SQL to ER diagram

How it works

1. 01

   Paste or drop your SQL

   CREATE TABLE statements, a migration file, or a full dump from pg_dump, mysqldump, SSMS, or sqlite3. Several .sql files can be dropped at once and are read together, up to 8 MB in all.

2. 02

   See the ER diagram

   Each table becomes a card that lists its columns with their types and PK, FK, and UK markers. Each foreign key becomes a line from the column that holds it to the key it points to.

3. 03

   Open it in Archboard

   Save it as a new diagram to rearrange, annotate, share a link, or export SVG and PNG. Signing up takes a Google or GitHub account.

What’s drawn from your SQL

The diagram is built from these parts of the CREATE TABLE statements:

• CREATE TABLE

  Every CREATE TABLE as a card named after the table, with one row per column: its name, its type, and a question mark when it allows NULL. Cards show up to 24 columns, then a count of the rest.

• PRIMARY KEYUNIQUE

  Primary keys, marked PK on every column they cover, even when they span several. A UNIQUE constraint, or a unique index made with CREATE UNIQUE INDEX, on a single column marks that column UK.

• FOREIGN KEYREFERENCESALTER TABLE

  Foreign keys written inline with REFERENCES, as table constraints, or added afterwards with ALTER TABLE, each drawn as a line between the two columns. A key that points back to its own table, such as parent_id, is shown as well.

• CREATE SCHEMA

  Schemas, whether created with CREATE SCHEMA or written in names such as sales.orders, drawn as frames around their tables when the SQL uses more than one. A schema with no tables gets no frame.

• NOT NULLUNIQUE

  Crow’s foot ends that read the relationship from the columns: a NOT NULL foreign key is required, a nullable one is optional, and a foreign key that is also unique or the primary key makes the relationship one-to-one.

SQL dialects it reads

PostgreSQL, MySQL and MariaDB, SQL Server, and SQLite are each read with their own quoting, comments, and statement separators, so a pg_dump file, a mysqldump with DELIMITER blocks, or an SSMS script full of GO lines works as it is. Oracle is read in part: tables, keys, and foreign keys are drawn, and PL/SQL blocks that end with a slash line are skipped. When the clues are mixed, a generic reader takes over, and you can pick the dialect yourself at any time.

What the ER diagram leaves out

• Rows of data. INSERT, COPY, and other data statements are skipped without being read, so a dump with data gives the same diagram as a schema-only one.
• Views, functions, procedures, triggers, and sequences. They’re skipped, and the preview tells you how many of each were left out; the views and triggers that mysqldump writes inside version comments are skipped as comments and not counted.
• Default values, CHECK expressions, comments, and the allowed values of ENUM and SET columns. Columns show their name, type, and key markers only.
• Anything from a running server. The tool reads the SQL text you give it and doesn’t connect to a database, so row counts, table sizes, and query plans aren’t shown.