Thread: duplicate key violates unique constraint

List,

 

We're using 8.2.1 in RedHat ES 4 here and I have a simple two column table: data_keys1.

 

                         Table "public.data_keys1"
 Column |  Type  |                        Modifiers
--------+--------+---------------------------------------------------------
 id     | bigint | not null default nextval('data_keys1_id_seq'::regclass)
 key1   | text   |
Indexes:
    "data_keys1_pkey" PRIMARY KEY, btree (id)

I also have a C program using libpq that populates this table via a PREPARE'd statement within a transaction.  Recently, while this table was being loaded, it encountered an error I hadn't seen before and is a bit confusing.

 

2007-05-09 09:27:07 EDT [22853] : LOG:  statement: EXECUTE insertKey1('vgdisplay');
2007-05-09 09:27:07 EDT [22853] : DETAIL:  prepare: PREPARE insertKey1(text) AS INSERT INTO public.data_keys1 (key1) values ($1);
2007-05-09 09:27:07 EDT [22853] : ERROR:  duplicate key violates unique constraint "data_keys1_pkey"
2007-05-09 09:27:07 EDT [22853] : STATEMENT:  EXECUTE insertKey1('vgdisplay');

 

As you can see, it's only providing the key1 column and the sequence is providing the value for the column with the constraint.  How can this be happening?  There were no other transactions, commits, individual inserts happenning at the same time or within several seconds of this one.

 

I'd like to explore sequences a bit more b/c as they are implemented in PostgreSQL is a little confusing.  Normally, I'd start another thread but it may have some bearing here.

 

 

Sequences...

 

Now, I don't know if this just hasn't been tested or is a documented feature (a.k.a. bug) but something does not seem right here.  In this test case I'm able to get the same sequence ID's via two psql connections to the same database on the same sequence.

 

Connect to "db" in two different psql sessions (I'll prefix them below with 1: and 2:) and in one create the table

 

1: db=# CREATE TABLE t1 ( s serial, i int);

1: db=# \d t1
                         Table "public.t1"
 Column |  Type   |                   Modifiers
--------+---------+------------------------------------------------
 s      | integer | not null default nextval('t1_s_seq'::regclass)
 i      | integer |

1: db=# SELECT * FROM t1_s_seq;

 sequence_name | last_value | increment_by |      max_value      | min_value | cache_value | log_cnt | is_cycled | is_called
---------------+------------+--------------+---------------------+-----------+-------------+---------+-----------+-----------
 t1_s_seq      |         12 |            1 | 9223372036854775807 |         1 |           1 |      32 | f         | t
(1 row)

 

So, cache on t1_s_seq is set to 1.  Not sure why the max_value is so high when the column was specified as a 'serial' not a 'serial8' but perhaps I should check the code out and submit a patch for that.  I digress.

 

Check out the value for t1_s_seq on connection #1.

 

1: db=# select nextval('t1_s_seq');
 nextval
---------
       1
(1 row)

 

And check the value for t1_s_seq on connection #2.

 

2: db=# select nextval('t1_s_seq');
 nextval
---------
       2
(1 row)

 

So far, so good.  Now start a transaction on connection #1, advance t1_s_seq by 10 but don't commit;

 

1: db=# begin;
BEGIN
1: db=# select setval('t1_s_seq', currval('t1_s_seq')+10);
 setval
--------
     11
(1 row)

 

And check the current value for t1_s_seq on connection #2.

 

2: db=# select currval('t1_s_seq');
 nextval
---------
       2
(1 row)

 

 

That's expected since the transaction on connection #1 hasn't been commited.  Now commit the transaction on connection #1 and check it's current value.

 

1: db=# commit;

COMMIT

1: db=# select currval('t1_s_seq');
 currval
---------
      11
(1 row)

 

Again, expected.  Now let's check the current value on connection #2 again.

 

2: db=# select currval('t1_s_seq');
 currval
---------
       2
(1 row)

 

This is where I take issue with the output.  I'm not sure what benefit cache value has as the client should, in this case, consulted with the backend as too the value of the sequence.

 

Sequences are suppose to be unique but in this case it seems that may not always be the case.  Sequences have some kind of odd relationship (no pun intended) with transactions in that they are in some cases in sync regardless of the connection or query and in other situations, such as above, are out of sync.

 

Just for fun, select on connection #2 the nextval of the sequence.

 

2: db=# select nextval('t1_s_seq');
 nextval
---------
      12
(1 row)

 

*boogle*

 

What have I done wrong here?  Does it have any bearing on my unique constraint error?  I wouldn't think so but I haven't found any other possible explanation.

 

TIA,

Greg

 

--

 Greg Spiegelberg

 Manager, Product Development

 ISOdx Solutions, a division of Cranel, Inc.

 gspiegelberg@isodxsolutions.com

 614.318.4314, office

 614.431.8388, fax