Kubernetes deployment
Language Weaver Edge 8.5 can be deployed on a Kubernetes cluster, using provided Helm Charts.
- Simplified configuration of Language Weaver Edge using
yamlfiles. - Automatic deployment of Language Weaver Edge controller, job engines, and translation engines as pods.
- Job and translation engines are available as StatefulSets and can be scaled manually or automatically, based on needs.
- With auto-scaling you can define the minimum and maximum number of pods that can be started for a specific language pair, and what CPU usage threshold must be met to start a new pod. Pods are then automatically started based on current workload. This feature can be combined with Kubernetes nodes auto-scaling and ensure nodes are only started when more pods need to be deployed.
The following are the high level steps taken during the four parts of the installation process to achieve this deployment setup.
Part 1: Deploying prerequisites
- A block storage of "ReadWriteOnce"
- A file storage of "ReadWriteMany"
- An ingress controller
- A secret containing Admin username/password
- A secret containing Language Weaver Edge license (optional)
- A GPU node to run the training engine pod (optional)
Part 2: Deploying storage classes, persistent volumes, and copying language pairs
sdl-mtedge-data helm chart creates storage classes, persistent volumes, copies all language pairs listed in the Values.yaml file to the mtedge-lp-pvc volume.
Part 3: Deploying Language Weaver Edge
sdl-mtedge helm chart deploys controller pod and all the engine pods in the specified namespace as per Values.yaml file.
Part 4: Obtaining the Language Weaver Edge license
Only the Language Weaver Edge controller pod requires a license. All engine pods get their license from the controller pod.
- The desired Language Weaver Edge language pairs.
- Processing Units count.
After the license file is provided by the licensing team, it is uploaded through the web Ui.
See Deploying Language Weaver Edge on Kubernetes for detailed information on the deployment process.