Recommendations for configuring custom images in Compute

If you work with custom images, we recommend optimizing them so your boot disks and virtual machines (VMs) work with higher productivity and speed. When you create a VM based on a custom image, this VM doesn’t include some of the settings that a VM based on a public image includes. You can add these settings to your custom-image VM and optimize it. The additional settings aren’t required for every custom image, but we recommend them for at least production-oriented images unless you have workload-specific reasons for not using these settings.

Install the monitoring agent by Nebius AI Cloud

The monitoring agent is installed by default on VMs with public images. If you use a custom image, install the agent to collect and view metrics for the VM. To install the monitoring agent, connect to the VM and then run the following commands:

sudo curl https://dr.nebius.cloud/public.gpg -o /etc/apt/keyrings/nebius-public.gpg.pub

echo deb [signed-by=/etc/apt/keyrings/nebius-public.gpg.pub] https://dr.nebius.cloud/ stable main | sudo tee /etc/apt/sources.list.d/nebius-public.list

sudo apt-get update
sudo apt-get install -y nebius-observability-agent nebius-observability-agent-updater

Collect serial logs

To collect and view serial logs of a VM on a custom image, configure the VM’s GRUB to initialize and use a serial port. GRUB is a boot loader package that boots an operating system and kernel configuration. To configure GRUB for Ubuntu-based disk images:

Connect to the VM.
In the /etc/default/grub file, add the following line:
GRUB_CMDLINE_LINUX="console=tty1 console=ttyS0"
Update GRUB:
update-grub
Reboot the VM: disconnect from it, and stop and start it.

Change sysctl settings

Apply the recommended sysctl settings to improve the networking and kernel characteristics of the VM. To do so, connect to the VM and update the following settings in two files:

File /etc/sysctl.d/30-network.conf

# Decrease default tcp keepalive time
net.ipv4.tcp_keepalive_time = 120
net.ipv4.tcp_keepalive_intvl = 60
net.ipv4.tcp_keepalive_probes = 4
# Make conntrack more liberal to tcp out of window packets to prevent
# spurious connection resets when masquerading is in use
net.netfilter.nf_conntrack_tcp_be_liberal = 1

File /etc/sysctl.d/90-kernel.conf

kernel.panic = 10
kernel.printk = 7 4 1 7

Adjust the virtiofs settings

To speed up reading from volumes and writing to them, adjust the virtiofs settings. Connect to the VM and run the Bash script below. The script increases the read_ahead_kb value to improve the performance of the VM volumes.

#!/usr/bin/env bash
set -euo pipefail

install -d /usr/local/bin
install -d /etc/udev/rules.d

cat > /usr/local/bin/tune_virtiofs_bdi.sh <<'EOF'
#!/usr/bin/env bash

kernel="${1:-}"

while read -r dev on mp type fs opts; do
    dev=$(/bin/mountpoint -d "$mp")
    if [[ "$fs" == "virtiofs" ]]; then
        if [[ "$kernel" == "$dev" ]]; then
            exit 0
        fi
    fi
done < <(mount | grep virtiofs)

exit 1
EOF

chmod +x /usr/local/bin/tune_virtiofs_bdi.sh

cat > /etc/udev/rules.d/99-virtiofs.rules <<'EOF'
SUBSYSTEM=="bdi", ACTION=="add", PROGRAM="/usr/local/bin/tune_virtiofs_bdi.sh $kernel", ATTR{read_ahead_kb}="8192"
EOF

Enable failure reporting by using pvpanic

The pvpanic tool allows the guest OS kernel to report panic and crash events to the hypervisor. This helps with failure detection and troubleshooting. To start using pvpanic on Ubuntu, connect to the VM and install the package:

sudo apt-get install -y linux-image-generic

Virtual machines

Slurm and Soperator in Nebius AI Cloud

Managed Service for Kubernetes®

Recommendations for configuring custom images in Compute

Install the monitoring agent by Nebius AI Cloud

Collect serial logs

Change sysctl settings

Adjust the virtiofs settings

Enable failure reporting by using pvpanic

​Install the monitoring agent by Nebius AI Cloud

​Collect serial logs

​Change sysctl settings

​Adjust the virtiofs settings

​Enable failure reporting by using pvpanic

Install the monitoring agent by Nebius AI Cloud

Collect serial logs

Change sysctl settings

Adjust the virtiofs settings

Enable failure reporting by using pvpanic