Question

Given a table (TableA) that contains the following data;

Id    Date    Status    RecordId
1    01/06/11    2      REC001
2    01/06/11    2      REC002
3    01/06/11    2      REC003
4    01/07/11    1      REC001

How can I return all records with a status of 2 except records with a given RecordId where a status of 2 is followed by a record of 1 at a later date (and there are no further records with a status of 2.

So for example, the query should return REC002 and REC003 as REC001 had a status of 2 in the past, but that was superseeded by record Id 4 with a status of 1 at a later date. If, at some later point in time, another record was added for REC001 with a status of 2, then this should again be present in the result set (assuming there are no later records with a status of 1).

My feeble attempt at messing about with this is;

DECLARE @TableA TABLE
(
    Id INT,
    Dt DATETIME,
    Stat INT,
    RecId VARCHAR(6)
)

INSERT INTO @TableA 

SELECT   1,    DATEADD(day, -5, current_timestamp),  2,   'REC001'
UNION
SELECT   2,    DATEADD(day, -4, current_timestamp),  2,   'REC002'
UNION
SELECT   3,    DATEADD(day, -3, current_timestamp),  2,   'REC003'
UNION
SELECT   4,    DATEADD(day, -2, current_timestamp),  1,   'REC001'

   SELECT * 
     FROM @TableA t1
LEFT JOIN @TableA t2 ON t1.RecId = t2.RecId 
    WHERE t1.Stat = 2 
      AND (t1.Dt >= t2.Dt 
      AND t2.Stat != 1)

This kinda works, but returns values where t1.Id = t2.Id. I know I can exclude this through my where clause, but if I add a bunch more records to the table it fails again. For example;

INSERT INTO @TableA 
SELECT   1,    DATEADD(day, -15, current_timestamp),  2,   'REC004'
UNION
SELECT   2,    DATEADD(day, -14, current_timestamp),  2,   'REC002'
UNION
SELECT   3,    DATEADD(day, -13, current_timestamp),  1,   'REC003'
UNION
SELECT   4,    DATEADD(day, -12, current_timestamp),  1,   'REC001'
UNION
SELECT   11,    DATEADD(day, -5, current_timestamp),  2,   'REC004'
UNION
SELECT   21,    DATEADD(day, -4, current_timestamp),  2,   'REC002'
UNION
SELECT   31,    DATEADD(day, -3, current_timestamp),  1,   'REC003'
UNION
SELECT   41,    DATEADD(day, -2, current_timestamp),  1,   'REC001'

Any ideas are appreciated.

EDIT: I tried the two answers given, and while neither gave me exactly what I needed, they certainly pointed me in the right direction. Using the answers given, I came up with the following that seems to do what I require;

;WITH lastSuccess(recid, dt) AS (
    select recid, max(dt) from @tableA
    where stat = 1
    group by recid
),
lastFailure(recid, dt) AS (
    select recid, max(dt) from @tableA
    where stat = 2
    group by recid
)
select a.* from @tablea a
-- Limit results to those that include a failure
INNER JOIN lastFailure lf ON lf.recid = a.recid AND lf.dt = a.dt
-- If the recid also has a success, show this along with it's latest success date
LEFT JOIN lastSuccess ls ON ls.recid = lf.recid 
-- Limit records to where last failure is > last success or where there is no last success.
WHERE (lf.dt > ls.dt OR ls.dt IS NULL)

The only drawback I can see here is if there are two records with exactly the same timestamp then it would appear in the result set twice. For example, if Id 21 was repicated as 22 then, it would appear twice. This isn’t a real problem as in reality, the timestamp will always be unique.

Answer 1

WITH ranked AS (
  SELECT
    *,
    rn = ROW_NUMBER() OVER (PARTITION BY RecId ORDER BY Dt DESC)
  FROM TableA
)
SELECT
  r1.Id,
  r1.Dt,
  r1.Stat,
  r1.RecId
FROM ranked r1
  INNER JOIN ranked r2 ON r1.RecId = r2.RecId AND r2.rn = 1
WHERE r1.Stat = 2

---

UPDATE after the question’s update

WITH ranked AS (
  SELECT
    *,
    rn = ROW_NUMBER() OVER (PARTITION BY RecId ORDER BY Dt DESC)
  FROM TableA
)
SELECT
  Id,
  Dt,
  Stat,
  RecId
FROM ranked
WHERE Stat = 2 AND rn = 1