9.x86_64.exlist
/opt/xcat/share/xcat/netboot/rh/service.rhels9.x86_64.otherpkgs.pkglist
/opt/xcat/share/xcat/netboot/rh/service.rhels9.x86_64.pkglist
/opt/xcat/share/xcat/netboot/rh/service.rhels9.x86_64.postinstall
/opt/xcat/share/xcat/install/rh/compute.rhels9.pkglist
/opt/xcat/share/xcat/install/rh/compute.rhels9.tmpl
/opt/xcat/share/xcat/install/rh/service.rhels9.pkglist
/opt/xcat/share/xcat/install/rh/service.rhels9.tmpl
/opt/xcat/share/xcat/install/rh/service.rhels9.x86_64.otherpkgs.pkglist

[Obihoernchen]

@lbgracioso that's fine. This issue is about all EL9 distros. I'll clarify this.

Good catch. I'll create the missing links.
Usually I use rhels9 for all EL distros, that's why I didn't notice :D

[ddj-brown]

ibpostscripts.tar.gz
This contains two bash scripts that we're using on RHEL9.2

1. ipoib sets IP address for first IB interface with link. Presumes boot net is 172.20.0.0/16 and IPoIB is 172.25.0.0/16
2. ibpkey looks for a configured IB partition PKEY # 1 and creates a subinterface, requesting IP from DHCP server on master node.
   We have MOFED installed, only real dependency is use of ibdev2netdev.

[samveen]

@Obihoernchen As per the github workflow cofiguration, the PR test checks the changes in the ubuntu 20.04 environment, which may not catch issues against RHEL and derivative environments. Can the PR testing be extended to also include any one of the RHEL and derivatives environment. Never mind. Only ubuntu is available as a runner as per the docs, unless self hosting the runners

[Obihoernchen]

yes unfortunately :(
But the testing in the background is way bigger testing all different OSs already.
And also way more tests with real OS deployment etc.

[ZAM-]

Is this change only going to be for RHEL 9.x+ or earlier versions as well?

[samveen]

@ZAM- RHEL/Centos 7 family and RHEL/Centos/Rocky 8 family are already very well supported. This ticket is to track extending that support to include the RHEL9 and derivatives.

[rlcto]

@Obihoernchen Hi Markus, one question we have, and I think @ZAM- was alluding to, is how we approach fixes related to networking. By that I mean, a lot of the tools in xCAT were written to use the now deprecated ifcg commands and directory structure, e.g. ifcfg files vs. keyfiles. My assumption is that we’re not replacing the ifcfg nomenclature but rather including the keyfile nomenclature. I also assume that we would be doing a check to see if the host is RHEL9. Do you have an example of how you have addressed this already if you have?

[Obihoernchen]

@rlcto @ZAM- Yes, often xCAT is using nmcli already, but in some parts (for instance nics.nicextraparams and I think IPv6) it relies on ifcfg files. In my opinion it would be best to use nmcli for everything and not rely on files at all.
To keep it simple, I guess current logic for EL<8 can stay as is, just add a special case for EL>=9 with full nmcli support if needed.
So far 90% is working with current code already (on EL9), just some minor parts are not using nmcli yet or use nmcli, but still rely on ifcfg files, too.
OS detection logic is included in networks scritps already. For instance: https://github.com/xcat2/xcat-core/blob/master/xCAT/postscripts/configeth#L706
Perhaps it is also easier to simply recreate the logic currently applied to the ifcfg files for the new keyfiles. But I would assume using nmcli should be easier.

[alexrichert]

@Obihoernchen in what you're envisioning as far as using nmcli for EL9, would the idea be that the user-provided settings would follow the current syntax/setting names/etc., or would the idea be that the user configs would directly specify nmcli arguments?

[Obihoernchen]

@alexrichert In my opinion everything in the nics table should stay as is. 90% of it is using nmcli already.
But there are some expections. For instance nics.nicextraparams relies on ifcfg files. For this nics column users should just be able to input any nmcli connection key-value pair into this field and the network scripts should just pass this 1:1 to nmcli I guess.
Furthermore even when using nmcli the scripts still write ifcfg files. This shouldn't happen on EL9+.

There might be other instances not using nmcli yet though. For now I only know about nics.nicextraparams.

[ocfmatt]

If you want to render something as a file or amend a file you could consider using keyfile format output before issuing a nmcli con reload command to make it active on a managed host. The use of a keyfile will help with extraparams logic but it will, as with ifcfg, rely on sane user input.

Network manager has been recommended by Red Hat since the introduction of RHEL8 and should, in my opinion, be the go to network management utility in EL8+ but I know some are familiar and reluctant to move away from ifcfg.

1. https://www.redhat.com/en/blog/rhel-9-networking-say-goodbye-ifcfg-files-and-hello-keyfiles
2. https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/8/html/configuring_and_managing_networking/assembly_networkmanager-connection-profiles-in-keyfile-format_configuring-and-managing-networking#proc_manually-creating-a-networkmanager-profile-in-keyfile-format_assembly_networkmanager-connection-profiles-in-keyfile-format

[LukeLR]

Is installing RHEL 9 already possible using xcat? Or do these issues still prevent an automated installation? We are still on CentOS 7 and want to switch to RockyLinux 9, but are waiting for xcat support. Thanks a lot!

[Obihoernchen]

Yes it's already possible with 2.16.5, but you might hit one of the confignetwork related issues mentioned above.
But overall stateless/stateful installation is working just fine.

[ddj-brown]

We have our cluster of 400+ nodes booted off rhel9 image from xcat server.  We replaced some of the postscripts to deal with NetworkManager, etc. There was a bit of a learning curve but nothing insurmountable. — setting static address on boot nic after getting dhcp— setting ipoib address based on boot nic address— adding a separate IB fabric key and requesting dhcp address on that subnet. I’m willing to share the scripts, let me know.   -- ddjDave ***@***.*** Jul 17, 2024, at 7:53 AM, L ***@***.***> wrote: Is installing RHEL 9 already possible using xcat? Or do these issues still prevent an automated installation? We are still on CentOS 7 and want to switch to RockyLinux 9, but are waiting for xcat support. Thanks a lot! —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you commented.Message ID: ***@***.***>