DB2 v9.5/AIX 6.1
I always thought that PKs are needed for referencial integrity/constraint only. I am new in this company and I find we are using import with insert_update option which updates existing records in the table with matching PKs. I was told we are using PKs only for this purpose.

Is there any other way to do same task like import...insert_update without creating/using PKs on the table? Thanks

Commands manual says

INSERT_UPDATE
Adds rows of imported data to the target table, or updates existing rows (of the target table) with matching primary keys.

Looks like the import is looking for a primary key (or maybe an unique key) to match imported records to existing records. I don't think it will work without the keys.

So this is the question: do I have to have primary keys for this or it's enough to have unique index on these cols?

Thank you

Technically spoken DB2 can do without a Primary Key definition. Personally I do not accept tables without a PK definition.
Unique key - Wikipedia, the free encyclopedia
There is a catch with DB2: you can add the keywords "primary key" in the column defintion as part of the CREATE TABLE. DB2 will automaticly create an index for you. When you create you own index later (with unique/cluster/include parameters) you will have 2 almost-the-same indexes on that table causing only overhead and no gain what-so-ever.
So what they taught me:
- create table without the PK definition
- create the unique index (can be more columns)
- alter table to add the PK contraint
Now DB2 will check if there is already an index which can be used and will not create its own index. If you own index cannot be used, DB2 will create an index with the same name as you contraint.

The basic point is that an index is not a constraint.

A unique index is not a declared constraint, although it acts as a defacto constraint.

That's a muddy area. An index (which is a concept for the physical layer of the database design) is just an access method. Personally, I find it really unfortunate to have something like an "unique index" where the index drags a constraint into the picture, which belongs to the conceptional database layer. My approach to avoid any misconceptions is therefore:
- if you need an access method, define an index
- if you need a unique constraint, define a primary key or a unique key

what ever dr te z said.

To answer your original question is you DO NOT need UK nor PK to do the insert_update.

I'm afraid you are wrong. Check SQL3203N, for example.

I am not afraid to be wrong thanks for pointing this out. I guess I better refresh my memory on it.