Most efficient way to select record if a value has changed - sql

Question

What would be the most efficient way to select a record when one of the value has changed?
Ex:
I have an account history table like below where records are being created when the account change:
Id AcctNb Active Created
8 123456 1 01/03/2012
6 123456 0 01/01/2012
I like to find an efficient way to return the record where the active status has changed since the last entry.
UPDATE
The query I am using at the moment which works but inefficiently"
select d1.acctNb,d1.active, d2.active
from d044 d1 , d044 d2
where d1.created = '2012-04-14'
and d1.acctNb = d2.acctNb
and d2.created = (select max(d.created) from d044 d where d.acctNb = d2.acctNb and d.id != d1.id)
and (d1.active != d2.active)

Answer 1

Try this:
create table log
(
log_id int identity(1,1) primary key,
acct_id int not null,
active bit not null,
created datetime not null
);
insert into log(acct_id, active,created)
values
(1,1,'January 1, 2012'),
(1,1,'January 2, 2012'),
(1,0,'January 3, 2012'),
(1,0,'January 4, 2012'),
(1,1,'January 5, 2012'),
(2,0,'February 1, 2012'),
(2,1,'February 2, 2012'),
(2,0,'February 3, 2012'),
(2,1,'February 4, 2012'),
(2,1,'February 5, 2012');
The solution:
with serialize as
(
select row_number()
over(partition by acct_id order by created) rx,
*
from log
)
select ds.acct_id,
ds.active ds_active,
pr.active pr_active,
ds.created
from serialize ds -- detect second row
join serialize pr -- previous row
on pr.acct_id = ds.acct_id
and ds.rx = pr.rx + 1
where ds.rx >= 2 and
pr.active <> ds.active
Query output: January 3, January 5, February 2, February 3, February 4
Those are the dates when changes on active had occurred(detected)
Basically the logic is, starting from second row, we scan its previous row, if their active's value didn't match (via WHERE pr.active <> ds.active), we show them on results
Live test: http://sqlfiddle.com/#!3/68136/4

Answer 2

2 ways
1)
add a column
update_tsmp timestamp
put a trigger on the table that runs after update or insert
-- checks the Active field
-- if it has changed update update_tsmp to the current timestamp
now you must define "since the last entry" to determine whether you want to return the record
2)
create a history table
Id AcctNb Active Created change_tsmp updating_user delete_flag
put a trigger on the table that runs before update or delete
-- copies the record to the history table, checking the delete flag as appropriate

Answer 3

If future SQL Server will have LAG windowing function, you can simplify the comparison of previous row to current by using LAG
This works now on Postgresql(since 8.4), it already has LAG and LEAD windowing function:
create table log
(
log_id serial primary key,
acct_id int not null,
active boolean not null,
created timestamp not null
);
insert into log(acct_id, active,created)
values
(1,true,'January 1, 2012'),
(1,true,'January 2, 2012'),
(1,false,'January 3, 2012'),
(1,false,'January 4, 2012'),
(1,true,'January 5, 2012'),
(2,false,'February 1, 2012'),
(2,true,'February 2, 2012'),
(2,false,'February 3, 2012'),
(2,true,'February 4, 2012'),
(2,true,'February 5, 2012');
LAG approach is elegantly simpler than ROW_NUMBER and JOIN combo approach:
with merge_prev as
(
select
acct_id, created,
lag(active) over(partition by acct_id order by created) pr_active, -- previous row's active
active sr_active -- second row's active
from log
)
select *
from merge_prev
where
pr_active <> sr_active
Live test: http://sqlfiddle.com/#!1/b1eb0/25
EDIT
LAG is already available on SQL Server 2012: http://sqlfiddle.com/#!6/d17c0/1