Qdrant

The qdrant output upserts each batch's rows as Qdrant points over the REST API. Each row becomes one point: the vector comes from a FixedSizeList(Float32)/List(Float32) column (typically produced by the embedding processor), an optional id column keys the point, and every remaining column lands in the point payload.

Configuration

Field	Type	Required	Default	Description
type	string	yes	—	qdrant
url	string	yes	—	Qdrant base URL (e.g. http://localhost:6333).
collection	string	yes	—	Target collection name.
vector_field	string	no	embedding	Column holding the vector.
id_field	string	no	—	Column used as the point id (unsigned integer or string). When omitted, the console generates a random UUID v4 per row — redelivery inserts duplicates, so prefer an id column for at-least-once idempotency.
payload_fields	array	no	all other columns	Columns included in the point payload.
api_key	string	no	—	Sent as Authorization: Bearer. Supports secret references.
timeout_ms	integer	no	30000	HTTP request timeout.
retry_count	integer	no	3	Retry attempts for connection errors and 5xx responses; 4xx responses fail immediately.
headers	map	no	—	Extra HTTP headers.

Semantics

• Writes use PUT /collections/{collection}/points?wait=true — one bounded request per slice of at most 1,000 points; larger batches are split sequentially.
• Upserts are idempotent per point id: with an id_field, redelivery overwrites the same point (at-least-once friendly); without one, a fresh UUID is generated per row, so every attempt inserts a new point.
• The vector's dimension must match the collection's; mismatches are returned by Qdrant as a 4xx and fail the batch (routed to error_output when configured).
• Requests to loopback endpoints (a local Qdrant instance) always bypass the system proxy.

Example

yaml

output:
  type: "qdrant"
  url: "http://localhost:6333"
  collection: "documents"
  vector_field: "embedding"
  id_field: "doc_id"

Complete Pipeline Example

yaml

logging:
  level: info

streams:
  - id: docs-upserts
    input:
      type: memory
    pipeline:
      processors:
        - type: json_to_arrow
    output:
      type: qdrant
      url: "http://localhost:6333"
      collection: "documents"
      vector_field: "embedding"
      id_field: "doc_id"