@olivierlambert I do.

I have also noticed something extremely weird.

I have 3 HDDs attached to one host.
2x2TB raid 1 (software raid done on the XCP-ng host)
1x4TB

lsblk shows:

... SNIP ...
sda                                                             8:0    0   1.8T  0 disk
├─sda2                                                          8:2    0   1.8T  0 part
└─sda1                                                          8:1    0     2G  0 part
...
sdb                                                             8:16   0   1.8T  0 disk
├─sdb2                                                          8:18   0   1.8T  0 part
└─sdb1                                                          8:17   0     2G  0 part
  └─md127                                                       9:127  0     2G  0 raid1
...
sde                                                             8:64   0   3.7T  0 disk
├─sde2                                                          8:66   0   3.7T  0 part
└─sde1                                                          8:65   0     2G  0 part
  └─md127                                                       9:127  0     2G  0 raid1

All 3 disks are passed through to a TrueNAS VM on the host, and all the data is properly stored, but I have no idea why mdadm shows that the 4TB disk is part of the raid, instead of the other one?

/dev/md127:
           Version : 1.2
     Creation Time : Sun Aug 27 14:32:08 2023
        Raid Level : raid1
        Array Size : 2094080 (2045.00 MiB 2144.34 MB)
     Used Dev Size : 2094080 (2045.00 MiB 2144.34 MB)
      Raid Devices : 2
     Total Devices : 2
       Persistence : Superblock is persistent

       Update Time : Sun Oct  8 12:07:28 2023
             State : clean
    Active Devices : 2
   Working Devices : 2
    Failed Devices : 0
     Spare Devices : 0

Consistency Policy : resync

              Name : november:swap0
              UUID : ae045fa0:74b00896:3134ede5:c837bec3
            Events : 27

    Number   Major   Minor   RaidDevice State
       0       8       65        0      active sync   /dev/sde1
       1       8       17        1      active sync   /dev/sdb1

Anyways, this doesn't seem to be the issue, since the other host which has no HDDs attached, only m.2 VM SR's, and it also took exactly 10 minutes for the toolstack to go up.

Now XO can't reach any of the hosts, even though all the VMs are up.

@olivierlambert Ah, I understand the naming convention now.

So XO, but XO is irellevant to this issue. The problem was the 10 minutes it took the toolstack to boot up, compared to the 1-2 minutes it always took.

I updated XCP-ng now, rebooted, and both hosts took 10 minutes for the stack to come up again. Any ideas what could be causing this delay and how we could troubleshoot it?

@olivierlambert Sorry for the delay, I'm not receiving emails for replies here, for some reason.

I'm using XO from the sources

Xen Orchestra, commit 3c047
xo-server 5.124.0
xo-web 5.126.0

XCP-ng was fully up to date when issues occurred, I do have a few updates pending now, but haven't rebooted since the issue:

software-version (MRO)    : product_version: 8.2.1; 
product_version_text: 8.2; 
product_version_text_short: 8.2; 
platform_name: XCP; 
platform_version: 3.2.1; 
product_brand: XCP-ng; 
build_number: release/yangtze/master/58; 
hostname: localhost; 
date: 2023-08-09; 
dbv: 0.0.1; 
xapi: 1.20;
xen: 4.13.5-9.36; 
linux: 4.19.0+1;
xencenter_min: 2.16; 
xencenter_max: 2.16;
network_backend: openvswitch; 
db_schema: 5.603
``

Hi, I use a startup script that starts VMs in a specific order, every time my main and single XCP-ng host is restarted.

The past few days I've been getting random failures, where at first the XOA VM just loses connectivity to the host toolstack, even though all VMs are up and the host is functional (I can ssh in).

The script was configured like this:

#!/bin/bash

# xe vm-list for name-label, add in start order
vms=(vm1 vm2 vm3 etc...)
wait=30s

# No need to modify below
initwait=3m
vmslength=${#vms[@]}
log=/root/scripts/startup.log

start_vm () {
   echo -n "[$(date +"[%Y-%m-%d %H:%M:%S]")] Starting $1 ... " >> ${log}
   /opt/xensource/bin/xe vm-start name-label=$1
   if [ $? -eq 0 ]
     then
       echo "Success" >> ${log}
     else
       echo "FAILED" >> ${log}
   fi

   # Wait if not the last vm
   if [ "$1" != "${vms[${vmslength}-1]}" ]
     then
       echo "Waiting ${wait}" >> ${log}
       sleep ${wait}
   fi
}

echo "[$(date +"[%Y-%m-%d %H:%M:%S]")] Running autostart script (Waiting ${initwait})" > ${log}
sleep ${initwait}

for vm in ${vms[@]}
do
  start_vm ${vm}
done

echo "[$(date +"%T")] Startup complete." >> ${log}
echo

As you can see the initwait is set to 3m, having the script wait for the XCP-ng toolstack to get ready, and I've had no issues with this config for the past year.

Now I have noticed that the toolstack takes about 10 minutes to start, where it took about 2 beforehand. I have no idea what's going wrong because I didn't do any updates in the meantime.

Does anyone have an idea where I should look to see what's causing this 10 minute hang?

Even after rebooting the host, after the XOA VM is up, it can't connect to the toolstack for some reason:
connect ETIMEDOUT host-ip:443

Update: the XOA error is due to a kernel issue. 5.10.0-25-amd64 works, 5.10.0-26-amd64 cannot connect to any XCP-ng host. This still leaves me wondering why the XCP-ng host toolstack startup time has increased so drastically.

@olivierlambert Understood, thank you.

So I don't need to enable NBD on all my VMs, although I'm curious is there any drawback or perhaps positive sides of enabling NBD for every network?

Hello,

Do all of my VMs need to use NBD in order to benefit from faster backup speeds, or only the XO VM?

The XO VM, the Hypervisor and TrueNAS are all on their own VLANs. So from my understanding, doesn't matter which VM I'm backing up on that Hypervisor, the only network that would require NBD would be the XO VM one, rather the one that talks to the Hypervisor?

Note that the XO VM resides on the same hypervisor. Now that I think of it, maybe I should use an internal network, just to skip steps 1,2,3,4?

So we still don't have a list of which "modern" GPUs support vGPU?

Use case: jellyfin/plex transcoding, Simple Windows VM workstation, etc.

@olivierlambert

Can you link the docs please?

This is what I was looking at initially:
https://xen-orchestra.com/docs/restapi.html#collections-request

I've been automating my infrastructure, and I've noticed that disk extending has become a repetitive action:

- power off vm
- extend disk
- power on vm
- extend disk
- extend lvm

The last 2 steps seem easy to implement with ansible, though the first 3 ones seem rather impossible without the API.

Any status updates on the API?