bulk_create on Postgres doesn't use the unnest strategy when foreign keys are involved since 5.2.1

[canardoFR]

Dear Django team,

I'm very fresh & green to Django, but we had to insert millions of lines into a PostgreSQLtable.

We noticed that the ORM was setting up a "INSERT with UNNEST" for our bulk_create as long as there was no ForeignKey.

Is related to code here ​https://github.com/django/django/blob/main/django/db/backends/postgresql/compiler.py, line 42 to 45

I made a quick a dirty test by monkey patching the assemble_as_sql function to created a new list of tested fields ignoring ForeignKey type. I would have done better by testing the ForeignKey type, but as said above, to yound and green on django (some days).

I do not know if code is intentional or if this is a bug, thus the ticket submission.

the patch is attached as a file...

Thanks for reading and cheers,

Alain R.

[Simon Charette]

From looking at your attached monkey patching solution you seem to want to make sure the optimization introduced in Django 5.2 on Postgres to make use of UNNEST (see #35936) is enabled when foreign keys are involved which appears to have been inadvertently disabled by 764af7a3d6c0b543dcf659a2c327f214da768fe4 (see #36289).

Going to move a release blocker since this a regression in 5.2.1.

[GitHub <noreply@…>]

In 0fe2188:

Fixed #36502 -- Restored UNNEST strategy for foreign key bulk inserts on PostgreSQL.

Regression in 764af7a3d6c0b543dcf659a2c327f214da768fe4.

[Mariusz Felisiak <felisiak.mariusz@…>]

In 3df1ad57:

[5.2.x] Fixed #36502 -- Restored UNNEST strategy for foreign key bulk inserts on PostgreSQL.

Regression in 764af7a3d6c0b543dcf659a2c327f214da768fe4.

Backport of 0fe218842e0e396e3ab3982bd21227968a9e7fd8 from main.