Maintained by: Gurobi Optimization
Where to get help: Gurobi Support, Gurobi Documentation
When building a production application, we recommend using an explicit version number instead of the latest tag.
This way, you are in control of the upgrade process of your application.
Supported architectures: linux/amd64, linux/arm64
Published image artifact details: https://github.com/Gurobi/docker-manager
Gurobi images:
- gurobi/optimizer: Gurobi Optimizer (full distribution)
- gurobi/python: Gurobi Optimizer (Python API only)
- gurobi/python-example: Gurobi Optimizer example in Python with a WLS license
- gurobi/modeling-examples: Optimization modeling examples (distributed as Jupyter Notebooks)
- gurobi/compute: Gurobi Compute Server
- gurobi/manager: Gurobi Cluster Manager
The Gurobi Optimizer is the fastest and most powerful mathematical programming solver available for your LP, QP and MIP (MILP, MIQP, and MIQCP) problems. More info at the Gurobi Website.
The Gurobi Cluster Manager is a new server component that helps to manage your Compute Server. It provides better security through user authentication and API keys. It also expands the capabilities of Compute Server cluster nodes with unified management of interactive and non-interactive optimization tasks. Gurobi Compute Server Documentation
The Web License Service (WLS) is a new Gurobi licensing service for containerized environments (Docker, Kubernetes...). Gurobi components can automatically request and renew license tokens to the WLS servers available in several regions worldwide. WLS only requires that your container has access to the Internet. Commercial users can request an evaluation and academic users can request a free license. Please register to access the Web License Manager and read the documentation
Note that other standard license types (NODE, Academic) do not work with containers. Please contact your sales representative at sales@gurobi.com to discuss licensing options.
The Cluster Manager itself does not need a license, but the Compute Server nodes will, and there are two options:
-
Mounting the client license file: You can store connection parameters in a client license file (typically called
gurobi.lic) and mount it to the container. This option provides a simple approach for testing with Docker. When using Kubernetes, the license file can be stored as a secret and mounted in the container. -
Setting parameters through environment variables for a WLS license: GRB_WLSACCESSID, GRB_WLSSECRET, and GRB_LICENSEID. These variables are used to pass the access ID, the secret, and the license ID respectively.
We do not recommend adding the license file to the Docker image itself. It is not a flexible solution as you may not reuse the same image with different settings. More importantly, it is not secure as some license files need to contain credentials in the form of API keys that should remain private.
The following command starts a Cluster Manager instance and connects to a MongoDB database
$ docker run gurobi/manager --database=MONGO_DB_URL
Then you can go to http://localhost:61080/manager and log in using the default credentials:
standard user: gurobi / pass
administrator: admin / admin
system administrator: sysadmin / cluster
However, this is for testing only because without at least one Compute Server node, you will not be able to submit optimization jobs. This is why using Docker Compose or Kubernetes is necessary as the deployment scripts will start all the required components (see below).
Example docker-compose.yml for a cluster manager:
version: '3.1'
services:
mongodb:
image: mongo:latest
restart: always
volumes:
- mongodb:/data/db
networks:
- back-tier
manager:
image: gurobi/manager:latest
restart: always
depends_on:
- "mongodb"
ports:
- "61080:61080"
command: --database=mongodb://mongodb:27017
networks:
- back-tier
compute:
image: gurobi/compute:latest
restart: always
depends_on:
- manager
command: --manager=http://manager:61080
networks:
- back-tier
volumes:
- ./gurobi.lic:/opt/gurobi/gurobi.lic:ro
volumes:
mongodb:
networks:
back-tier:
Run docker-compose up from the directory where the docker-compose.yml file was saved to download and start all the containers.
Then you can go to http://localhost:61080/manager and log in using the default credentials:
standard user: gurobi / pass
administrator: admin / admin
system administrator: sysadmin / cluster
Note that you can scale gurobi/compute instances using the scale flag:
$ docker-compose up --scale compute=2
If you want to mount a specific license with Kubernetes, it can be done using a secret.
kubectl create secret generic gurobi-lic --from-file="gurobi.lic=$PWD/gurobi.lic"
Then you can start multiple pods for the Cluster Manager, the Mongo Database, and the Compute Server nodes. A simple deployment file is provided as an example. To keep the demonstration simple, this deployment file will not persist the database. If you wish to do so, please refer to the MongoDB documentation or a hosted solution (for example MongoDB Atlas).
kubectl apply -f k8s.yaml
To check the deployment, you can list the created pods:
% kubectl get pods
NAME READY STATUS RESTARTS AGE
gurobi-compute-674f8dcbf4-v27nn 1/1 Running 0 8m18s
gurobi-compute-674f8dcbf4-x9kff 1/1 Running 0 8m23s
gurobi-manager-79dcbf5b74-c9dfv 1/1 Running 0 8m23s
gurobi-manager-79dcbf5b74-wcqhj 1/1 Running 0 8m17s
Then you can access the logs of one instance of the Cluster Manager:
% kubectl logs gurobi-manager-7f57dcc5bd-2vfn5
2024-03-05T16:12:51Z - info : Gurobi Cluster Manager starting...
2024-03-05T16:12:51Z - info : Platform is linux64 (linux) - "Ubuntu 20.04.6 LTS"
2024-03-05T16:12:51Z - info : Version is 12.0.0 (build v12.0.0rc1)
2024-03-05T16:12:51Z - info : Connecting to database grb_rsm on 10.97.58.19:27017...
2024-03-05T16:12:56Z - info : Connected to database grb_rsm (version 7.0.6, host gurobi-mongo-9fbb76c7b-6hhb2)
2024-03-05T16:12:57Z - info : Default gurobi user created
2024-03-05T16:12:57Z - info : Default admin user created
2024-03-05T16:12:57Z - info : Default sysadmin user created
2024-03-05T16:12:57Z - info : Default settings created
2024-03-05T16:12:57Z - info : Schema already migrated by another process
2024-03-05T16:12:57Z - info : Checking 0 cluster nodes
2024-03-05T16:12:57Z - info : Authentication Provider set to Gurobi Authentication Provider
2024-03-05T16:12:57Z - info : Creating proxy with MaxIdleConns=200 MaxIdleConnsPerHost=32 IdleConnTimeout=130
2024-03-05T16:12:57Z - info : Public root is /opt/gurobi_server/linux64/resources/grb_rsm/public
2024-03-05T16:12:57Z - info : Starting Cluster Manager server (HTTP) on port 61080...
2024-03-05T16:13:31Z - info : Node d1fc142a-1c41-4abf-b832-0153186e658e:10.1.0.59:61000, Compute Server registered
2024-03-05T16:13:31Z - info : Node 4506308c-ab79-4e5f-865d-5eea22e93f73:10.1.0.58:61000, Compute Server registered
As well as the logs of one of the Compute Server nodes:
% kubectl logs gurobi-compute-547b4fb898-xhxc6
2024-03-05T16:12:51Z - info : Gurobi Remote Services starting...
2024-03-05T16:12:51Z - info : Platform is linux64 (linux) - "Ubuntu 20.04.6 LTS"
2024-03-05T16:12:51Z - info : Version is 12.0.0 (build v12.0.0rc1)
2024-03-05T16:12:51Z - info : Variable GRB_LICENSE_FILE is not set
2024-03-05T16:12:51Z - info : Using license file /opt/gurobi/gurobi.lic
2024-03-05T16:12:51Z - info : Server starting WLS license
2024-03-05T16:12:51Z - info : Node address is 10.1.0.59:61000
2024-03-05T16:12:51Z - info : WLS container ID mhost:15V7SRCZ Tag 1709655171590318981120
2024-03-05T16:12:51Z - info : Node FQN is gurobi-compute-547b4fb898-xhxc6
2024-03-05T16:12:51Z - info : Node has 12 cores
2024-03-05T16:12:51Z - info : Using data directory /opt/gurobi_server/linux64/bin/data
2024-03-05T16:12:51Z - info : Data store created
2024-03-05T16:12:51Z - info : Node ID is d1fc142a-1c41-4abf-b832-0153186e658e
2024-03-05T16:12:51Z - info : Available runtimes: [10.0.0 10.0.1 10.0.2 10.0.3 11.0.0 11.0.1 11.0.2 11.0.3 12.0.0]
2024-03-05T16:12:51Z - info : Public root is /opt/gurobi_server/linux64/resources/grb_rs/public
2024-03-05T16:12:51Z - info : Starting API server (HTTP) on port 61000...
2024-03-05T16:12:51Z - info : Accepting worker registration on port 39029...
2024-03-05T16:13:31Z - info : Joining cluster from manager
2024-03-05T16:13:31Z - info : Node will join 10.1.0.58:61000
2024-03-05T16:13:32Z - info : Node 10.1.0.58:61000, transition from JOINING to ALIVE
Finally, you can retrieve the load balancer ingress of the gurobi-manager service
and open your browser to the exposed url and port. If you have applied the deployment file
to a local environment, you can go to http://localhost:61080/manager and log in using the
default credentials:
standard user: gurobi / pass
administrator: admin / admin
system administrator: sysadmin / cluster
Gurobi Compute Server will run the optimization jobs which are CPU-intensive tasks. To get better performance, we recommend deploying the Compute Server pods on dedicated Kubernetes worker nodes with high-end CPUs, adequate memory, and avoid resource contention:
- By using a daemon set, we run one and only one Gurobi Compute Server per worker node as an agent. Also, if you scale up or down the number of nodes in your node group, it will automatically adjust the number of Compute Server nodes.
- Then, the node selector has a label
app: computeto indicate that the daemon will only run on the dedicated worker nodes. - Finally, by defining a taint
app:computeon the selected worker nodes, and applying the tolerations to the daemon set, we make sure no other pods will run on these nodes, unless they have the required toleration.
A simple deployment file is provided as an example.
By downloading and using this image, you agree with the End-User License Agreement for the Gurobi software contained in this image.
As with all Docker images, these likely also contain other software which may be under other licenses (such as Bash, etc from the base distribution, along with any direct or indirect dependencies of the primary software being contained).
As for any pre-built image usage, it is the image user's responsibility to ensure that any use of this image complies with any relevant licenses for all software contained within.