To get started using kubernetes, we defer to the kubernetes documentation.
Kubernetes can be useful in your ML workflows if you need access to other infrastructure resources. For example:
- Deploying and serving models with a service mesh
- Deploying and serving arbitrary applications, such as Jupyter Notebooks or documentation sites.
- Accessing resources that are visible from your internal k8s cluster, such as:
- Databases, i.e. Presto, Hive
- Storage, i.e. HDFS
- Distributed data processing, such as Dask or Spark
- Specialized computing such as GPUs.
- Integration with experiment tracking systems.
- Access to secrets.
- Native resource management and kubernetes for scalability and resiliance.
- Interoperability with machine learning or data pipelines, such as Argo, MLFlow, Prefect, etc.
- … and much more.