Features overview

django-data-seed is a pipeline: it works out what to generate, makes each value fit the column, wires up relationships correctly, keeps each row coherent, and inserts everything fast — after checking your schema is sound.

• Data generation

  Every field type produces a valid value. Values are refined by the field's name (city → a city) and clamped to its constraints (max_length, choices, decimal precision, integer ranges, IP protocol).

• Relationships

  A foreign-key dependency graph seeds parents before children, reuses rows instead of exploding trees, and handles OneToOne, self-FKs, M2M, and cycles.

• Row coherence

  Fields within a row agree: ordered lifecycle dates, one persona per name set, one address per city/state/zip.

• Uniqueness

  Single- and multi-column uniqueness tracked in memory — no per-value SELECT, no collisions.

• Pre-flight checks

  Unapplied migrations, schema drift, and missing tables are caught before any row is written.

• File & media fields

  File/Image/FilePath fields get real, tiny files under your MEDIA_ROOT, honouring each field's upload_to.

The generation pipeline

For every field, the engine resolves a value in this order — first match wins:

1. User override — an explicit value/callable/pool you supplied.
2. NULL injection — for nullable, non-required columns (configurable).
3. Weighted choice — for choices fields (skewed, not uniform).
4. Field-name inference — email, city, price, latitude, …
5. Field-type provider — dispatched by the field's real MRO.

Whatever comes out is then decorated to fit the column (length, numeric range, decimal shape) and, for unique columns, checked against an in-memory set.

Performance & reproducibility

• bulk_create in batches; one transaction per model (--atomic all|model|none).
• Uniqueness handled in memory — zero per-value round-trips.
• --seed N makes the entire run byte-for-byte reproducible.
• Realism has an off switch: --realism uniform for dumb-fast noise.