Installation
Follow these simple steps to install a new instance of Supervisely to your server.
Before you start installation of Supervisely to your host, please make sure your server meets the hardware & software requirements below.
The minimum hardware requirements are as following:
CPU: 4 vCPUs / cores
Memory: 8 GB RAM
Hard drive: 100 GB storage or more
Choose the hard drive size according to your needs. There is also an option to attach any S3 compatible storage to distribute the data in your cloud.
If you don't have enough free space on a system drive, but you have mounted another one, you can change the place where the data is stored by changing DATA_PATH in .env file
Never set DATA_PATH pointing to a network share (NFS or SMB), because it affects performance significantly. Set DATA_PATH pointing to an actual directory on your local hard drive and then mount storage subfolder of DATA_PATH directory as a network share - this way "hot" subfolders like db or proxy_cache will be on a local hard drive and image / videos storage on a network share.
If you plan to use smart tool (AI powered semantic segmentation) or neural networks (human-in-the-loop) you will need a server with GPU. The requirements are as following:
GPU Memory: 8 GB or more (GeForce GTX 1080, Tesla K80, Tesla P100)
You can install Supervisely and run GPU computations on different machines using agents (see cluster).
We recommend to use the following EC2-instances for deployment in AWS:
1 × m5.2xlarge — the platform itself
1 × p2.xlarge — (for Neural Network Module only)
To run Supervisely you will need Linux OS with kernel 3.10 or newer. The following configuration is recommended:
OS: Ubuntu 18.04 or later
NVIDIA graphics driver: 460.27.03 or later (for Neural Network module only)
nvidia-docker2 (for Neural Network module only)
All pre-requirements can be automatically installed via supervisely-cli utility for Ubuntu 18.04 or later. In case of a different linux distribution, you can install Supervisely and pre-requirements manually, but we won't be able to provide installation support.
When all requirements above have been installed, you can deploy Supervisely. Please follow those steps one by one. Read each step carefully and complete it fully before going to the next step.
Before the installation we will send you a license.key, a unique installation key and command to install supervise-cli. Please run it on the machine where you want to install Supervisely and make sure sudo supervisely gives you help information.
Run sudo supervisely install-all in your terminal. We will detect the necessary dependencies and install them. The following software may be installed:
CUDA 9.0 & Nvidia Driver (for Training Module)
NVIDIA Container Runtime (for Training Module)
If don't have NVIDIA drivers and CUDA, you will be asked if you need to install it. Choose y if you have GPU on your server. Your computer will be rebooted during the installation of the NVIDIA driver.
We have sent you installation key that looks like "hdhUssJskOskAA". Run sudo supervisely auth <installation key> to authenticate yourself. Now, run sudo supervisely update to download the latest version of Supervisely. We will ask you to provide a directory to store configuration in and some required variables. Be default, we use /opt/supervisely folder. You can always check where is your configuration directory using supervisely where command.
Though we will ask you to provide required configuration variables on the previous step, it's a good idea to make a final overview.
Open file /opt/supervisely/.env. There are some configuration variables in it that you might want to change. Look for <please, insert the value here> - you need to provide values here explicitly because we don't know those values in advance.
Here are some variables that you can find here:
SERVER_ADDRESS- Public or local network address of the server where you deploy Supervisely. That value will be used by agents to connect to the cluster. Example: 192.168.1.42 or supervisely.intranet. If you have a very restrictive network environment with firewall, you can useip addr list docker0to get your docker0 IP address - it is some kind oflocalhostfor docker.STORAGE_ACCESS_KEY- Access key for S3 integration (optional)STORAGE_SECRET_KEY- Secret key for S3 integration (optional)
If you will decide to change those values later after the installation is complete, do not forget to run supervisely up -d to apply the changes.
Run sudo supervisely login to login to our private docker registry. We provide the necessary credentials in .env file so you don't have to input anything.
If you see "syntax error", you forgot to set one of the variables in .env. Please go back to the previous step to edit the required variables.
You should see "Login Succeeded" in your terminal.
Now it's time to start the Supervisely instance. Run sudo supervisely upgrade --skip-backup. We will pull docker images from registry and run every service defined in docker-compose.yml in detached mode.
This can take a while.
When the script has finished, you can open Supervisely in your web browser. Go to http://localhost page. You will see the following:
Enter your license key from the file license.key here and press "Update license".
You should see a "Success" message - your license has been updated.
Now you can refresh the current page. You will see the login box:
Enter the default credentials: login and password admin. Now your are in. Welcome!
We strongly advice you to complete post-installation steps after the initial setup.
I open "http://localhost" but see an error
Database is still initializing. Wait a minute and try again.
One of our internal services haven't got the license key. Refresh the page and try again.
It seems the agent has tried to wait for Supervisely to start. Restart agent manually: sudo supervisely restart agent.
Nvidia container runtime has not been installed. Try to run command docker run --runtime=nvidia --rm nvidia/cuda nvidia-smi to see if it works.
If you don't want to run train & inference tasks, please remove checkbox "Use nvidia runtime" under advances settings in the new node modal window and run the deploy command again.
The models are still downloading. It can take a while, just wait some time and refresh the Models list.
