k8s-deploy: add DocumentDB backend (MongoDB-compatible)

[hossain-rayhan]

Description

Adds an optional DocumentDB backend to the k8s-deploy/ flow as a single MongoDB-compatible alternative to running PostgreSQL and Neo4j side-by-side. DocumentDB (documentdb/documentdb-kubernetes-operator) ships its own operator, so this PR wires it in alongside the existing KubeBlocks-based databases rather than as a KubeBlocks addon.

🙏 Looking for an early sanity-check on the approach before I polish further. The DocumentDB project provides its own Kubernetes operator (Helm chart at https://documentdb.github.io/documentdb-kubernetes-operator), so I deliberately did not package it as a KubeBlocks addon, instead 01-prepare.sh installs the upstream operator directly when ENABLE_DOCUMENTDB=true. Happy to refactor if maintainers would prefer a different integration shape.

DocumentDB itself is not GA yet (upstream notes: "under active development but not yet recommended for production use"), so the README places it under a clearly-labeled Preview / Experimental section, separate from the Production Deployment flow.

Related Issues

N/A — inspired by documentdb/documentdb-kubernetes-operator#362.

Changes Made

New folder (mirrors siblings like databases/postgresql/, databases/neo4j/):

• k8s-deploy/databases/documentdb/values.yaml — single-node DocumentDB CR + docdb-credentials Secret (placeholder password), images pinned to ghcr.io/documentdb/documentdb-kubernetes-operator/{documentdb,gateway}:0.110.0.

Wiring into the existing scripts (all gated on ENABLE_DOCUMENTDB=false by default):

• databases/00-config.sh — added ENABLE_DOCUMENTDB flag with a note that it uses its own operator (not KubeBlocks).
• databases/01-prepare.sh — installs cert-manager (only if missing) and the DocumentDB operator Helm chart into the documentdb-operator namespace.
• databases/02-install-database.sh — applies the CR + Secret and waits for status.status="Cluster in healthy state".
• databases/03-uninstall-database.sh — deletes the CR.
• databases/04-cleanup.sh — uninstalls the DocumentDB operator (cert-manager intentionally left in place).
• lightrag/values.yaml — adds MONGO_DATABASE: lightrag plus comments documenting the override env vars.
• install_lightrag.sh — sources 00-config.sh, gates PG/Neo4j password lookups on their ENABLE_* flags, and adds a DocumentDB block that:
  1. Reads status.connectionString from the DocumentDB resource,
  2. eval-resolves the embedded $(kubectl get secret …) substitution,
  3. Swaps the ClusterIP for the in-cluster DNS name and strips replicaSet=rs0 (which conflicts with directConnection=true in pymongo),
  4. Sets LIGHTRAG_KV_STORAGE=MongoKVStorage, LIGHTRAG_GRAPH_STORAGE=MongoGraphStorage, LIGHTRAG_DOC_STATUS_STORAGE=MongoDocStatusStorage, and LIGHTRAG_VECTOR_STORAGE=NanoVectorDBStorage.

Docs:

• k8s-deploy/README.md — new top-level Preview / Experimental: DocumentDB Backend section after Production Deployment, with a ⚠️ "not for production use" warning, enable instructions, and caveats.
• k8s-deploy/databases/README.md — short note in step 3 about the auto-install.

Checklist

• Changes tested locally (full install/uninstall cycle on a Kubernetes 1.35 cluster — DocumentDB cluster reaches healthy state, LightRAG connects, KV/graph/doc-status collections appear in DocumentDB)
• Code reviewed
• Documentation updated (k8s-deploy/README.md, k8s-deploy/databases/README.md)
• Unit tests added (n/a — pure deployment/scripting changes; bash -n passes on all modified scripts)

Additional Notes

• Vectors are not stored in DocumentDB. DocumentDB does not implement the MongoDB Atlas $vectorSearch operator that MongoVectorDBStorage requires, so embeddings stay on the LightRAG PVC via NanoVectorDBStorage. KV / graph / doc-status data do live in DocumentDB collections.
• Kubernetes 1.35+ is required because the DocumentDB operator uses the ImageVolume feature.
• Two non-fatal startup warnings from DocumentDB are expected and safe to ignore: createIndex.collation is not implemented yet and Pipeline stage name not recognized: $listSearchIndexes.
• All existing PG/Neo4j flows are unchanged when ENABLE_DOCUMENTDB=false (the default).

[danielaskdd]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 0a5295bab0

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Install DocumentDB operator from its OCI chart

When ENABLE_DOCUMENTDB=true, this path tries to install the operator from a conventional Helm repo, but the current upstream install docs publish the chart as an OCI artifact (oci://ghcr.io/documentdb/documentdb-operator) and note that OCI registries do not support repo-style discovery (see https://github.com/documentdb/documentdb-kubernetes-operator#install). In a fresh cluster following the new README, helm repo update documentdb / helm upgrade ... documentdb/documentdb-operator can fail to install the CRDs, leaving 02-install-database.sh unable to apply the DocumentDB resource.

Useful? React with 👍 / 👎.