How to set primary key, then convert to autofield?

You cannot have a FK from DocumentCluster.pk to Document and at the same time make DocumentCluster.pk a serial column. That's a contradiction. It would have to be two separate columns. I suppose you mean the other way round: from Document to DocumentCluster.

This can be done with SQL DDL commands:

BEGIN;CREATE TABLE DocumentCluster (pk serial, lots text, of_fields int, ...);INSERT INTO DocumentCluster (pk, lots, of_fields, ...)SELECT pk, lots, of_fields, ... FROM DocumentORDER  BY pk; -- optionalALTER TABLE DocumentCluster ADD CONSTRAINT DocumentCluster_pkeyPRIMARY KEY (pk);  -- could be integrated in CREATE TABLEALTER TABLE Document  -- and not the other way round!ADD   CONSTRAINT pk_fk FOREIGN KEY (pk) REFERENCES DocumentCluster(pk);SELECT setval(pg_get_serial_sequence('DocumentCluster', 'pk'), max(pk))FROM   DocumentCluster;  -- update SEQUENCE to highest valueCOMMIT;

• Your obfuscation layer (Django) may be using double-quoted names like "Document", in which case you'd have to do the same ...

• About setval().

• About pg_get_serial_sequence()

• The PK could be declared in the CREATE TABLE statement, but it's cheaper to add it after rows have been inserted - assuming pk is unique not null.

More explanation and links:

• Manually update the counter used for generation of primary key in Hibernate?
• How to reset postgres' primary key sequence when it falls out of sync?
• Modify Django AutoField start value

Using south this is much simpler than I thought it would be.

Use the south schemamigration command as usual to build the schema migration that adds the DocumentCluster model.

Then use the datamigration command to build a skeleton for the next migration:

./manage.py datamigration yourapp populate_clusters

Then fill in the forwards method in the resulting python migration file so it looks like:

def forwards(self, orm):    max_id = -1    clusters_added = 0                                                                                                             for document in orm.Document.objects.all():        cluster = orm.DocumentCluster()        cluster.id = document.id        cluster.sub_document = document        cluster.save()        max_id = max(max_id, document.id)        clusters_added +=1    if max_id >= clusters_added:        orm.DocumentCluster.objects.raw("SELECT "\           "setval(pg_get_serial_sequence('yourapp_documentcluster',"\                                         "'id'), %d)" % max_id+1)

(The reverse method in this datamigration would simply delete all the DocumentModel instances.)

If this 'populate_clusters' migration step is run right after the migration step that added the DocumentCluster table then the DocumentCluster table is empty and the serial/autokey counter starts at zero. If you never deleted any Documents then the serial/autokey counter value will end up at the next unused value for Document pk values and you don't even have to bump it up as shown in Erwin's answer.

If, however, you have deleted Document instances or if somehow an id value was skipped then you'll need to bump up that serial/autokey counter using SQL. Django provides a way to directly execute raw SQL. (I don't think you can do without using raw SQL).

For safety sake you should check if cluster.id is <= the maximum document.id value seen in the loop.