PostgreSQL sequence based on another column

Question

Lets say I have a table as such:

Column   |     Type    |                        Notes
---------+------------ +----------------------------------------------         


        

Answer 1

You could use a window function to assign your SEQ values, something like:

INSERT INTO YourTable
    (ID, SEQ, DATA)
    SELECT ID, ROW_NUMBER() OVER(PARTITION BY ID ORDER BY DATA), DATA
        FROM YourSource

Answer 2

I had the same need to dynamicaly store a tree-like structure, not to add all IDs at once.
I prefer not use sequence table for each group as there could be thousands of them.
It run in an intensive multi-processing environment, so it has to be race-condition-proof.
Here the insert fonction for the 1st level. Other levels follow the same principle.

Each group as independent non-reusable sequencial IDs, the function receives a group name & sub-group name, gives you the existing ID or creates it & returns the new ID.
I tried a loop to have a single select, but the code is as long & harder to read.

CREATE OR REPLACE FUNCTION getOrInsert(myGroupName TEXT, mySubGroupName TEXT)
  RETURNS INT AS
$BODY$
DECLARE
   myId INT;
BEGIN -- 1st try to get it if it already exists
   SELECT id INTO myId FROM myTable
      WHERE groupName=myGroupName AND subGroupName=mySubGroupName;
   IF NOT FOUND THEN
      -- Only 1 session can get it but others can read
      LOCK TABLE myTable IN SHARE ROW EXCLUSIVE MODE; 
      -- 2nd try in case of race condition
      SELECT id INTO myId FROM myTable
         WHERE groupName=myGroupName AND subGroupName=mySubGroupName;
      IF NOT FOUND THEN -- Doesn't exist. Get next ID for this group.
         SELECT COALESCE(MAX(id), 0)+1 INTO myId FROM myTable
            WHERE groupName=myGroupName;
         INSERT INTO myTable (groupName, id, subGroupName)
            VALUES (myGroupName, myId, mySubGroupName);
      END IF;
   END IF;
   RETURN myId;
END;
$BODY$
  LANGUAGE plpgsql VOLATILE COST 100;

To try it:

CREATE TABLE myTable (GroupName TEXT, SubGroupName TEXT, id INT);
SELECT getOrInsert('groupA', 'subgroupX'); -- Returns 1
...
SELECT * FROM myTable;
 groupname | subgroupname | id 
-----------+--------------+----
 groupA    | subgroupX    |  1
 groupA    | subgroupY    |  2
 groupA    | subgroupZ    |  3
 groupB    | subgroupY    |  1

Answer 3

If seq reflects (or should reflect) the order in which the rows are inserted, I'd rather use a timestamp that gets populated automatically and generate the sequence number on the fly when selecting the rows using row_number():

create table some_table
( 
  id          integer   not null,
  inserted_at timestamp not null default current_timestamp,
  data text
);

The to get the seq column, you can do:

select id,  
       row_number() over (partition by id order by inserted_at) as seq,
       data
from some_table
order by id, seq;

The select is however going to be a bit slower compared to using a persisted seq column (especially with an index on id, seq).

If that becomes a problem you can either look into using a materialized view, or adding the seq column and then updating it on a regular basis (I would not do this in a trigger for performance reasons).

SQLFiddle example: http://sqlfiddle.com/#!15/db69b/1

Answer 4

Here's a simple way using standard SQL:

INSERT INTO mytable (id, seq, data)
SELECT << your desired ID >>,
       COUNT(*) + 1,
       'Quick brown fox, lorem ipsum, lazy dog, etc etc.'
FROM mytable
WHERE id = << your desired ID (same as above) >>;

See SQL Fiddle Demo.

(If you wanted to be a bit cleverer you could consider creating a trigger to update the row using the same method immediately after an insert.)

Answer 5

I don't have any postgresql-specific experience, but can you use a subquery in your insert statement? Something like, in Mysqlish,

INSERT INTO MYTABLE SET 
   ID=4, 
   SEQ=(  SELECT MAX(SEQ)+1 FROM MYTABLE WHERE ID=4  ),
   DATA="Quick brown fox, lorem ipsum, lazy dog, etc etc."

Answer 6

PostgreSQL supports grouped unique columns, as such:

CREATE TABLE example (
    a integer,
    b integer,
    c integer,
    UNIQUE (a, c)
);

See PostgreSQL Documentation - Section 5.3.3

Easy :-)