Managed Qdrant Vector Database

Okustera Managed Qdrant delivers ultra-fast, high-dimensional vector similarity search and document indexing for Retrieval-Augmented Generation (RAG), semantic search, and AI recommendation engines.

Deployed as container-native Kubernetes StatefulSets backed by Ceph RBD NVMe storage, Okustera Qdrant instances combine Rust performance with zero-trust tenant network isolation and automated API key escrow via OpenStack Barbican KMS.

---

Architectural Highlights

Key Capabilities

• On-Disk NVMe Vectors & Payloads: Payloads and vector indices are persisted to NVMe block volumes (QDRANT__STORAGE__ON_DISK_PAYLOAD: "true"), keeping memory footprint under 400 MB per instance while maintaining sub-15ms search latencies.
• Scalar Quantization: Compresses 32-bit floating-point embeddings (FP32) into 8-bit integers (INT8), reducing RAM consumption by up to 90% with less than 1% precision loss.
• Native Hybrid Search (Dense + Sparse BM25): Supports dense neural embeddings combined with sparse BM25 keyword vectors in a single retrieval query for superior ranking accuracy.
• Payload Metadata Pre-Filtering: Filters vectors directly during HNSW graph traversal based on metadata tags (tenant_id, document_type, timestamps) rather than filtering after retrieval.
• Zero-Trust Private Access: Vector clusters are provisioned strictly on private internal ClusterIP networks (:6333 HTTP, :6334 gRPC). Public subdomains (qdrant.okustera.com) are eliminated, guaranteeing vectors and proprietary embeddings never traverse public ingress routes.
• Barbican KMS Key Escrow: 32-character high-entropy API keys are automatically generated, stored securely in OpenStack Barbican KMS, and injected into the container environment.

---

Provisioning via OMC Portal & REST API

Portal Cloud Console

1. Navigate to Database Services (DBaaS) in the OMC Portal.
2. Click Create Database Cluster and select Qdrant Vector DB.
3. Specify your cluster name, tenant namespace, CPU/memory limits, and Ceph NVMe storage allocation (default: 10 GiB).
4. Click Create Cluster. The cluster endpoint (http://api_key:****@qdrant-<name>.<namespace>.svc.cluster.local:6333) and Barbican API key will be displayed with one-click copy.

REST API Example

curl -X POST "https://portal.opencloud.local/api/v1/dbaas/qdrant" \
  -H "Authorization: Bearer ${OKUSTERA_API_TOKEN}" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "enterprise-rag-kb",
    "namespace": "default",
    "storage_size": "20Gi",
    "cpu_limit": "2",
    "memory_limit": "4Gi"
  }'

---

Connecting from Python (qdrant-client)

To query your private Qdrant cluster from an application pod running in the same Kubernetes cluster or over WireGuard VPN:

import os
from qdrant_client import QdrantClient
from qdrant_client.models import Distance, VectorParams, PointStruct

# Connect via private internal ClusterIP service
client = QdrantClient(
    url="http://qdrant-enterprise-rag-kb.default.svc.cluster.local:6333",
    api_key=os.environ.get("QDRANT_API_KEY"),
)

collection_name = "knowledge_base"

# Create a collection with cosine distance metric
if not client.collection_exists(collection_name):
    client.create_collection(
        collection_name=collection_name,
        vectors_config=VectorParams(size=1536, distance=Distance.COSINE),
    )

# Upsert sample embedding points
client.upsert(
    collection_name=collection_name,
    points=[
        PointStruct(
            id=1,
            vector=[0.05] * 1536,
            payload={"source": "sovereign_cloud_spec.pdf", "tenant_id": "acme-corp"},
        )
    ],
)

# Perform similarity search with metadata pre-filtering
results = client.search(
    collection_name=collection_name,
    query_vector=[0.05] * 1536,
    limit=5,
)

for hit in results:
    print(f"Point ID: {hit.id}, Score: {hit.score}, Payload: {hit.payload}")

---

Connecting via cURL / REST API

Within the tenant private network:

# Verify cluster health
curl -s -H "api-key: ${QDRANT_API_KEY}" \
  http://qdrant-enterprise-rag-kb.default.svc.cluster.local:6333/readyz

# List all collections
curl -s -H "api-key: ${QDRANT_API_KEY}" \
  http://qdrant-enterprise-rag-kb.default.svc.cluster.local:6333/collections

---

Technical Specifications

Parameter	Specification
Engine	Qdrant Vector Database (qdrant/qdrant:latest)
Language	Rust
Persistence	Ceph RBD Block Volume (ceph-rbd StorageClass)
Network Exposure	Private ClusterIP (TCP 6333 HTTP, TCP 6334 gRPC)
Public Routing	None (Zero-Trust isolated)
Authentication	API Key stored in OpenStack Barbican KMS
Distance Metrics	Cosine, Dot Product, Euclidean, Manhattan
Index Types	HNSW (Hierarchical Navigable Small World) with on-disk payload