v0.1.1

A Kubernetes operator for Chroma, written in .NET. Community project, not affiliated with Chroma.

kubectl apply -f https://github.com/ChromaDotNet/ChromaDB.Operator/releases/latest/download/chroma-operator.yaml

Single-node Chroma. One resource Chroma gives a server with its PersistentVolumeClaim, its configuration and its Service.

• The operator keeps the highest version of Chroma that has opened the data, and refuses an older image, since Chroma does not open
  data migrated by a newer version.
• A claim is kept by default when the Chroma is deleted, with its data version.

Distributed Chroma, preview. With mode: distributed the operator creates the objects of the Helm chart distributed-chroma of
Chroma, on an external Postgres and an external S3 bucket.

• A test compares them with helm template of the chart.
• The operator checks Postgres and the bucket before anything starts.
• The migration Job runs first, then the components tier by tier.
• Going back to an earlier version is refused without an annotation.
• The images of Chroma for the distributed mode need CPUs with AVX-512.

VecturaDB, preview. A single-node Chroma can run VecturaDB, which adds sparse vectors,
BM25, the Search API and collection fork to the single node of Chroma, with the same API and on the same data.

• The data passes from Chroma to VecturaDB only along the passages the operator knows, each one tried on a volume first.
• Back to Chroma needs an annotation, and the status says what Chroma does not see of the data.

Known limit. On AKS a change of a Chroma once reached the operator and was applied only after a restart of the operator. The cause
is not known. Two safety nets cover it:

• every Chroma goes back into the queue of the operator once a minute;
• when a Chroma has had no pass for 5 minutes, the liveness probe of the operator fails and Kubernetes restarts the operator.

See the README for the fields, the status and the limits.