Add MIG overlay support with configuration and service files

This commit is contained in:
Josh Burks
2026-01-30 14:48:05 -07:00
parent 1e87f43dfe
commit cc20a65848
6 changed files with 201 additions and 0 deletions

View File

@@ -0,0 +1,112 @@
package mig
import (
"bytes"
"testing"
"github.com/stretchr/testify/assert"
"github.com/warewulf/warewulf/internal/app/wwctl/overlay/show"
"github.com/warewulf/warewulf/internal/pkg/testenv"
"github.com/warewulf/warewulf/internal/pkg/wwlog"
)
func Test_migOverlay(t *testing.T) {
env := testenv.New(t)
defer env.RemoveAll()
env.ImportFile("etc/warewulf/nodes.conf", "nodes.conf")
env.ImportFile("var/lib/warewulf/overlays/mig/rootfs/etc/systemd/system/nvidia-mig.service.ww", "../rootfs/etc/systemd/system/nvidia-mig.service.ww")
tests := []struct {
name string
args []string
log string
}{
{
name: "mig:nvidia-mig.service (default)",
args: []string{"--render", "node1", "mig", "etc/systemd/system/nvidia-mig.service.ww"},
log: migServiceDefault,
},
{
name: "mig:nvidia-mig.service (per-gpu)",
args: []string{"--render", "node2", "mig", "etc/systemd/system/nvidia-mig.service.ww"},
log: migServicePerGpu,
},
{
name: "mig:nvidia-mig.service (custom)",
args: []string{"--render", "node3", "mig", "etc/systemd/system/nvidia-mig.service.ww"},
log: migServiceCustom,
},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
cmd := show.GetCommand()
cmd.SetArgs(tt.args)
stdout := bytes.NewBufferString("")
stderr := bytes.NewBufferString("")
logbuf := bytes.NewBufferString("")
cmd.SetOut(stdout)
cmd.SetErr(stderr)
wwlog.SetLogWriter(logbuf)
err := cmd.Execute()
assert.NoError(t, err)
assert.Empty(t, stdout.String())
assert.Empty(t, stderr.String())
assert.Equal(t, tt.log, logbuf.String())
})
}
}
const migServiceDefault string = `backupFile: true
writeFile: true
Filename: etc/systemd/system/nvidia-mig.service
[Unit]
DefaultDependencies=no
Description=Configure NVIDIA MIG (Multi-Instance GPU) partitions at boot time
Before=nvidia-persistenced.service nvidia-dcgm.service dcgm_exporter.service
[Service]
Type=oneshot
RemainAfterExit=yes
ExecStart=/usr/local/sbin/_config_mig "2,2"
ExecStop=sh -c 'nvidia-smi mig -dci ; nvidia-smi mig -dgi ; nvidia-smi -mig 0'
[Install]
WantedBy=multi-user.target
`
const migServicePerGpu string = `backupFile: true
writeFile: true
Filename: etc/systemd/system/nvidia-mig.service
[Unit]
DefaultDependencies=no
Description=Configure NVIDIA MIG (Multi-Instance GPU) partitions at boot time
Before=nvidia-persistenced.service nvidia-dcgm.service dcgm_exporter.service
[Service]
Type=oneshot
RemainAfterExit=yes
ExecStart=/usr/local/sbin/_config_mig "14,14,14,15 7,7,7,8"
ExecStop=sh -c 'nvidia-smi mig -dci ; nvidia-smi mig -dgi ; nvidia-smi -mig 0'
[Install]
WantedBy=multi-user.target
`
const migServiceCustom string = `backupFile: true
writeFile: true
Filename: etc/systemd/system/nvidia-mig.service
[Unit]
DefaultDependencies=no
Description=Configure NVIDIA MIG (Multi-Instance GPU) partitions at boot time
Before=nvidia-persistenced.service nvidia-dcgm.service dcgm_exporter.service
[Service]
Type=oneshot
RemainAfterExit=yes
ExecStart=/usr/local/sbin/_config_mig "1,1,1,1"
ExecStop=sh -c 'nvidia-smi mig -dci ; nvidia-smi mig -dgi ; nvidia-smi -mig 0'
[Install]
WantedBy=multi-user.target
`

View File

@@ -0,0 +1,8 @@
nodes:
node1: {}
node2:
tags:
gpuMigType: "14,14,14,15 7,7,7,8"
node3:
tags:
gpuMigType: "1,1,1,1"

View File

@@ -0,0 +1 @@
../nvidia-mig.service

View File

@@ -0,0 +1,13 @@
[Unit]
DefaultDependencies=no
Description=Configure NVIDIA MIG (Multi-Instance GPU) partitions at boot time
Before=nvidia-persistenced.service nvidia-dcgm.service dcgm_exporter.service
[Service]
Type=oneshot
RemainAfterExit=yes
ExecStart=/usr/local/sbin/_config_mig "{{ .Tags.gpuMigType | default "2,2" }}"
ExecStop=sh -c 'nvidia-smi mig -dci ; nvidia-smi mig -dgi ; nvidia-smi -mig 0'
[Install]
WantedBy=multi-user.target

View File

@@ -0,0 +1,37 @@
#!/usr/bin/bash
set -euo pipefail
# This script configures NVIDIA MIG (Multi-Instance GPU) partitions based on the provided
# gpuMigType string argument. It supports both a single configuration for all GPUs or
# individual configurations per GPU.
# Example inputs:
# "14,14,14,15" # single config for all GPUs
# "14,14,14,15 7,7,7,8" # per-GPU configs for GPU0 and GPU1
# See https://docs.nvidia.com/datacenter/tesla/mig-user-guide/supported-mig-profiles.html for details on supported MIG profiles
# Get the gpuMigType string from the first argument
MIG_STRING="${1}"
if [[ -z "$MIG_STRING" ]]; then
echo "Usage: $0 <gpuMigType string>" >&2
exit 1
fi
# Check if nvidia-smi command exists
if ! command -v nvidia-smi &> /dev/null; then
echo "Error: nvidia-smi command not found" >&2
exit 1
fi
# enable MIG mode
nvidia-smi -mig 1
# per-GPU space-separated list (one token per GPU), otherwise single-string common case
if [[ "$MIG_STRING" == *" "* ]]; then
read -a PER_GPU_MIG_CONFIGS <<< "$MIG_STRING"
for gpu_index in "${!PER_GPU_MIG_CONFIGS[@]}"; do
nvidia-smi mig -cgi "${PER_GPU_MIG_CONFIGS[$gpu_index]}" -i "$gpu_index" -C
done
else
nvidia-smi mig -cgi "$MIG_STRING" -C
fi

View File

@@ -0,0 +1,30 @@
#!/usr/bin/bash
set -euo pipefail
# Generates NVIDIA MIG device entries for Slurm's gres.conf.
# Queries MIG GPU instances and maps them to /dev/nvidia-caps/ device files.
#
# Process: nvidia-smi -> extract GPU/GI/profile -> lookup minor in
# /proc/driver/nvidia-caps/mig-minors -> map to device file -> output gres format
if ! command -v nvidia-smi &> /dev/null; then
echo "Error: nvidia-smi command not found" >&2
exit 1
fi
hostname=$(hostname -s)
nvidia-smi mig -lgi | grep -E '^\|' | grep -v '===' | grep -v 'GPU.*Name' | while read line; do
gpu=$(echo "$line" | awk '{print $2}')
name=$(echo "$line" | awk '{print $4}')
gi=$(echo "$line" | awk '{print $6}') # Instance ID is column 6
[[ "$gpu" =~ ^[0-9]+$ ]] || continue
pattern="gpu${gpu}/gi${gi}/access"
minor=$(grep "^${pattern} " /proc/driver/nvidia-caps/mig-minors | awk '{print $2}')
echo "NodeName=${hostname} Name=gpu Type=${name} File=/dev/nvidia-caps/nvidia-cap${minor}"
done