File
Read a CSV, JSON or Parquet file by URL. The relation is memoized: it is fetched once.
Syntax
{ "id": "n1", "op": "file", "name": "file", "x": 0, "y": 0, "params": {} }
One out of it starts at "sourcePort": "out".
Inputs
This operator has no input ports.
Parameters
| Parameter | Kind | Default | Choices |
|---|---|---|---|
url | url | "" | |
format | menu | "csv" | csv, json, parquet |
header | toggle | true | |
names | string | "" | |
nullstr | string | "" |
Outputs
| Port | Type |
|---|---|
out | table |
Remarks
File reads one CSV, JSON or Parquet file into a table in DuckDB, in the engine worker. The table is read once and kept: a change downstream never reads the file again, and changing the URL, the format or a CSV option reads it anew.
A URL that starts with @/ names a file in the project's data directory, so @/data.csv is the data.csv beside noodles.json. Any other URL is fetched by the browser, so a server on another domain has to allow cross-origin requests (CORS).
DuckDB detects each column's type from the file. The Header row, Column names and Null marker options apply to CSV only: turn Header row off and list Column names for a file whose first row is data. A JSON file is an array of objects, one row each.
To compute a value per row, add an Attribute Create after it, or a DuckDB SQL node for anything a query does better.
Examples
A CSV stored with the project, its positions made into a column for a layer:
{
"nodes": [
{ "id": "n1", "op": "file", "name": "stations", "x": 0, "y": 0, "params": { "url": "@/stations.csv" } },
{ "id": "n2", "op": "attribute", "name": "position", "x": 300, "y": 0, "params": { "name": "position", "expression": "[lng, lat]" } }
],
"edges": [{ "id": "e1", "source": "n1", "sourcePort": "out", "target": "n2", "targetPort": "in" }]
}
Details
- Category: Data
- Operator:
file
Older projects
Projects saved before documents used operator classes. These convert to file when they open: FileOp.