qdrant

Qdrant → NornicDB Migration Scripts

Three runnable migrations to move collections + points from a source Qdrant instance into NornicDB. NornicDB exposes a Qdrant-compatible gRPC surface, so most of these scripts use the same Qdrant client library on both ends — only the connection target differs.

The Python and Go scripts implement a same-API replay: list collections, replicate vector configs (single and named), scroll points, upsert with wait=true, verify counts. Re-running is idempotent — Upsert is keyed by point ID, and existing collections can be skipped with --skip-existing.

The Node script implements a bridge: it reads from Qdrant over REST and writes into NornicDB over Bolt as nodes (one label per source collection, vector on n.embedding, payload merged into the node properties). Named-vector collections are not handled by the Node script — use Python or Go for that case.

Choosing a script

• Python — quickest to set up, talks gRPC end-to-end. Use this unless you need otherwise.
• Go — no Python runtime, vendorable for CI pipelines. Talks gRPC end-to-end.
• Node — pick this only if you already have a Node/TS pipeline. Different shape: writes into NornicDB through Bolt rather than gRPC, so the result is graph nodes (with vector + payload) rather than Qdrant-collection-shaped data.

Common flags

Example

# Python, gRPC end-to-end
python scripts/migration/qdrant/migrate.py \
    --source-url http://qdrant.prod:6333 \
    --target-host nornicdb.local --target-grpc-port 6334 \
    --batch-size 512 --skip-existing

What gets migrated

• Vector configs (single and named vectors), distance metric, dimension.
• All points: ID, vector(s), payload.
• Idempotently: re-running upserts the same IDs, no duplicates.

What does NOT get migrated

• Snapshots, replication settings, sharding parameters — NornicDB manages these internally; the source values are advisory.
• HNSW tuning parameters (M, ef_construct, ef_search) — NornicDB auto-tunes.
• Quantization configs — NornicDB does not currently use Qdrant-shaped quantization.
• Per-collection RBAC — re-author against NornicDB's Auth.Enabled model if needed.

Verification after migration

# Spot-check a few points
python -c "
from qdrant_client import QdrantClient
c = QdrantClient(host='nornicdb.local', grpc_port=6334, prefer_grpc=True)
print(c.get_collection('docs'))
print(c.count('docs', exact=True))
points, _ = c.scroll('docs', limit=5, with_vectors=True, with_payload=True)
for p in points:
    print(p.id, len(p.vector or []), list((p.payload or {}).keys()))
"

See also

• docs/skills/grpc.skill.md — the gRPC surface the scripts call.
• docs/skills/qdrant-migration.skill.md — the consumer-facing migration skill.
• docs/user-guides/qdrant-grpc.md — full proto compatibility matrix.