In order to replicate a table using transactional replication the table must have a primary key. The replication uses the primary key for tracking all changes on the published database and the subscriber. In fact, if you modify the primary key on the subscriber (I am talking about simple transactional replication) and afterward you modify the same record on the publisher the replication will fail, since it will not be able to find that record on the subscriber. In merge replication a rowguid column is added to each published table for that purpose.

XML indexes require PKs because under the covers a hidden table is created (called the “nodes table”) which has a 1:m relationship with the original table. The nodes table includes the PK from the original table so that if a query can perform best by doing a seek into the nodes table first, then finding related attributes in the base table, that can be done as efficiently as possible. The requirement for PK vs unique is probably just one of ease, so they don’t have to figure out which unique constraint/index to use if there are multiple…

@Jason – how does that affect transactional replication? I’m a replication novice.

@Damon – I don’t care about diagrams.

@SQLDenis – good point about the nullability!

@Adam – why do XML indexes require PK’s? I’m not familiar with XML indexes.

@Carolyn – I’m not sure why the DB engine would treat a clustered unique index any different than a clustered primary key. Could you provide an example or link to more information?

Thanks everyone for your comments and input!

A primary key does not allow nulls while a unique index allows 1 null, this might be problematic if you are trying to create a foreign key
Also by default a PK is clustered while a unique constraint is not

People just have to be aware of the differences…hey for all you know someone puts 2 unique constraints on the same column on sorted desc the other asc (I have seen this!!!)

create table bla(id int)

alter table bla add constraint uq_con unique (id asc)
alter table bla add constraint uq_con2 unique (id desc)