Select rows ignoring narrow times - sql-server

Question

I have a table [EventLog] that contains reads data, recorded by a card reader that controls a gate. However, the same card code [epc] can be read multiple times, during card holder holding for some time near the reader.
I want to show reads for the same code, on the same reader, but ignoring reads for 2 minutes for example.
Example: EventLog
ID EPC ReaderID LogTime
1 1234 1 2016-04-15 12:33:55
2 1234 1 2016-04-15 12:34:05
3 1234 1 2016-04-15 12:34:10
4 4321 2 2016-04-15 12:34:12
5 4321 2 2016-04-15 12:34:14
Desired result:
ID EPC ReaderID LogTime
1 1234 1 2016-04-15 12:33:55
4 4321 2 2016-04-15 12:34:12
What I am using now is the windows function LAG to determine the difference in minutes between each read and it previous one:
SELECT EPC, ReaderName, PersonName, LogTime
FROM (
SELECT EPC, ReaderName, PersonName, LogTime,
DATEDIFF(MINUTE, LAG(LogTime) OVER (PARTITION BY EPC, ReaderID ORDER BY LogTime), LogTime) diff_prev
FROM EventLog l
LEFT OUTER JOIN Person p ON p.EPC = l.EPC
INNER JOIN Reader r ON r.ID = l.ReaderID
) tbl
WHERE diff_prev IS NULL OR diff_prev >= #ignoreMinutes
ORDER BY LogTime
Where #ignoreMinutes is a parameter that specifies how many minutes to ignore the same read.
But this solution is not correct in cases where the card is read once per second, for 3 hours. for Example:
ID EPC ReaderID LogTime diff_prev
1 1234 1 2016-04-15 12:33:55 NULL
2 1234 1 2016-04-15 12:34:05 0
3 1234 1 2016-04-15 12:34:10 0
4 1234 1 2016-04-15 12:34:32 0
5 1234 1 2016-04-15 12:34:54 0
6 1234 1 2016-04-15 12:35:14 0
7 1234 1 2016-04-15 12:35:34 0
8 1234 1 2016-04-15 12:35:54 0
9 1234 1 2016-04-15 12:36:04 0
10 1234 1 2016-04-15 12:36:15 0
11 4321 2 2016-04-15 12:44:12 NULL
12 4321 2 2016-04-15 12:44:14 0
As you see, my solution when executed with #ignoreMinutes = 1, will result in only 2 rows selected ID = 1, 11 since the rest are all diff_prev = 0. But the correct result set should be ID = 1, 6, 10, 11
Can you help? Thanks!

Answer 1

Here's a 'candidate' solution I came up with. At least it works correctly on your last example, returning records 1, 6, 10, 11.
DECLARE #intervalSeconds INT
SET #intervalSeconds = 60;
WITH EL AS
(
-- Select first record for each EPC, this is the baseline for recursion
SELECT
ID,
EPC,
LogTime
FROM EventLog
WHERE LogTime = (SELECT MIN(LogTime) FROM EventLog IEL WHERE IEL.EPC = EventLog.EPC)
-- Add following events
UNION ALL
SELECT
ID,
EPC,
LogTime
FROM
(
SELECT
NextEvent.ID,
NextEvent.EPC,
NextEvent.LogTime,
ROW_NUMBER() OVER(PARTITION BY NextEvent.EPC ORDER BY NextEvent.LogTime) eventNumber
FROM EventLog NextEvent
JOIN
(
SELECT
ID,
ROW_NUMBER() OVER(PARTITION BY EPC ORDER BY LogTime DESC) eventNumber, -- Reverse numbering to get last row by readNumber = 1
EPC,
LogTime
FROM EL -- Recursion
) PreviousEvent -- Here we have all already selected events wich we're interested in
ON PreviousEvent.EPC = NextEvent.EPC
AND PreviousEvent.eventNumber = 1 -- We need only the last one for each EPC
WHERE DATEDIFF(SECOND, PreviousEvent.LogTime, NextEvent.LogTime) > #intervalSeconds
) NextCandidateEvents -- Here we have all events with desired interval offset for each EPC
WHERE NextCandidateEvents.eventNumber = 1 -- We need only the first one for each EPC
)
SELECT * FROM EL
ORDER BY EPC, LogTime