Apprendre  /  Antisèches

Aide-mémoire DuckDB

Le SQL DuckDB de tous les jours, à jour pour DuckDB 2.0, sur une page à imprimer. Chaque requête a été exécutée sur DuckDB 2.0.0.

Antisèche5 min de lectureMis à jour le 16 novembre 2026
SQLBases de donnéesPython
Toutes les antisèches et guides
Tous les guides et antisèches

Interroger les fichiers directement

SQL
SELECT * FROM 'sales.csv';                    -- no import step
SELECT * FROM read_csv('sales_*.csv', union_by_name = true);
SELECT * FROM 'sales/*.parquet';
SELECT * FROM read_json('events.json');
DESCRIBE SELECT * FROM 'sales.parquet';       -- column names and types

Tables

SQL
CREATE TABLE sales AS FROM 'sales.csv';
CREATE OR REPLACE TABLE sales AS FROM read_csv('sales_*.csv');
INSERT INTO sales BY NAME SELECT 5 AS order_id, 'East' AS region;
SUMMARIZE sales;                 -- min, max, nulls, distinct, quartiles

SQL convivial

SQL
FROM sales;                                  -- = SELECT * FROM sales
FROM sales SELECT region, amount WHERE amount > 100;
SELECT * EXCLUDE (notes) FROM sales;
SELECT * REPLACE (round(amount) AS amount) FROM sales;
SELECT region, sum(amount) FROM sales GROUP BY ALL ORDER BY ALL;
SELECT max(COLUMNS('amount|tax')) FROM sales;   -- regex on names

Fenêtres et QUALIFY

SQL
SELECT region, date, amount,
       sum(amount) OVER (PARTITION BY region ORDER BY date) AS running
FROM sales;

-- top order per region, no subquery
SELECT * FROM sales
QUALIFY row_number() OVER (PARTITION BY region ORDER BY amount DESC) = 1;

Jointure ASOF

SQL
-- the latest price valid at each order date
SELECT s.order_id, p.rate
FROM sales s
ASOF JOIN prices p
  ON s.region = p.region AND s.date >= p.valid_from;
-- ASOF LEFT JOIN keeps orders with no match

Pivot et unpivot

SQL
CREATE TABLE wide AS
PIVOT (SELECT region, amount, month(date) AS month FROM sales)
ON region USING sum(amount) GROUP BY month;

UNPIVOT wide ON COLUMNS(* EXCLUDE (month))
INTO NAME region VALUE total;

Texte et dates

SQL
SELECT date_trunc('month', date)          AS month,
       strftime(date, '%d/%m/%Y')         AS uk_date,
       date_diff('day', date, DATE '2026-12-31') AS days_left,
       regexp_extract(code, '\d+')        AS number,
       region ILIKE 'n%'                  AS starts_with_n
FROM sales;

Listes et lambdas (2.0)

SQL
SELECT list_transform([1, 2, 3], lambda x: x * 10);   -- [10, 20, 30]
SELECT list_filter([1, 2, 3], lambda x: x > 1);       -- [2, 3]
SELECT [x * 10 FOR x IN [1, 2, 3]];                   -- comprehension
-- 2.0 disables the old x -> x + 1 arrow form by default

JSON et VARIANT (2.0)

SQL
SELECT payload ->> '$.user.id' AS user_id FROM raw_events;
SELECT json_set('{"a": 1}', '$.b', '2');               -- {"a":1,"b":2}

CREATE TABLE events (payload VARIANT);                 -- fast semi-structured
INSERT INTO events VALUES ('{"event": "click", "x": 10}'::JSON::VARIANT);
SELECT variant_type(payload), variant_keys(payload) FROM events;

Variables

SQL
SET VARIABLE min_amount = 100;
SELECT * FROM sales WHERE amount > getvariable('min_amount');

Exporter

SQL
COPY (FROM sales) TO 'out.parquet' (FORMAT parquet);
COPY sales TO 'out' (FORMAT parquet, PARTITION_BY (region));
COPY sales TO 'out.csv' (HEADER);
SELECT count(*) FROM 'out/**/*.parquet';               -- read it back

Cloud et autres bases

SQL
CREATE SECRET (TYPE s3, PROVIDER credential_chain);   -- uses your AWS login
SELECT * FROM 's3://my-bucket/sales/*.parquet';

ATTACH 'shop.duckdb' AS shop;                          -- another DuckDB file
ATTACH 'dbname=shop host=localhost' AS pg (TYPE postgres, READ_ONLY);
CREATE TABLE shop.sales_copy AS FROM sales;

Triggers (2.0)

SQL
CREATE TRIGGER trg_audit
  AFTER UPDATE ON target
  REFERENCING OLD TABLE AS o NEW TABLE AS n
  FOR EACH STATEMENT
    INSERT INTO audit
    SELECT n.id, o.val, n.val FROM o JOIN n ON o.id = n.id;

Mode serveur (2.0)

SQL
-- on the server
CALL quack_serve(token = 'my_token');

-- on a client
ATTACH 'quack:localhost' AS qk (TOKEN 'my_token');
CONNECT qk;
SELECT * FROM events;
-- listens on localhost only; use a TLS proxy for remote access

Python

Python
import duckdb

duckdb.sql("SELECT region, sum(amount) FROM 'sales.csv' GROUP BY ALL").df()

df = duckdb.sql("FROM 'sales.parquet'").df()   # to pandas
duckdb.sql("SELECT count(*) FROM df")           # query a DataFrame by name
duckdb.sql("FROM df").pl()                      # to Polars

con = duckdb.connect("shop.duckdb")             # a file on disk
con.sql("CREATE OR REPLACE TABLE t AS FROM df")

Bonnes habitudes

Interrogez du Parquet plutôt que du CSV dès que possible : c'est plus petit et beaucoup plus rapide. Filtrez et choisissez les colonnes tôt, car DuckDB ne lit que ce que vous demandez. Les fichiers créés avec DuckDB 2.0 ne s'ouvrent pas avec les versions précédentes : mettez à jour en même temps tous ceux qui partagent un fichier de base. Utilisez ORDER BY quand l'ordre compte ; sans lui, l'ordre des lignes n'est pas garanti.

Alessandro Ecclesie AgazziÉcrit par Alessandro Ecclesie Agazzi, analytics engineer à Londres. Mis à jour le 16 novembre 2026. À propos

Cet article vous a été utile ?

De nouveaux guides et décryptages des nouveautés en data et en IA paraissent ici régulièrement. Suivez-moi sur LinkedIn ou partagez cet article avec un collègue.