mirror of https://git.collinwebdesigns.de/oscar.krause/fastapi-dls.git synced 2024-12-27 14:09:33 +03:00

Oscar Krause ea8a66d449 removed "PatchMalformedJsonMiddleware" because its not working on driver site

ref. oscar.krause/fastapi-dls#1

2024-12-02 08:37:01 +01:00

35 KiB

Raw Blame History

FastAPI-DLS

Minimal Delegated License Service (DLS).

Compatibility tested with official NLS 2.0.1, 2.1.0, 3.1.0, 3.3.1. For Driver compatibility see compatibility matrix.

This service can be used without internet connection. Only the clients need a connection to this service on configured port.

Official Links

https://git.collinwebdesigns.de/oscar.krause/fastapi-dls (Private Git)
https://gitea.publichub.eu/oscar.krause/fastapi-dls (Public Git)
https://hub.docker.com/r/collinwebdesigns/fastapi-dls (Docker-Hub collinwebdesigns/fastapi-dls:latest)

All other repositories are forks! (which is no bad - just for information and bug reports)

Releases & Release Notes

Further Reading

NVIDIA vGPU Guide - This document serves as a guide to install NVIDIA vGPU host drivers on the latest Proxmox VE version
vgpu_unlock - Unlock vGPU functionality for consumer-grade Nvidia GPUs.
vGPU_Unlock Wiki - Guide for vgpu_unlock
Proxmox All-In-One Installer Script - Also known as proxmox-installer.sh

[TOC]

Setup (Service)

System requirements

256mb ram
4gb hdd
maybe IPv6 must be disabled

Tested with Ubuntu 22.10 (EOL!) (from Proxmox templates), actually its consuming 100mb ram and 750mb hdd.

Prepare your system

Make sure your timezone is set correct on you fastapi-dls server and your client

This guide does not show how to install vGPU host drivers! Look at the official documentation packed with the driver releases.

Docker

Docker-Images are available here for Intel (x86), AMD (amd64) and ARM (arm64):

Docker-Hub: collinwebdesigns/fastapi-dls:latest
GitLab-Registry: registry.git.collinwebdesigns.de/oscar.krause/fastapi-dls:latest

The images include database drivers for postgres, mariadb and sqlite.

Run this on the Docker-Host

WORKING_DIR=/opt/docker/fastapi-dls/cert
mkdir -p $WORKING_DIR
cd $WORKING_DIR
# create instance private and public key for singing JWT's
openssl genrsa -out $WORKING_DIR/instance.private.pem 2048 
openssl rsa -in $WORKING_DIR/instance.private.pem -outform PEM -pubout -out $WORKING_DIR/instance.public.pem
# create ssl certificate for integrated webserver (uvicorn) - because clients rely on ssl
openssl req -x509 -nodes -days 3650 -newkey rsa:2048 -keyout  $WORKING_DIR/webserver.key -out $WORKING_DIR/webserver.crt

Start container

To test if everything is set up properly you can start container as following:

docker volume create dls-db
docker run -e DLS_URL=`hostname -i` -e DLS_PORT=443 -p 443:443 -v $WORKING_DIR:/app/cert -v dls-db:/app/database collinwebdesigns/fastapi-dls:latest

Docker-Compose / Deploy stack

See examples directory for more advanced examples (with reverse proxy usage).

Adjust REQUIRED variables as needed

version: '3.9'

x-dls-variables: &dls-variables
  TZ: Europe/Berlin # REQUIRED, set your timezone correctly on fastapi-dls AND YOUR CLIENTS !!!
  DLS_URL: localhost # REQUIRED, change to your ip or hostname
  DLS_PORT: 443
  LEASE_EXPIRE_DAYS: 90  # 90 days is maximum
  DATABASE: sqlite:////app/database/db.sqlite
  DEBUG: false

services:
  dls:
    image: collinwebdesigns/fastapi-dls:latest
    restart: always
    environment:
      <<: *dls-variables
    ports:
      - "443:443"
    volumes:
      - /opt/docker/fastapi-dls/cert:/app/cert
      - dls-db:/app/database
    logging:  # optional, for those who do not need logs
      driver: "json-file"
      options:
        max-file: 5
        max-size: 10m
            
volumes:
  dls-db:

Debian / Ubuntu / macOS (manual method using `git clone` and python virtual environment)

Tested on Debian 11 (bullseye), Debian 12 (bookworm) and macOS Ventura (13.6), Ubuntu may also work. Please note that setup on macOS differs from Debian based systems.

Make sure you are logged in as root.

Install requirements

apt-get update && apt-get install git python3-venv python3-pip

Install FastAPI-DLS

WORKING_DIR=/opt/fastapi-dls
mkdir -p $WORKING_DIR
cd $WORKING_DIR
git clone https://git.collinwebdesigns.de/oscar.krause/fastapi-dls .
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
deactivate
chown -R www-data:www-data $WORKING_DIR

Create keypair and webserver certificate

WORKING_DIR=/opt/fastapi-dls/app/cert
mkdir -p $WORKING_DIR
cd $WORKING_DIR
# create instance private and public key for singing JWT's
openssl genrsa -out $WORKING_DIR/instance.private.pem 2048 
openssl rsa -in $WORKING_DIR/instance.private.pem -outform PEM -pubout -out $WORKING_DIR/instance.public.pem
# create ssl certificate for integrated webserver (uvicorn) - because clients rely on ssl
openssl req -x509 -nodes -days 3650 -newkey rsa:2048 -keyout  $WORKING_DIR/webserver.key -out $WORKING_DIR/webserver.crt
chown -R www-data:www-data $WORKING_DIR

Test Service

This is only to test whether the service starts successfully.

cd /opt/fastapi-dls/app
sudo -u www-data /opt/fastapi-dls/venv/bin/uvicorn main:app --app-dir=/opt/fastapi-dls/app
# or
su - www-data -c "/opt/fastapi-dls/venv/bin/uvicorn main:app --app-dir=/opt/fastapi-dls/app"

Create config file

Adjust DLS_URL as needed (accessing from LAN won't work with 127.0.0.1)

mkdir /etc/fastapi-dls
cat <<EOF >/etc/fastapi-dls/env
DLS_URL=127.0.0.1
DLS_PORT=443
LEASE_EXPIRE_DAYS=90
DATABASE=sqlite:////opt/fastapi-dls/app/db.sqlite

EOF

Create service

cat <<EOF >/etc/systemd/system/fastapi-dls.service
[Unit]
Description=Service for fastapi-dls
After=network.target

[Service]
User=www-data
Group=www-data
AmbientCapabilities=CAP_NET_BIND_SERVICE
WorkingDirectory=/opt/fastapi-dls/app
EnvironmentFile=/etc/fastapi-dls/env
ExecStart=/opt/fastapi-dls/venv/bin/uvicorn main:app \\
  --env-file /etc/fastapi-dls/env \\
  --host \$DLS_URL --port \$DLS_PORT \\
  --app-dir /opt/fastapi-dls/app \\
  --ssl-keyfile /opt/fastapi-dls/app/cert/webserver.key \\
  --ssl-certfile /opt/fastapi-dls/app/cert/webserver.crt \\
  --proxy-headers
Restart=always
KillSignal=SIGQUIT
Type=simple
NotifyAccess=all

[Install]
WantedBy=multi-user.target

EOF

Now you have to run systemctl daemon-reload. After that you can start service with systemctl start fastapi-dls.service and enable autostart with systemctl enable fastapi-dls.service.

openSUSE Leap (manual method using `git clone` and python virtual environment)

Tested on openSUSE Leap 15.4, openSUSE Tumbleweed may also work.

Install requirements

zypper in -y python310 python3-virtualenv python3-pip

Install FastAPI-DLS

BASE_DIR=/opt/fastapi-dls
SERVICE_USER=dls
mkdir -p ${BASE_DIR}
cd ${BASE_DIR}
git clone https://git.collinwebdesigns.de/oscar.krause/fastapi-dls .
python3.10 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
deactivate
useradd -r ${SERVICE_USER} -M -d /opt/fastapi-dls
chown -R ${SERVICE_USER} ${BASE_DIR}

Create keypair and webserver certificate

CERT_DIR=${BASE_DIR}/app/cert
SERVICE_USER=dls
mkdir ${CERT_DIR}
cd ${CERT_DIR}
# create instance private and public key for singing JWT's
openssl genrsa -out ${CERT_DIR}/instance.private.pem 2048 
openssl rsa -in ${CERT_DIR}/instance.private.pem -outform PEM -pubout -out ${CERT_DIR}/instance.public.pem
# create ssl certificate for integrated webserver (uvicorn) - because clients rely on ssl
openssl req -x509 -nodes -days 3650 -newkey rsa:2048 -keyout  ${CERT_DIR}/webserver.key -out ${CERT_DIR}/webserver.crt
chown -R ${SERVICE_USER} ${CERT_DIR}

Test Service

This is only to test whether the service starts successfully.

BASE_DIR=/opt/fastapi-dls
SERVICE_USER=dls
cd ${BASE_DIR}
sudo -u ${SERVICE_USER} ${BASE_DIR}/venv/bin/uvicorn main:app --app-dir=${BASE_DIR}/app
# or
su - ${SERVICE_USER} -c "${BASE_DIR}/venv/bin/uvicorn main:app --app-dir=${BASE_DIR}/app"

Create config file

Adjust DLS_URL as needed (accessing from LAN won't work with 127.0.0.1)

BASE_DIR=/opt/fastapi-dls
cat <<EOF >/etc/fastapi-dls/env
DLS_URL=127.0.0.1
DLS_PORT=443
LEASE_EXPIRE_DAYS=90
DATABASE=sqlite:///${BASE_DIR}/app/db.sqlite

EOF

Create service

BASE_DIR=/opt/fastapi-dls
SERVICE_USER=dls
cat <<EOF >/etc/systemd/system/fastapi-dls.service
[Unit]
Description=Service for fastapi-dls vGPU licensing service
After=network.target

[Service]
User=${SERVICE_USER}
AmbientCapabilities=CAP_NET_BIND_SERVICE
WorkingDirectory=${BASE_DIR}/app
EnvironmentFile=/etc/fastapi-dls/env
ExecStart=${BASE_DIR}/venv/bin/uvicorn main:app \\
  --env-file /etc/fastapi-dls/env \\
  --host \$DLS_URL --port \$DLS_PORT \\
  --app-dir ${BASE_DIR}/app \\
  --ssl-keyfile ${BASE_DIR}/app/cert/webserver.key \\
  --ssl-certfile ${BASE_DIR}/app/cert/webserver.crt \\
  --proxy-headers
Restart=always
KillSignal=SIGQUIT
Type=simple
NotifyAccess=all

[Install]
WantedBy=multi-user.target

EOF

Now you have to run systemctl daemon-reload. After that you can start service with systemctl start fastapi-dls.service and enable autostart with systemctl enable fastapi-dls.service.

Debian / Ubuntu (using `dpkg` / `apt`)

Packages are available here:

GitLab-Registry

Successful tested with:

Debian 12 (Bookworm) (EOL: June 06, 2026)
Ubuntu 22.10 (Kinetic Kudu) (EOL: July 20, 2023)
Ubuntu 23.04 (Lunar Lobster) (EOL: January 2024)
Ubuntu 23.10 (Mantic Minotaur) (EOL: July 2024)
Ubuntu 24.04 (Noble Numbat) (EOL: April 2036)
Ubuntu 24.10 (Oracular Oriole) (EOL: tba.)

Not working with:

Debian 11 (Bullseye) and lower (missing python-jose dependency)
Ubuntu 22.04 (Jammy Jellyfish) (not supported as for 15.01.2023 due to fastapi - uvicorn version missmatch)

Run this on your server instance

First go to GitLab-Registry and select your version. Then you have to copy the download link of the fastapi-dls_X.Y.Z_amd64.deb asset.

apt-get update
FILENAME=/opt/fastapi-dls.deb
wget -O $FILENAME <download-url>
dpkg -i $FILENAME
apt-get install -f --fix-missing

Start with systemctl start fastapi-dls.service and enable autostart with systemctl enable fastapi-dls.service. Now you have to edit /etc/fastapi-dls/env as needed.

ArchLinux (using `pacman`)

Shout out to samicrusader who created build file for ArchLinux!

Packages are available here:

GitLab-Registry

pacman -Sy
FILENAME=/opt/fastapi-dls.pkg.tar.zst

curl -o $FILENAME <download-url>
# or
wget -O $FILENAME <download-url>

pacman -U --noconfirm fastapi-dls.pkg.tar.zst

Start with systemctl start fastapi-dls.service and enable autostart with systemctl enable fastapi-dls.service. Now you have to edit /etc/default/fastapi-dls as needed.

unRAID

Download this xml file
Put it in /boot/config/plugins/dockerMan/templates-user/
Go to Docker page, scroll down to Add Container, click on Template list and choose FastAPI-DLS
Open terminal/ssh, follow the instructions in overview description
Setup your container IP, Port, DLS_URL and DLS_PORT
Apply and let it boot up

Unraid users must also make sure they have Host access to custom networks enabled if unraid is the vgpu guest.

Continue here for docker guest setup.

NixOS

Tanks to @mrzenc for fastapi-dls-nixos.

Let's Encrypt Certificate (optional)

If you're using installation via docker, you can use traefik. Please refer to their documentation.

Note that port 80 must be accessible, and you have to install socat if you're using standalone mode.

acme.sh --issue -d example.com \
  --cert-file /etc/fastapi-dls/webserver.donotuse.crt \
  --key-file /etc/fastapi-dls/webserver.key \
  --fullchain-file /etc/fastapi-dls/webserver.crt \
  --reloadcmd "systemctl restart fastapi-dls.service"

After first success you have to replace --issue with --renew.

Configuration

Variable	Default	Usage
`DEBUG`	`false`	Toggles `fastapi` debug mode
`DLS_URL`	`localhost`	Used in client-token to tell guest driver where dls instance is reachable
`DLS_PORT`	`443`	Used in client-token to tell guest driver where dls instance is reachable
`TOKEN_EXPIRE_DAYS`	`1`	Client auth-token validity (used for authenticate client against api, not `.tok` file!)
`LEASE_EXPIRE_DAYS`	`90`	Lease time in days
`LEASE_RENEWAL_PERIOD`	`0.15`	The percentage of the lease period that must elapse before a licensed client can renew a license *1
`DATABASE`	`sqlite:///db.sqlite`	See official SQLAlchemy docs
`CORS_ORIGINS`	`https://{DLS_URL}`	Sets `Access-Control-Allow-Origin` header (comma separated string) *2
`SITE_KEY_XID`	`00000000-0000-0000-0000-000000000000`	Site identification uuid
`INSTANCE_REF`	`10000000-0000-0000-0000-000000000001`	Instance identification uuid
`ALLOTMENT_REF`	`20000000-0000-0000-0000-000000000001`	Allotment identification uuid
`INSTANCE_KEY_RSA`	`<app-dir>/cert/instance.private.pem`	Site-wide private RSA key for singing JWTs *3
`INSTANCE_KEY_PUB`	`<app-dir>/cert/instance.public.pem`	Site-wide public key *3

*1 For example, if the lease period is one day and the renewal period is 20%, the client attempts to renew its license every 4.8 hours. If network connectivity is lost, the loss of connectivity is detected during license renewal and the client has 19.2 hours in which to re-establish connectivity before its license expires.

*2 Always use https, since guest-drivers only support secure connections!

*3 If you recreate your instance keys you need to recreate client-token for each guest!

Setup (Client)

The token file has to be copied! It's not enough to C&P file contents, because there can be special characters.

This guide does not show how to install vGPU guest drivers! Look at the official documentation packed with the driver releases.

Linux

Download client-token and place it into /etc/nvidia/ClientConfigToken:

curl --insecure -L -X GET https://<dls-hostname-or-ip>/-/client-token -o /etc/nvidia/ClientConfigToken/client_configuration_token_$(date '+%d-%m-%Y-%H-%M-%S').tok
# or
wget --no-check-certificate -O /etc/nvidia/ClientConfigToken/client_configuration_token_$(date '+%d-%m-%Y-%H-%M-%S').tok https://<dls-hostname-or-ip>/-/client-token

Restart nvidia-gridd service:

service nvidia-gridd restart

Check licensing status:

nvidia-smi -q | grep "License"

Output should be something like:

vGPU Software Licensed Product
    License Status                    : Licensed (Expiry: YYYY-M-DD hh:mm:ss GMT)

Done. For more information check troubleshoot section.

Windows

Power-Shell (run as administrator!)

Download client-token and place it into C:\Program Files\NVIDIA Corporation\vGPU Licensing\ClientConfigToken:

curl.exe --insecure -L -X GET https://<dls-hostname-or-ip>/-/client-token -o "C:\Program Files\NVIDIA Corporation\vGPU Licensing\ClientConfigToken\client_configuration_token_$($(Get-Date).tostring('dd-MM-yy-hh-mm-ss')).tok"

Restart NvContainerLocalSystem service:

Restart-Service NVDisplay.ContainerLocalSystem

Check licensing status:

& 'nvidia-smi' -q  | Select-String "License"

Output should be something like:

vGPU Software Licensed Product
    License Status                    : Licensed (Expiry: YYYY-M-DD hh:mm:ss GMT)

Done. For more information check troubleshoot section.

unRAID Guest

Make sure you create a folder in a linux filesystem (BTRFS/XFS/EXT4...), I recommend /mnt/user/system/nvidia (this is where docker and libvirt preferences are saved, so it's a good place to have that)
Edit the script to put your DLS_IP, DLS_PORT and TOKEN_PATH, properly
Install User Scripts plugin from Community Apps (the Apps page, or google User Scripts Unraid if you're not using CA)
Go to Settings > Users Scripts > Add New Script
Give it a name (the name must not contain spaces preferably)
Click on the gear icon to the left of the script name then edit script
Paste the script and save
Set schedule to At First Array Start Only
Click on Apply

API Endpoints

show

GET /

Redirect to /-/readme.

GET /-/health

Status endpoint, used for healthcheck.

GET /-/config

Shows current runtime environment variables and their values.

GET /-/readme

HTML rendered README.md.

GET /-/manage

Shows a very basic UI to delete origins or leases.

GET /-/origins?leases=false

List registered origins.

Query Parameter	Default	Usage
`leases`	`false`	Include referenced leases per origin

DELETE /-/origins

Deletes all origins and their leases.

GET /-/leases?origin=false

List current leases.

Query Parameter	Default	Usage
`origin`	`false`	Include referenced origin per lease

DELETE /-/lease/{lease_ref}

Deletes an lease.

GET /-/client-token

Generate client token, (see installation).

Others

There are many other internal api endpoints for handling authentication and lease process.

Troubleshoot / Debug

Please make sure that fastapi-dls and your guests are on the same timezone!

Maybe you have to disable IPv6 on the machine you are running FastAPI-DLS.

Docker

Logs are available with docker logs <container>. To get the correct container-id use docker container ls or docker ps.

Linux

Logs are available with journalctl -u nvidia-gridd -f.

Windows

Logs are available in C:\Users\Public\Documents\Nvidia\LoggingLog.NVDisplay.Container.exe.log.

Known Issues

Linux

Invalid HTTP request

This error message: uvicorn.error:Invalid HTTP request received. can be ignored.

Ref. https://github.com/encode/uvicorn/issues/441

Log example

WARNING:uvicorn.error:Invalid HTTP request received.
Traceback (most recent call last):
  File "/usr/lib/python3/dist-packages/uvicorn/protocols/http/h11_impl.py", line 129, in handle_events
    event = self.conn.next_event()
  File "/usr/lib/python3/dist-packages/h11/_connection.py", line 485, in next_event
    exc._reraise_as_remote_protocol_error()
  File "/usr/lib/python3/dist-packages/h11/_util.py", line 77, in _reraise_as_remote_protocol_error
    raise self
  File "/usr/lib/python3/dist-packages/h11/_connection.py", line 467, in next_event
    event = self._extract_next_receive_event()
  File "/usr/lib/python3/dist-packages/h11/_connection.py", line 409, in _extract_next_receive_event
    event = self._reader(self._receive_buffer)
  File "/usr/lib/python3/dist-packages/h11/_readers.py", line 84, in maybe_read_from_IDLE_client
    raise LocalProtocolError("no request line received")
h11._util.RemoteProtocolError: no request line received

Windows

Required cipher on Windows Guests (e.g. managed by domain controller with GPO)

It is required to enable SHA1 (TLS_ECDHE_RSA_WITH_AES_256_CBC_SHA_P521) in windows cipher suite.

Multiple Display Container LS Instances

On Windows on some machines there are running two or more instances of NVIDIA Display Container LS. This causes a problem on licensing flow. As you can see in the logs below, there are two lines with NLS initialized, each prefixed with <1> and <2>. So it is possible, that daemon 1 fetches a valid license through dls-service, and daemon 2 only gets a valid local license.

Log example

Display-Container-LS

Tue Dec 20 17:25:11 2022:<1>:NLS initialized
Tue Dec 20 17:25:12 2022:<2>:NLS initialized
Tue Dec 20 17:25:16 2022:<1>:Valid GRID license not found. GPU features and performance will be restricted. To enable full functionality please configure licensing details.
Tue Dec 20 17:25:17 2022:<1>:License acquired successfully. (Info: 192.168.178.110, NVIDIA RTX Virtual Workstation; Expiry: 2022-12-21 16:25:16 GMT)
Tue Dec 20 17:25:17 2022:<2>:Valid GRID license not found. GPU features and performance will be restricted. To enable full functionality please configure licensing details.
Tue Dec 20 17:25:38 2022:<2>:License acquired successfully from local trusted store. (Info: 192.168.178.110, NVIDIA RTX Virtual Workstation; Expiry: 2022-12-21 16:25:16 GMT)

fastapi-dls

> [  origin  ]: 41720000-FA43-4000-9472-0000E8660000: {'candidate_origin_ref': '41720000-FA43-4000-9472-0000E8660000', 'environment': {'fingerprint': {'mac_address_list': ['5E:F0:79:E6:DE:E1']}, 'hostname': 'PC-Windows', 'ip_address_list': ['2003:a:142e:c800::1cc', 'fdfe:7fcd:e30f:40f5:ad5c:e67b:49a6:cfb3', 'fdfe:7fcd:e30f:40f5:6409:db1c:442b:f90b', 'fe80::a32e:f736:8988:fe45', '192.168.178.110'], 'guest_driver_version': '527.41', 'os_platform': 'Windows 10 Pro', 'os_version': '10.0.19045', 'host_driver_version': '525.60.12', 'gpu_id_list': ['1E3010DE-133210DE'], 'client_platform_id': '00000000-0000-0000-0000-000000000113', 'hv_platform': 'Unknown', 'cpu_sockets': 1, 'physical_cores': 8}, 'registration_pending': False, 'update_pending': False}
> [  origin  ]: 41720000-FA43-4000-9472-0000E8660000: {'candidate_origin_ref': '41720000-FA43-4000-9472-0000E8660000', 'environment': {'fingerprint': {'mac_address_list': ['5E:F0:79:E6:DE:E1']}, 'hostname': 'PC-Windows', 'ip_address_list': ['2003:a:142e:c800::1cc', 'fdfe:7fcd:e30f:40f5:ad5c:e67b:49a6:cfb3', 'fdfe:7fcd:e30f:40f5:6409:db1c:442b:f90b', 'fe80::a32e:f736:8988:fe45', '192.168.178.110'], 'guest_driver_version': '527.41', 'os_platform': 'Windows 10 Pro', 'os_version': '10.0.19045', 'host_driver_version': '525.60.12', 'gpu_id_list': ['1E3010DE-133210DE'], 'client_platform_id': '00000000-0000-0000-0000-000000000113', 'hv_platform': 'Unknown', 'cpu_sockets': 1, 'physical_cores': 8}, 'registration_pending': False, 'update_pending': False}
> [   code   ]: 41720000-FA43-4000-9472-0000E8660000: {'code_challenge': 'bTwcOn17SD5mtwmFdKDgufnceGXeGYcnFfMHqmjtReo', 'origin_ref': '41720000-FA43-4000-9472-0000E8660000'}
> [   code   ]: 41720000-FA43-4000-9472-0000E8660000: {'code_challenge': 'FCVDfgKmgr+lyvSpOxr4fZnDZv8VrNtNEAZPUuLAr7A', 'origin_ref': '41720000-FA43-4000-9472-0000E8660000'}
> [   auth   ]: 41720000-FA43-4000-9472-0000E8660000 (bTwcOn17SD5mtwmFdKDgufnceGXeGYcnFfMHqmjtReo): {'auth_code': 'eyJhbGciOiJSUzI1NiIsImtpZCI6IjAwMDAwMDAwLTAwMDAtMDAwMC0wMDAwLTAwMDAwMDAwMDAwMCIsInR5cCI6IkpXVCJ9.eyJpYXQiOjE2NzE1NTcwMzMsImV4cCI6MTY3MTU1NzkzMywiY2hhbGxlbmdlIjoiYlR3Y09uMTdTRDVtdHdtRmRLRGd1Zm5jZUdYZUdZY25GZk1IcW1qdFJlbyIsIm9yaWdpbl9yZWYiOiJiVHdjT24xN1NENW10d21GZEtEZ3VmbmNlR1hlR1ljbkZmTUhxbWp0UmVvIiwia2V5X3JlZiI6IjAwMDAwMDAwLTAwMDAtMDAwMC0wMDAwLTAwMDAwMDAwMDAwMCIsImtpZCI6IjAwMDAwMDAwLTAwMDAtMDAwMC0wMDAwLTAwMDAwMDAwMDAwMCJ9.m5M4h9HRYWkItHEdYGApJVM7TgBH0qyDXCxPkaG2-Km5SviRMk0_3er5Myjq3rYGlr88JBviA07Pc3cr7fV-tDAXaSGalxLNfFtVRcnzqbtgnkodep1PHRUXYkiQgfaJ36m02zZucu4qMyYfQTpZ_-x67eycFKyN9T9cRJ4PYFe5W_6_zjzz6D0qeLACDhXt4ns980URttKfn2vACE8gPP5-EC-7lSY1g1mAWJKB_X9OlYRFE2mkCxnde6z5I2qmCXE_awimkigjo5LYvDcjCz60QDsOD2Ojgz4Y9xgjPbKnup4c2orKTWLUfT8_o4toKbaSfuLzPtD-41b3E8NqHQ', 'code_verifier': 'NCkAAB0+AACEHAAAIAAAAEoWAACAGAAArGwAAOkkAABfTgAAK0oAADFiAAANXAAAHzwAAKg4AAC/GwAAkxsAAEJHAABiDwAAaC8AAFMYAAAOLAAAFUkAAEheAAALOwAAHmwAAIJtAABpKwAArmsAAGM8AABnVwAA5FkAAP8mAAA'}
> [   auth   ]: 41720000-FA43-4000-9472-0000E8660000 (FCVDfgKmgr+lyvSpOxr4fZnDZv8VrNtNEAZPUuLAr7A): {'auth_code': 'eyJhbGciOiJSUzI1NiIsImtpZCI6IjAwMDAwMDAwLTAwMDAtMDAwMC0wMDAwLTAwMDAwMDAwMDAwMCIsInR5cCI6IkpXVCJ9.eyJpYXQiOjE2NzE1NTcwMzQsImV4cCI6MTY3MTU1NzkzNCwiY2hhbGxlbmdlIjoiRkNWRGZnS21ncitseXZTcE94cjRmWm5EWnY4VnJOdE5FQVpQVXVMQXI3QSIsIm9yaWdpbl9yZWYiOiJGQ1ZEZmdLbWdyK2x5dlNwT3hyNGZabkRadjhWck50TkVBWlBVdUxBcjdBIiwia2V5X3JlZiI6IjAwMDAwMDAwLTAwMDAtMDAwMC0wMDAwLTAwMDAwMDAwMDAwMCIsImtpZCI6IjAwMDAwMDAwLTAwMDAtMDAwMC0wMDAwLTAwMDAwMDAwMDAwMCJ9.it_UKCHLLd25g19zqryZ6_ePrkHljXJ3uX-hNdu-pcmnYD9ODOVl2u5bRxOrP6S2EUO4WLZIuvLOhbFBUHfZfXFRmmCv4NDJoZx36Qn6zszePK9Bngej40Qf8Wu3JGXMVrwfC6WNW6WFeUT-s9jos5e1glFk_E3ZhOYQjXljWOcfcNvZ-PVJFBi5OzyQqLuL43GQH_PSF66N2gq0OyKgxTvg2q6SzGD3YAxsbjy2mD0YOUv8pW8Dr_9L4hmnNHg2DdM_lCwmy4qIBaDkAQDq8VCw1-4RcXROiLlYwhvHRalsXnmREPXaOUiUrr8rrCX8jgc7Fcd1uhY5jnouWbwEAg', 'code_verifier': 'tFAAAKQSAAAqOQAAhykAANJxAAA9PQAAyFwAALNsAAB/VQAA4GQAAB5fAAA2JgAApWIAAKMeAAB3YwAAggQAAPsEAAAuAgAAblIAABR/AAAfAgAAenoAAKZ3AABUTQAA5CQAANkTAAC8JwAAvUQAAO0yAAA3awAAegIAAD1iAAA'}
> [  leases  ]: 41720000-FA43-4000-9472-0000E8660000 (bTwcOn17SD5mtwmFdKDgufnceGXeGYcnFfMHqmjtReo): found 0 active leases
> [  leases  ]: 41720000-FA43-4000-9472-0000E8660000 (FCVDfgKmgr+lyvSpOxr4fZnDZv8VrNtNEAZPUuLAr7A): found 0 active leases
> [  create  ]: 41720000-FA43-4000-9472-0000E8660000 (bTwcOn17SD5mtwmFdKDgufnceGXeGYcnFfMHqmjtReo): create leases for scope_ref_list ['1e9335d0-049d-48b2-b719-e551c859f9f9']

in comparison to linux

nvidia-grid.service

Dec 20 17:53:32 ubuntu-grid-server nvidia-gridd[10354]: vGPU Software package (0)
Dec 20 17:53:32 ubuntu-grid-server nvidia-gridd[10354]: Ignore service provider and node-locked licensing
Dec 20 17:53:32 ubuntu-grid-server nvidia-gridd[10354]: NLS initialized
Dec 20 17:53:32 ubuntu-grid-server nvidia-gridd[10354]: Acquiring license. (Info: 192.168.178.110; NVIDIA RTX Virtual Workstation)
Dec 20 17:53:34 ubuntu-grid-server nvidia-gridd[10354]: License acquired successfully. (Info: 192.168.178.110, NVIDIA RTX Virtual Workstation; Expiry: 2022-12-21 16:53:33 GMT)

fastapi-dls

> [  origin  ]: B210CF72-FEC7-4440-9499-1156D1ACD13A: {'candidate_origin_ref': 'B210CF72-FEC7-4440-9499-1156D1ACD13A', 'environment': {'fingerprint': {'mac_address_list': ['d6:30:d8:de:46:a7']}, 'hostname': 'ubuntu-grid-server', 'ip_address_list': ['192.168.178.114', 'fdfe:7fcd:e30f:40f5:d430:d8ff:fede:46a7', '2003:a:142e:c800::642', 'fe80::d430:d8ff:fede:46a7%ens18'], 'guest_driver_version': '525.60.13', 'os_platform': 'Ubuntu 20.04', 'os_version': '20.04.5 LTS (Focal Fossa)', 'host_driver_version': '525.60.12', 'gpu_id_list': ['1E3010DE-133210DE'], 'client_platform_id': '00000000-0000-0000-0000-000000000105', 'hv_platform': 'LINUX_KVM', 'cpu_sockets': 1, 'physical_cores': 16}, 'registration_pending': False, 'update_pending': False}
> [   code   ]: B210CF72-FEC7-4440-9499-1156D1ACD13A: {'code_challenge': 'hYSKI4kpZcWqPatM5Sc9RSCuzMeyz2piTmrRQKnnHro', 'origin_ref': 'B210CF72-FEC7-4440-9499-1156D1ACD13A'}
> [   auth   ]: B210CF72-FEC7-4440-9499-1156D1ACD13A (hYSKI4kpZcWqPatM5Sc9RSCuzMeyz2piTmrRQKnnHro): {'auth_code': 'eyJhbGciOiJSUzI1NiIsImtpZCI6IjAwMDAwMDAwLTAwMDAtMDAwMC0wMDAwLTAwMDAwMDAwMDAwMCIsInR5cCI6IkpXVCJ9.eyJpYXQiOjE2NzE1NTUyMTIsImV4cCI6MTY3MTU1NjExMiwiY2hhbGxlbmdlIjoiaFlTS0k0a3BaY1dxUGF0TTVTYzlSU0N1ek1leXoycGlUbXJSUUtubkhybyIsIm9yaWdpbl9yZWYiOiJoWVNLSTRrcFpjV3FQYXRNNVNjOVJTQ3V6TWV5ejJwaVRtclJRS25uSHJvIiwia2V5X3JlZiI6IjAwMDAwMDAwLTAwMDAtMDAwMC0wMDAwLTAwMDAwMDAwMDAwMCIsImtpZCI6IjAwMDAwMDAwLTAwMDAtMDAwMC0wMDAwLTAwMDAwMDAwMDAwMCJ9.G5GvGEBNMUga25EeaJeAbDk9yZuLBLyj5e0OzVfIjS70UOvDb-SvLSEhBv9vZ_rxjTtaWGQGK0iK8VnLce8KfqsxZzael6B5WqfwyQiok3WWIaQarrZZXKihWhgF49zYAIZx_0js1iSjoF9-vNSj8zan7j-miOCOssfPzGgfJqvWNnhR6_2YkCQgJssHMjGT1QxaJBZDVOuvY0ND7r6jxlS_Xze1nWtau1mtC6bu2hM8cxbYUtM-XOC8welCZ8ZOCKkutmVix0weV3TVNfR5vuBUz1QS6B9YC8R-eVVBhN2hl4j7kGZLmZ4TpyLViYEUVZsqGBayVIPeN2BhtqTO9g', 'code_verifier': 'IDiWUb62sjsNYuU/YtZ5YJdvvxE70gR9vEPOQo9+lh/DjMt1c6egVQRyXB0FAaASNB4/ME8YQjGQ1xUOS7ZwI4tjHDBbUXFBvt2DVu8jOlkDmZsNeI2IfQx5HRkz1nRIUlpqUC/m01gAQRYAuR6dbUyrkW8bq9B9cOLSbWzjJ0E'}
> [  leases  ]: B210CF72-FEC7-4440-9499-1156D1ACD13A (hYSKI4kpZcWqPatM5Sc9RSCuzMeyz2piTmrRQKnnHro): found 0 active leases
> [  create  ]: B210CF72-FEC7-4440-9499-1156D1ACD13A (hYSKI4kpZcWqPatM5Sc9RSCuzMeyz2piTmrRQKnnHro): create leases for scope_ref_list ['f27e8e79-a662-4e35-a728-7ea14341f0cb']

Error on releasing leases on shutdown (can be ignored and/or fixed with reverse proxy)

The driver wants to release current leases on shutting down windows. This endpoint needs to be a http endpoint. The error message can safely be ignored (since we have no license limitation :P) and looks like this:

Log example

<1>:NLS initialized
<1>:License acquired successfully. (Info: 192.168.178.110, NVIDIA RTX Virtual Workstation; Expiry: 2023-3-30 23:0:22 GMT)
<0>:Failed to return license to 192.168.178.110 (Error: Generic network communication failure)
<0>:End Logging

log with nginx as reverse proxy (see docker-compose-http-and-https.yml)

<1>:NLS initialized
<2>:NLS initialized
<1>:Valid GRID license not found. GPU features and performance will be fully degraded. To enable full functionality please configure licensing details.
<1>:License acquired successfully. (Info: 192.168.178.33, NVIDIA RTX Virtual Workstation; Expiry: 2023-1-4 16:48:20 GMT)
<2>:Valid GRID license not found. GPU features and performance will be fully degraded. To enable full functionality please configure licensing details.
<2>:License acquired successfully from local trusted store. (Info: 192.168.178.33, NVIDIA RTX Virtual Workstation; Expiry: 2023-1-4 16:48:20 GMT)
<2>:End Logging
<1>:End Logging
<0>:License returned successfully. (Info: 192.168.178.33)
<0>:End Logging

vGPU Software Compatibility Matrix

Successfully tested with this package versions.

vGPU Suftware	Driver Branch	Linux vGPU Manager	Linux Driver	Windows Driver	Release Date	EOL Date
`17.4`	R550	`550.127.06`	`550.127.05`	`553.24`	October 2024	February 2025
`17.3`	R550	`550.90.05`	`550.90.07`	`552.74`	July 2024
`17.2`	R550	`550.90.05`	`550.90.07`	`552.55`	June 2024
`17.1`	R550	`550.54.16`	`550.54.15`	`551.78`	March 2024
`17.0`	R550	`550.54.10`	`550.54.14`	`551.61`	February 2024
`16.8`	R535	`535.216.01`	`535.216.01`	`538.95`	October 2024	July 2026
`16.7`	R535	`535.183.04`	`535.183.06`	`538.78`	July 2024
`16.6`	R535	`535.183.04`	`535.183.01`	`538.67`	June 2024
`16.5`	R535	`535.161.05`	`535.161.08`	`538.46`	February 2024
`16.4`	R535	`535.161.05`	`535.161.07`	`538.33`	February 2024
`16.3`	R535	`535.154.02`	`535.154.05`	`538.15`	January 2024
`16.2`	R535	`535.129.03`	`535.129.03`	`537.70`	October 2023
`16.1`	R535	`535.104.06`	`535.104.05`	`537.13`	August 2023
`16.0`	R535	`535.54.06`	`535.54.03`	`536.22`	July 2023
`15.4`	R525	`525.147.01`	`525.147.05`	`529.19`	June 2023	December 2023
`14.4`	R510	`510.108.03`	`510.108.03`	`514.08`	December 2022	February 2023

*To get the latest drivers, visit Nvidia or search in Discord-Channel GPU Unlocking (Server-ID: 829786927829745685) on channel licensing

Credits

Thanks to vGPU community and all who uses this project and report bugs.

Special thanks to:

@samicrusader who created build file for ArchLinux
@cyrus who wrote the section for openSUSE
@midi who wrote the section for unRAID
@polloloco who wrote the NVIDIA vGPU Guide
@DualCoder who creates the vgpu_unlock functionality vgpu_unlock
Krutav Shah who wrote the vGPU_Unlock Wiki
Wim van 't Hoog for the Proxmox All-In-One Installer Script
@mrzenc who wrote fastapi-dls-nixos

And thanks to all people who contributed to all these libraries!

35 KiB Raw Blame History

FastAPI-DLS

Setup (Service)

Docker

Debian / Ubuntu / macOS (manual method using git clone and python virtual environment)

openSUSE Leap (manual method using git clone and python virtual environment)

Debian / Ubuntu (using dpkg / apt)

ArchLinux (using pacman)

unRAID

NixOS

Let's Encrypt Certificate (optional)

Configuration

Setup (Client)

Linux

Windows

unRAID Guest

API Endpoints

Troubleshoot / Debug

Docker

Linux

Windows

Known Issues

Linux

Invalid HTTP request

Windows

Required cipher on Windows Guests (e.g. managed by domain controller with GPO)

Multiple Display Container LS Instances

Error on releasing leases on shutdown (can be ignored and/or fixed with reverse proxy)

log with nginx as reverse proxy (see docker-compose-http-and-https.yml)

vGPU Software Compatibility Matrix

Credits

35 KiB

Raw Blame History

Debian / Ubuntu / macOS (manual method using `git clone` and python virtual environment)

openSUSE Leap (manual method using `git clone` and python virtual environment)

Debian / Ubuntu (using `dpkg` / `apt`)

ArchLinux (using `pacman`)