Uniqueness

Seeding a table with unique columns naively means a database round-trip per value to check for collisions — an N+1 storm. django-data-seed tracks uniqueness in memory instead.

How it works

For each unique column, the engine loads the existing distinct values once, then generates against that in-memory set with zero further round-trips. UUID columns skip the check entirely — their collision odds are astronomical.

When a generated value collides, the engine retries; if a column's value space is tight (say a unique CharField(max_length=13)), it folds in seeded entropy so the run can't dead-end. Because the entropy comes from the seeded RNG, the result stays reproducible.

Multi-column uniqueness

unique_together and plain UniqueConstraint(fields=[...]) are tracked too. The engine builds the whole value-tuple for a row and checks it against the set of tuples already used before accepting the row, retrying the build until the tuple is free.

class Review(models.Model):
    book = models.ForeignKey(Book, on_delete=models.CASCADE)
    reviewer = models.ForeignKey(Author, on_delete=models.CASCADE)
    # ...
    class Meta:
        constraints = [
            models.UniqueConstraint(fields=["book", "reviewer"], name="one_review"),
        ]

Seeding Review produces at most one row per (book, reviewer) pair.

Per target database

All uniqueness checks run against the connection you're seeding (--using), not always default — so a multi-database run is correct.

Conditional & expression constraints

Plain, non-conditional constraints are tracked in memory. Conditional or expression constraints are left to the database to enforce.