Bulk load points into Qdrant efficiently with batch upsert and indexing disabled during ingestion

Documented steps

1. Upsert with PUT /collections/{collection_name}/points. Any point whose id already exists is overwritten, so insert and update are the same call.
2. Column-oriented batch format (compact for bulk): {"batch":{"ids":[1,2,3],"vectors":[[0.9,0.1,0.1],[0.1,0.9,0.1],[0.1,0.1,0.9]],"payloads":[{"color":"red"},{"color":"green"},{"color":"blue"}]}}.
3. Record-oriented alternative: {"points":[{"id":1,"vector":[0.9,0.1,0.1],"payload":{"color":"red"}}]}.
4. Point ids must be a 64-bit unsigned integer or a UUID string (hyphenated, simple, or URN form). Nothing else is accepted.
5. Query params on the PUT: ?wait=true to block until the write is applied, ordering=weak|medium|strong, timeout=<seconds>.
6. Optionally set body field update_mode (v1.17.0+): 'upsert' (default), 'insert_only', or 'update_only'.
7. Before a large load, disable indexing: PATCH the collection with {"optimizers_config":{"indexing_threshold":0}} via PUT /collections/{collection_name} update, or set indexing_threshold_kb: 0 in the YAML config for a fresh node.
8. Run the batched PUT calls without wait=true for throughput, or use a client SDK's upload_collection / upload_points with a parallel worker count.
9. After the load, restore indexing: set optimizers_config.indexing_threshold back to 10000 (the default) so Qdrant builds indexes on the now-static data.
10. Verify with GET /collections/{collection_name} that points_count matches and status returns to 'green'.
11. Reference: https://qdrant.tech/documentation/manage-data/points/
12. Reference: https://qdrant.tech/documentation/ops-optimization/optimizer/