Finding last "Non-Null" value in a Pivot query - php

Question

I have the following query:
SELECT * FROM
(
SELECT unix_timestamp, input_raw, tag_id
from [200030].[dbo].inputs
WHERE inputs.date_time > dateadd(day,-1,getdate())
AND
(tag_id = 92164 or tag_id = 92149)
) src
pivot
(
max(input_raw)
FOR tag_id IN ([92164], [92149])
) piv
ORDER by unix_timestamp DESC
which is great and works. It gives me the results:
However, I would like the query to do one more thing for me.
Whenever there is a 'NULL' result, I would like the query to replace the 'NULL' with the last 'non-NULL' value in the column.
For example, the first NULL seen in column '92164' would be replaced with '211'.
Also, it is possible that there will be several 'NULL's i a row, so the query would need to keep going up the column until it finds a none NULL.
I have been able to accomplish this with php. Putting the results into a 2D associative array and running a function that finds nulls then loops through to find the last non NULL, but I really want to do this all in SQL if possible. I'd much rather use the
while( $row = sqlsrv_fetch_array( $stmt, SQLSRV_FETCH_ASSOC)){
//code
}
method than assigning several arrays.
Any help?
Thank you
//EDIT
Forgot to add that this only applies if there is any non-nulls above the null value. For example it is acceptable if the first row is NULL.

Answer 1

You can put your pivot query in a CTE and reuse the CTE when you figure out what value to show. You data have null values only in the first column but here is a version that deals with the nulls in both columns.
WITH C AS
(
SELECT *
FROM (
SELECT unix_timestamp, input_raw, tag_id
FROM inputs
WHERE date_time > dateadd(day,-1,getdate()) AND
(tag_id = 92164 OR tag_id = 92149)
) src
PIVOT
(
MAX(input_raw)
FOR tag_id IN ([92164], [92149])
) piv
)
SELECT C1.unix_timestamp,
(
SELECT TOP(1) C2.[92164]
FROM C AS C2
WHERE C1.unix_timestamp <= C2.unix_timestamp AND
C2.[92164] IS NOT NULL
ORDER BY C2.unix_timestamp
) AS [92164],
(
SELECT TOP(1) C2.[92149]
FROM C AS C2
WHERE C1.unix_timestamp <= C2.unix_timestamp AND
C2.[92149] IS NOT NULL
ORDER BY C2.unix_timestamp
) AS [92149]
FROM C AS C1
ORDER by C1.unix_timestamp DESC;
If you have a performance problem it might be faster for you to store the result from the CTE in a temp table with a useful clustered key index.
CREATE TABLE #C
(
unix_timestamp int PRIMARY KEY,
[92164] int,
[92149] int
);
INSERT INTO #C
SELECT *
FROM (
SELECT unix_timestamp, input_raw, tag_id
FROM inputs
WHERE date_time > dateadd(day,-1,getdate()) AND
(tag_id = 92164 OR tag_id = 92149)
) src
PIVOT
(
MAX(input_raw)
FOR tag_id IN ([92164], [92149])
) piv;
SELECT C1.unix_timestamp,
(
SELECT TOP(1) C2.[92164]
FROM #C AS C2
WHERE C1.unix_timestamp <= C2.unix_timestamp AND
C2.[92164] IS NOT NULL
ORDER BY C2.unix_timestamp
) AS [92164],
(
SELECT TOP(1) C2.[92149]
FROM #C AS C2
WHERE C1.unix_timestamp <= C2.unix_timestamp AND
C2.[92149] IS NOT NULL
ORDER BY C2.unix_timestamp
) AS [92149]
FROM #C AS C1
ORDER by C1.unix_timestamp DESC;
DROP TABLE #C;

Answer 2

You can use an adaptation of the "running total" calculation method detailed here. Create a temp table or table variable to hold the results of your pivot query, with an extra column to hold the last not-null value of the 92164 column. I can't get SQLFiddle to do this, but it works on my machine (famous last words)
CREATE TABLE #pivot_results ([unix_timestamp] int, [92164] int, [92149] int)
INSERT INTO #pivot_results ([unix_timestamp], [92164], [92149])
VALUES
(1361893407, NULL, 294),
(1361893218, 207, 294),
(1361893108, NULL, 292),
(1361892807, 211, 292),
(1361892799, NULL, 292)
CREATE TABLE #update_me
([unix_timestamp] int, [92164] int, [92149] int, last_not_null_92164 int)
DECLARE #last_not_null_92164 INT = NULL;
INSERT INTO #update_me([unix_timestamp], [92164], [92149], last_not_null_92164)
SELECT unix_timestamp, [92164], [92149], NULL
FROM #pivot_results
ORDER BY unix_timestamp DESC
UPDATE #update_me
SET #last_not_null_92164 = last_not_null_92164 = ISNULL([92164],#last_not_null_92164)
FROM #update_me
SELECT
unix_timestamp
,last_not_null_92164 AS [92164]
,[92149]
FROM #update_me
ORDER BY unix_timestamp DESC