File Imports

import { Aside } from ‘@astrojs/starlight/components’;

A type: "import" action loads a Parquet / CSV / JSON file into a table in the warehouse — the symmetric inverse of type: "export". Unlike an export (a sink), an import is a producer: the table it creates is ref()-able by downstream models, just like a table or a declaration that actually carries data.

It’s a config-only action — there’s no SQL body (like a declaration). The source is the file at location; the destination is the action’s own target.

Need to produce the file first — download it, unzip it, reshape it? Pair the import with a python script action (1.20+) it depends on: the script stages the file in-DAG, the import loads it.

-- definitions/orders_in.sqlx
config {
  type: "import",
  import: {
    location: "s3://my-bucket/orders/*.parquet",   // exact source file or glob
    format: "parquet",                              // parquet | csv | json (json = JSONL)
    overwrite: true                                 // default true
  }
}

-- definitions/clean_orders.sqlx — ref() the imported table like any other model
config { type: "table" }
SELECT order_id, total FROM ${ref("orders_in")}

How it runs per warehouse

BigQuery — compiles to native LOAD DATA OVERWRITE|INTO <table> FROM FILES(format=…, uris=['gs://…']). In-engine, using the warehouse’s own GCS access. gs:// only.
Postgres / Supabase — the CLI runs the import via DuckDB: it attaches the database read-write, then CREATE TABLE … AS SELECT * FROM read_parquet('<location>') (replace) or INSERT INTO … SELECT … (append). DuckDB decodes the file; the table lands in your warehouse. Reads from s3://, gs://, Supabase Storage (its S3-compatible endpoint), or a local path — filling the gap where managed Postgres/Supabase can’t read Parquet itself.

MySQL/MariaDB import isn’t supported yet.

`import` options

Field	Description
`location`	Required. The exact source file/glob/URI to read (e.g. `gs://b/orders/*.parquet`). Used verbatim — unlike export, no filename is derived. `${}` interpolation works.
`format`	Required. `parquet`, `csv`, or `json` (JSONL).
`overwrite`	`true` (default) replaces the table (drop + create). `false` appends (`INSERT … SELECT`) into an existing table.
`options`	Format-specific passthrough (reserved).

Schemes & per-warehouse validity

Scheme	BigQuery	Postgres / Supabase
`gs://`	✓ (native)	✓ (DuckDB)
`s3://` (incl. Supabase Storage)	✗	✓
`local://…` or a path	✗	✓ (handy for debugging)

Storage credentials

Cloud sources on Postgres/Supabase read credentials from the storage section of the gitignored .df-credentials.json — the same section exports use (BigQuery uses its own GCS access; local:// needs none).