I missed this posting first time around – sorry about that.

I’ve just run a script like yours on 10.2.0.3 and 11.1.0.6 and Oracle doesn’t create a new index to protect the primary key.

It looks like it’s using the system-generated index on the hash key to support the primary key – possibly on the assumption that it should be quick enough to check for a duplicate by using the basic mechanisms of handling sorted hash clusters.

My comment was related to your conclusion that there was something very specific about sorted hash clusters.
That reminded me of something I discovered a long time ago.
But was I wrong?
Do you have an index when you create the primary key constraint at the same time as the table?
I do not see any index created with the following script:

drop table sorted_hash_table
;
select count(*) from dba_indexes
;
create table sorted_hash_table (
hash_value number,
sort_value varchar2(2), – sort,
v1 varchar2(10),
padding varchar2(100),
constraint sht_pk primary key(hash_value,sort_value)
)
cluster sorted_hash_cluster (
hash_value, sort_value
)
;
select count(*) from dba_indexes
;

The user_constraints view shows an index named “_NEXT_OBJECT” which must be a weakness of the view definition.

I’m not sure I understand your comment. In my example (hash_value, sort_value) is not intended to be the primary key. If I add a constraint to make it a primey key I expect to see (and did in a quick test on 10.2.0.1) a unique index created on those two columns.

Don’t worry, I was joking.

I also run a little blog and I know that life+work+study+blog can be equal to less spare time… :(

You are doing a very interesting work!
Thumbs up! :lol:

Sorry, finding the time to write this stuff can be a little difficult. But I’m going to be spending about 12 hours on trains with Wifi next week – so the wait might not be too long.

The sort is required for the cluster but is optional for the table – that’s why I’ve commented it only for the table.

We must wait the next one?

But I want to know it now! ^_^

Great article as always.

A very little error: on the first create cluster, the create cluster sorted_hash_cluster one, you miss a ‘–’ before the word “sort”