Duckdb

DuckDB JSON Processing

DuckDB excels at JSON analytics: schema inference, SQL queries, complex joins, and streaming. Use when Nushell pipeline commands aren't enough.

Loading JSON

read_json — Auto-Detect Schema

Reads JSON files or URLs, auto-detects structure into typed columns. Arrays and objects become composite types.

-- Load from URL or file
select * from read_json('https://api.example.com/data');
select * from read_json('data.json');

-- In-memory (no .db file)
duckdb :memory:

Date-like columns auto-cast to timestamp. Use explicit schema for control:

-- Explicit column types
select * from read_json('films.json', columns = {
  title: varchar, release_date: date, created: timestamp
});

-- Cast specific columns
select title, created::datetime from read_json('films.json');

Unstructured JSON Mode

Returns raw JSON blobs instead of auto-detected columns:

-- Single column of JSON blobs
select json_data from read_json('data.json',
  format = 'unstructured', columns = {json_data: 'json[]'});

-- Unfold into rows
select unnest(json_data) as raw from read_json('data.json',
  format = 'unstructured', columns = {json_data: 'json[]'});

Inspecting JSON Structure

json_structure — Schema Detection

Returns the inferred schema of a JSON blob:

select json_structure(raw_data) from my_table limit 1;
-- {"title":"VARCHAR","episode_id":"UBIGINT",...}

Use with from_json for type conversion:

set variable json_schema = (select json_structure(raw_data) from my_table limit 1);
select from_json(raw_data, getvariable('json_schema')) as typed from my_table;

json_keys — Top-Level Keys

select unnest(json_keys(raw_data)) as keys from my_table limit 1;

unnest — Struct to Columns

Converts a struct into actual columns:

select unnest(from_json(raw_data, getvariable('json_schema'))) as row from my_table;

Extracting Values

json_extract (->) and json_extract_string (->>)

select
  raw_data->>'title' as title,
  (raw_data->>'episode_id')::uint64 as episode_id,
  (raw_data->'characters')::varchar[] as characters
from my_table;

-- Equivalent function form
select json_extract_string(raw_data, 'title') as title from my_table;

Unfolding Arrays into Rows

select
  json_extract_string(raw_data, 'title') as title,
  unnest(cast(json_extract(raw_data, 'characters') as varchar[])) as character_id
from my_table;

Warning: Using unnest on two different columns zips them by position. Normalize one table at a time.

Creating JSON

Struct Literal Syntax

Compose nested structures with { key: value } syntax:

select {
  type: 'character',
  name: character.name,
  homeworld: { type: 'planet', name: planet.name, climate: planet.climate },
  species: species.name
} as character
from character
join planet on planet.url = character.homeworld
left join species on species.url = character.species[1];

array_agg — Fold Values into Arrays

Inverse of unnest — groups rows back into arrays:

select
  { type: 'film', title: title, characters: array_agg(character_blob) } as film
from film_character
join character_blob on character_blob.url = film_character.character
group by film_character.url, film_character.title;

to_json — Cast to JSON String

select to_json(film_struct) as film_json from film_blob;
-- json() and json_string() are equivalent

Writing JSON

copy — Export JSON

-- Line-delimited JSON (default)
copy film to 'output.ndjson';

-- JSON array
copy film to 'output.json' (format json, array true);

Streaming JSON

Process JSON without intermediate files:

# stdin → transform → stdout
echo '[{"a":1,"b":[2,3]}]' | duckdb -json -c \
  "select a, unnest(b) as b from read_json('/dev/stdin')"

# Database → stdout
duckdb -c "copy (select title, unnest(characters) from film) to '/dev/stdout' (format json, array true)" db.db

Key Functions Reference

Resources

• DuckDB JSON Documentation
• MotherDuck: Analyze JSON with SQL
• DuckDB Blog: Shredding Nested JSON
• Wrangling JSON with DuckDB