A search feature may need to find documents or images with similar meaning rather than matching the same words. Milvus stores vectors, which are numerical representations produced by models, and searches for nearby representations. Applications can use the returned records as relevant examples or context. It is the storage and search component, not the model that creates those vectors; a Kubernetes installation also brings metadata, messaging and persistence dependencies to operate.
Chart ownership
Milvus identifies zilliztech/milvus-helm as its official chart source. The project installation guide uses the zilliztech/milvus chart. Review the documentation for the selected application and chart versions together; the repository records compatibility restrictions and changes to messaging dependencies across chart generations.
Before adoption
Inventory the metadata store, object storage and message queue alongside Milvus itself. Decide whether those services are supplied by chart dependencies or operated externally. Capacity tests should include ingestion, index construction, filtered search and recovery after a component fails. Protect database access and credentials, and test a backup that includes the required metadata and stored objects. Choose standalone or distributed mode from measured requirements, and read the version-specific upgrade path before changing images or values.
Sources & further reading
Spotted something that needs another look?
Help improve this page →