Order of records - finding preceding and next records

Question

I have looked around a few forums for a similar question but I'm not really even sure what to search for!
I have a table with some events, client names, timestamps for the event, and a category.  There are four steps in a cycle.
CREATE TABLE #events (EventID INT IDENTITY(1, 1),
Client VARCHAR(10),
RecordTimestamp DATETIME,
Step TINYINT)
INSERT INTO #events (Client,
RecordTimestamp,
Step)
VALUES ('Client A', '01-01-2014', 1),
('Client A', '01-02-2014', 2),
('Client A', '01-03-2014', 3),
('Client A', '01-04-2014', 4),
('Client A', '01-05-2014', 1),
('Client A', '01-06-2014', 2),
('Client B', '01-04-2014', 1),
('Client B', '01-06-2014', 2)
SELECT *
FROM #events
ORDER BY RecordTimestamp, Client
So let's assume that step 2 cannot start for a given client until step 1 completes, category 3 can't start until 2 completes, etc.  If today is 1/7/2014 we could say that Client A has completed one full cycle (steps 1-4), Client A is
half way through cycle 2, and Client B is half way through its first cycle.
I'm trying to identify all cycles that are in progress, meaning a step 1, 2, or 3 is present without a step 4 record (by client).  A query on the data above would return step 2 of Client A and Client B, both on 1/6.  I no longer care about the
first cycle from Client A because another cycle has started.
The table columns can be changed if necessary.  Can someone at least push me in the right direction?  It may involve a CTE and/or cursor, but I'm just not sure where to start.
Thanks!

Answer 1

>  I provided the example above as a slightly simplified version of what I'm working on
Not good, you should provide a something closer to what you are working on.
I read from your original post that "There are four steps in a cycle." and I see that you enumerate them 1 - 4 and also read that "So let's assume that step 2 cannot start
for a given client until step 1 completes, category 3 can't start until 2 completes, etc.", so based on those facts I decided to enumerate all steps for each customer (rn1) and used the formula (rn - 1) / 4 to identify blocks
of steps (groups of consecutive steps):
row                                   rn1  (rn1 - 1) / 4  
('Client A', '01-01-2014', 1)  1    0
('Client A', '01-02-2014', 2)  2    0
('Client A', '01-03-2014', 3)  3    0
('Client A', '01-04-2014', 4)  4    0
('Client A', '01-05-2014', 1)  5    1
('Client A', '01-06-2014', 2)  6    1
('Client B', '01-04-2014', 1)  1    0
('Client B', '01-06-2014', 2)  2    0
Then for each group we enumerate again (rn2) but this time in descending order (rn DESC) so the last step in each group will be the first row based on the new enumeration. Now we can select those rows where rn2 = 1 but only if the step number is not 4.
If your data is different, for example steps numbers go from 10 to 40 in increments of 10 then you have to change the logic slightly:
SET NOCOUNT ON;
USE tempdb;
GO
CREATE TABLE #events (
EventID INT IDENTITY(1, 1),
Client VARCHAR(10),
RecordTimestamp DATETIME,
Step tinyint,
UNIQUE CLUSTERED (Client, RecordTimestamp, Step)
INSERT INTO #events (Client,
RecordTimestamp,
Step)
VALUES
('Client A', '01-01-2014', 10),
('Client A', '01-02-2014', 20),
('Client A', '01-03-2014', 30),
('Client A', '01-04-2014', 40),
('Client A', '01-05-2014', 10),
('Client A', '01-06-2014', 20),
('Client B', '01-04-2014', 10),
('Client B', '01-06-2014', 20);
SELECT *
FROM #events
ORDER BY RecordTimestamp, Client
GO
WITH C1 AS (
SELECT
ROW_NUMBER() OVER(PARTITION BY Client ORDER BY RecordTimestamp, Step) AS rn1
FROM
#events
C2 AS (
SELECT
ROW_NUMBER() OVER(PARTITION BY Client, (rn1 - 1) / 4 ORDER BY rn1 DESC) AS rn2
FROM
C1
SELECT
EventID,
Client,
RecordTimestamp,
Step
FROM
C2
WHERE
rn2 = 1 AND Step < 40;
GO
DROP TABLE #events;
GO
If you can have 8 steps and the maximum one is 80 then the logic should be:
SET NOCOUNT ON;
USE tempdb;
GO
CREATE TABLE #events (
EventID INT IDENTITY(1, 1),
Client VARCHAR(10),
RecordTimestamp DATETIME,
Step tinyint,
UNIQUE CLUSTERED (Client, RecordTimestamp, Step)
INSERT INTO #events (Client,
RecordTimestamp,
Step)
VALUES
('Client A', '01-01-2014', 10),
('Client A', '01-02-2014', 20),
('Client A', '01-03-2014', 30),
('Client A', '01-04-2014', 40),
('Client A', '01-05-2014', 50),
('Client A', '01-06-2014', 60),
('Client A', '01-07-2014', 70),
('Client A', '01-08-2014', 80),
('Client A', '01-09-2014', 10),
('Client A', '01-10-2014', 20),
('Client B', '01-04-2014', 10),
('Client B', '01-06-2014', 20);
SELECT *
FROM #events
ORDER BY RecordTimestamp, Client
GO
WITH C1 AS (
SELECT
ROW_NUMBER() OVER(PARTITION BY Client ORDER BY RecordTimestamp, Step) AS rn1
FROM
#events
C2 AS (
SELECT
ROW_NUMBER() OVER(PARTITION BY Client, (rn1 - 1) / 8 ORDER BY rn1 DESC) AS rn2
FROM
C1
SELECT
EventID,
Client,
RecordTimestamp,
Step
FROM
C2
WHERE
rn2 = 1 AND Step < 80;
GO
DROP TABLE #events;
GO
If you want to show the last non-completed group for each client and from there the first and last steps then try:
SET NOCOUNT ON;
USE tempdb;
GO
CREATE TABLE #events (
EventID INT IDENTITY(1, 1),
Client VARCHAR(10),
RecordTimestamp DATETIME,
Step tinyint
INSERT INTO #events (Client,
RecordTimestamp,
Step)
VALUES
('Client A', '01-01-2014', 10),
('Client A', '01-02-2014', 20),
('Client A', '01-03-2014', 30),
('Client A', '01-04-2014', 40),
('Client A', '01-05-2014', 50),
('Client A', '01-06-2014', 60),
('Client A', '01-07-2014', 70),
('Client A', '01-08-2014', 80),
('Client A', '01-09-2014', 10),
('Client A', '01-10-2014', 20),
('Client A', '01-11-2014', 30),
('Client B', '01-04-2014', 10),
('Client B', '01-06-2014', 20);
SELECT *
FROM #events
ORDER BY RecordTimestamp, Client
GO
WITH C1 AS (
SELECT
ROW_NUMBER() OVER(PARTITION BY Client ORDER BY RecordTimestamp, Step) AS rn1
FROM
#events
C2 AS (
SELECT
DENSE_RANK() OVER(PARTITION BY Client ORDER BY (rn1 - 1) / 8 DESC) AS grp,
ROW_NUMBER() OVER(PARTITION BY Client, (rn1 - 1) / 8 ORDER BY rn1) AS rn2,
ROW_NUMBER() OVER(PARTITION BY Client, (rn1 - 1) / 8 ORDER BY rn1 DESC) AS rn3
FROM
C1
SELECT
EventID,
Client,
RecordTimestamp,
Step
FROM
C2
WHERE
grp = 1 AND 1 IN (rn2, rn3) AND Step < 80
ORDER BY
Client,
EventID;
GO
DROP TABLE #events;
GO
If you check the execution plan you will notice several SORT operators which could yield poor performance. We could try using an intermediate table and adding a POC (partition / ordering / covering - idea from ITzik Ben-Gan) index to improve it.
Does this help?
AMB
Some guidelines for posting questions...