XCP-ng 8.3 betas and RCs feedback 🚀

r0ssar00

@xerxist said in XCP-ng 8.3 beta :

Which kernel are you looking at since 4.19 will be EOL in 9 months?

So, the main blocker in the way to upgrade the kernel is a kernel module we use for storage access from the VMs.

I'm curious: which module?

r0ssar00

A couple of issues in my environment:

Stats are still broken for me, even after ensuring XO was updated (from source) and that the latest patches were applied to my pool. Was there a public GH issue tracking this breakage? (I hope to use the issue to track back to the code to find out why and see if there's something unique about my environment that might be the cause)
I know pure PV VMs are entirely unsupported, so I don't expect this to go anywhere, so no big deal if this remains broken (for me anyways!) but after the latest updates XAPI throws "shadow_allocation_set ${sz}MB invalid argument" (where ${sz} depends on static memory config for that VM) when I try to start a PV VM. To the best of my understanding, shadow pages are only supported for HVM, right?

Thanks!
Kevin

stormi

@r0ssar00 You should convert PV VMs to HVM. However, if you really can't, you can switch them from pv to pv-in-pvh. It's not documented yet but this blog post will give you an example: https://xcp-ng.org/blog/2022/01/17/removing-support-for-32-bit-pv-guests/

stormi

@r0ssar00 Regarding broken stats, you could try to find logs related to them. Failures related to "stat" or "rrd" keywords, possibly. There was no github issue opened, so most details about the issue others had will only be found in this thread.

Danp

@r0ssar00 Could this be your issue with missing stats?

If statistics (all VMs and hosts) are not showing for a specific pool, check if there is a Backup network configured on your pool (setting is in the Advanced tab of the pool) and make sure XO can access all hosts of the pool via this network.

r0ssar00

@Danp this was exactly it, thanks! Although I'm 100% sure the network is accessible to both hosts (they're directly connected to each other, no switch in between, ping to/from hosts/VMs-on-either-host works just fine), clearing it resolved the problem. I tried the other two networks I have set up (one is directly connected, one goes through a switch and is shared with the rest of my LAN) and only using the primary management interface worked (which kinda defeats the purpose...).

If this strikes you as a bug, happy to collect logs/etc, just let me know what and (a pointer to) how!

r0ssar00

@stormi My inner "I want to do something slightly off the beaten path" nerd will be disappointed, but that's really all this was, just trying something different to be different for the challenge. Just thought I'd note it since it's not outside the realm of possible (in my mind anyways) that this is a symptom of potentially incomplete code ("failing to handle or improperly handling a config value" -> "is this the only place that happened?")

gb.123

@olivierlambert @stormi

I have a suggestion w.r.t interface-rename.
Can we have a screen at the installation (using iso method) which lists all the NICs and we can rename the interfaces at install time itself ?

stormi

I just pushed several updates to the xcp-ng-base repository for XCP-ng 8.3.

Most notable changes

Addition of a new SR driver, largeblock, which emulates 512B blocks for disks which only support 4KiB blocks. More at: https://xcp-ng.org/forum/topic/8901/largeblocksr-for-4kib-blocksize-disks
XO Lite updated to version 0.2.2, which changes the treeview component.
new API for PCI passthrough, that Xen Orchestra already knows how to use to let you configure a device for passthrough.
changes in the way configuration files are handled on OpenSSH. Previously, any update to sshd_config or ssh_config would overwrite any change you made to the files. This is what XenServer 8 does, but we wanted to do it differently, which is now the case. So, now, the update won't touch the files if you had modified them. To update default cipher lists, keys, algorithms, we now do so directly at build time, in the binaries. You can check this with sshd -T. If you do have local changes to these files, the update won't be able to remove the cipher lists and other definitions from the files. Instead, name_of_file.rpmnew files will be created in the /etc/ssh directory. Which means next time we update the default lists, your configuration file will still override them with the old settings. Make sure to remove the lines starting with Ciphers, MACs, KexAlgorithms and HostKeyAlgorithms if they are present.
changes in the way UEFI certificates are propagated to UEFI VMs at their first boot. In a few words: if you enabled Secure Boot for a VM but haven't setup your pool for Secure Boot first by running secureboot-certs install, then these VMs are currently in UEFI setup mode, which discards any Secure Boot setting. With the change made, Secure Boot will fail instead in this situation, thus avoiding a false sense of security. Future UI improvements in Xen Orchestra will help detecting the issue and fixing the UEFI certificates so that Secure Boot works in these VMs. You can also do it manually with: varstore-sb-state $VM_UUID user (yes, user, verbatim. It stands for user mode). This will install the necessary certificates to the VM provided you first have setup your pool for SecureBoot with secureboot-certs install.

ph7

@stormi
I can not see anything under Network throughput in XO-lite

edit:
When I hover I can see the numbers but no graph

ph7

@ph7
With another xcp-ng server that is not yet updated i can see the Network throughput graph from url/xolite.html which is on v0.2.2 (3a44b)

stormi

@gduperrey can you check?

Theoi-Meteoroi

@stormi Nice work on the PCI passthru. Very helpful. I like the treeview as well. Thanks!

gduperrey

@stormi and @ph7 I just tested with two xo-lites in 0.2.1 and 0.2.2 on just installed servers, and in both cases, the Network throughput graph seems to have lines and values.

ph7

@gduperrey
Both servers was last updated ~ 3 weeks ago

I updated the other server the same way i updated the first one and its showing the graph
The first one still no graph after another reboot.
Netdata and XO is showing the graph by the way.

Is there some log I can check

I have earlier ( a few month ago) had problems with no graphs shown at all but a change from dark to light mode and back to dark made them visible. But I dont remember what version it was and I have changed server after that.

gduperrey

@ph7 I've asked the xo-lite team if they can give us guidance on this or if they have any idea about the current state of this host.

ph7

@gduperrey
Its not importent to me, its just a simple homelab
I was just reporting if there was a bug.
The problematic "server":
Lenovo thinkcentre M715q
AMD Ryzen 5 2400GE
r8169 0000:01:00.0 eth0: RTL8168ep/8111ep

edit: thinkcentre

marcungeschikts

@ph7 : This issue seems to appear randomly. We are going to investigate further. Thank you for your feedback.

Houbsi

Greetings,

i have 8.3 running on my MS-01, it's working fine.

I upgraded my network setup with a pfsense VM and some VLAN's (still work in progress).
But, when i open my pfsense VM i'm getting this regular message:

xenstore: could not write path attr/eth1.2/ip

the 1.2 is my VLAN2 on this interface. Is this a known issue/bug? If you need some logs, then ping me

bufanda

@Houbsi Did you setup the pfsense VM according to the guide?
https://docs.xcp-ng.org/guides/pfsense/