Question

I have a table with columns like (PROPERTY_ID, GPSTIME, STATION_ID, PROPERTY_TYPE, VALUE) where PROPERTY_ID is primary key and STATION_ID is foreign key.

This table records state changes; each row represents property value of some station at given time. However, its data was converted from old table where each property was a column (like (STATION_ID, GPSTIME, PROPERTY1, PROPERTY2, PROPERTY3, ...)). Because usually only one property changed at time I have lots of duplicates.

I need to remove all successive rows with same values.

Example. Old table contained values like

time  stn   prop1  prop2
100   7     red    large
101   7     red    small
102   7     blue   small
103   7     red    small

The converted table is

(order by time,type)          (order by type,time)
time  stn type  value         time  stn type value
100   7   1     red           100   7   1    red
100   7   2     large         101   7   1    red
101   7   1     red           102   7   1    blue
101   7   2     small         103   7   1    red
102   7   1     blue          100   7   2    large
102   7   2     small         101   7   2    small
103   7   1     red           102   7   2    small
103   7   2     small         103   7   2    small

should be changed to

time  stn type  value
100   7   1     red
100   7   2     large
101   7   2     small
102   7   1     blue
103   7   1     red

The table contains about 22 mln rows.

My current approach is to use procedure to iterate over the table and remove duplicates:

BEGIN
    DECLARE done INT DEFAULT FALSE;
    DECLARE id INT;
    DECLARE psid,nsid INT DEFAULT null;
    DECLARE ptype,ntype INT DEFAULT null;
    DECLARE pvalue,nvalue VARCHAR(50) DEFAULT null;
    DECLARE cur CURSOR FOR 
        SELECT station_property_id,station_id,property_type,value 
        FROM station_property 
        ORDER BY station_id,property_type,gpstime;
    DECLARE CONTINUE HANDLER FOR NOT FOUND SET done = TRUE;
    OPEN cur;
    read_loop: LOOP
        FETCH cur INTO id,nsid,ntype,nvalue;
        IF done THEN 
            LEAVE read_loop;
        END IF;        
        IF (psid = nsid and ptype = ntype and pvalue = nvalue) THEN
            delete from station_property where station_property_id=id;
        END IF;
        SET psid = nsid;
        SET ptype = ntype;
        SET pvalue = nvalue;
    END LOOP;
    CLOSE cur;
END

However, it is too slow. On test table with 20000 rows it removes 10000 duplicates for 6 minutes. Is there a way to optimize the procedure?

P.S. I still have my old table intact, so maybe it is better to try and convert it without the duplicates rather than dealing with duplicates after conversion.

UPDATE.

To clarify which duplicates I want to allow and which not.

1. If a property changes, then changes back, I want all 3 records to be saved, even though first and the last contains same station_id, type, and value.
2. If there are several successive (by GPSTIME) records with same station_id, type, and value, I want only the first one (which represents the change to that value) to be saved.

In short, a -> b -> b -> a -> a should be optimized to a -> b -> a.

SOLUTION

As @Kickstart suggested, I’ve created new table, populated with filtered data. To refer previous rows, I’ve used approach similar to one used in this question.

rename table station_property to station_property_old;
create table station_property like station_property_old;

set @lastsid=-1;
set @lasttype=-1;
set @lastvalue='';

INSERT INTO station_property(station_id,gpstime,property_type,value)
select newsid as station_id,gpstime,newtype as type,newvalue as value from
-- this subquery adds columns with previous values
    (select station_property_id,gpstime,@lastsid as lastsid,@lastsid:=station_id as newsid,
    @lasttype as lasttype,@lasttype:=property_type as newtype,
    @lastvalue as lastvalue,@lastvalue:=value as newvalue
    from station_property_old
    order by newsid,newtype,gpstime) sub
-- we filter the data, removing unnecessary duplicates
where lastvalue != newvalue or lastsid != newsid or lasttype != newtype;

drop table station_property_old;

Answer 1

Possibly create a new table, populated with a select from the existing table using a GROUP BY. Something like this (not tested so excuse any typos):-

INSERT INTO station_property_new
SELECT station_property_id, station_id, property_type, value 
FROM (SELECT station_property_id, station_id, property_type, value, COUNT(*) FROM station_property GROUP BY station_property_id, station_id, property_type, value) Sub1