Tengo una tabla que registra el ID de usuario, el curso, el sessionid y la fecha de solicitud cada vez que se carga una página web. Quiero calcular la duración por usuario para un curso determinado. Es problemático hacer esto debido a timespans superpuestos.Consulta SQL para calcular parte de la duración de la visita de la tabla de registro
Los datos proporcionados aquí deberían dar como resultado 10 minutos de duración por usuario para el curso 1. Parece que no puedo hacerlo bien.
CREATE TABLE PageLogSample (
id INT NOT NULL PRIMARY KEY IDENTITY
, userid INT
, courseid INT
, sessionid INT
, requestdate DATETIME
);
TRUNCATE TABLE PageLogSample;
INSERT INTO PageLogSample (userid, courseid, sessionid, requestdate)
-- [0, 10] = 10 minutes
SELECT 1, 1, 1, '00:00:00'
UNION ALL SELECT 1, 1, 1, '00:10:00'
-- [0, 12] - [3, 5] = 10 minutes
-- or ... [0, 3] + [5, 12] = 10 minutes
UNION ALL SELECT 2, 1, 2, '00:00:00'
UNION ALL SELECT 2, 2, 2, '00:03:00'
UNION ALL SELECT 2, 2, 2, '00:05:00'
UNION ALL SELECT 2, 1, 2, '00:12:00'
-- [0, 12] - [3, 5] = 10 minutes
-- or ... [0, 3] + [5, 12] = 10 minutes
UNION ALL SELECT 3, 1, 3, '00:00:00'
UNION ALL SELECT 3, 2, 3, '00:03:00'
UNION ALL SELECT 3, 2, 3, '00:05:00'
UNION ALL SELECT 3, 1, 3, '00:12:00'
UNION ALL SELECT 3, 2, 3, '00:15:00'
-- [1, 13] - [3, 5] = 10 minutes
-- or ... [1, 3] + [5, 13] = 10 minutes
UNION ALL SELECT 4, 2, 4, '00:00:00'
UNION ALL SELECT 4, 1, 4, '00:01:00'
UNION ALL SELECT 4, 2, 4, '00:03:00'
UNION ALL SELECT 4, 2, 4, '00:05:00'
UNION ALL SELECT 4, 1, 4, '00:13:00'
UNION ALL SELECT 4, 2, 4, '00:15:00'
-- [0, 5] + [10, 15] = 10 minutes
UNION ALL SELECT 5, 1, 5, '00:00:00'
UNION ALL SELECT 5, 1, 5, '00:05:00'
UNION ALL SELECT 5, 1, 6, '00:10:00'
UNION ALL SELECT 5, 1, 6, '00:15:00'
-- [0, 10] = 10 minutes (ignoring everything inbetween)
UNION ALL SELECT 6, 1, 7, '00:00:00'
UNION ALL SELECT 6, 1, 7, '00:03:00'
UNION ALL SELECT 6, 1, 7, '00:05:00'
UNION ALL SELECT 6, 1, 7, '00:07:00'
UNION ALL SELECT 6, 1, 7, '00:10:00'
-- [0, 11] - [5, 6] = 10 minutes
-- or ... [0, 3] + [7, 11] = 6 minutes (good)
-- or ... [0, 5] + [7, 11] = 9 minutes (better)
UNION ALL SELECT 7, 1, 8, '00:00:00'
UNION ALL SELECT 7, 1, 8, '00:03:00'
UNION ALL SELECT 7, 2, 8, '00:05:00'
UNION ALL SELECT 7, 2, 8, '00:06:00'
UNION ALL SELECT 7, 1, 8, '00:07:00'
UNION ALL SELECT 7, 1, 8, '00:11:00'
-- [0, 1] + [2, 4] + [5, 7] + [8, 13] = 10
UNION ALL SELECT 8, 1, 9, '00:00:00'
UNION ALL SELECT 8, 2, 9, '00:01:00'
UNION ALL SELECT 8, 1, 9, '00:02:00'
UNION ALL SELECT 8, 1, 9, '00:03:00'
UNION ALL SELECT 8, 2, 9, '00:04:00'
UNION ALL SELECT 8, 1, 9, '00:05:00'
UNION ALL SELECT 8, 1, 9, '00:06:00'
UNION ALL SELECT 8, 2, 9, '00:07:00'
UNION ALL SELECT 8, 1, 9, '00:08:00'
UNION ALL SELECT 8, 1, 9, '00:13:00'
;
first trying the naive approach. Esto genera errores al superponer partes de la sesión.
DECLARE @courseid INT;
SET @courseid = 1;
SELECT subquery.userid
, COUNT(DISTINCT subquery.sessionid) AS sessioncount
, SUM(subquery.duration) AS duration
, CASE SUM(subquery.duration)
WHEN 10 THEN 'ok'
ELSE 'ERROR'
END
FROM (
SELECT userid
, sessionid
, DATEDIFF(MINUTE, MIN(requestdate), MAX(requestdate)) AS duration
FROM PageLogSample
WHERE courseid = @courseid
GROUP BY userid
, sessionid
) subquery
GROUP BY subquery.userid
ORDER BY subquery.userid;
-- userid sessioncount duration
-- 1 1 10 ok
-- 2 1 12 ERROR
-- 3 1 12 ERROR
-- 4 1 12 ERROR
-- 5 2 10 ok
Segunda oportunidad. Evita las superposiciones. Esto solo funciona parcialmente
DECLARE @courseid INT;
SET @courseid = 1;
WITH cte (userid, courseid, sessionid, start, finish, duration)
AS (
SELECT userid
, courseid
, sessionid
, MIN(requestdate)
, MAX(requestdate)
, DATEDIFF(MINUTE, MIN(requestdate), MAX(requestdate))
FROM PageLogSample
GROUP BY userid
, courseid
, sessionid
)
SELECT naive.userid
, naive.sessioncount
, naive.duration AS naiveduration
, correction.duration AS correctionduration
, naive.duration - ISNULL(correction.duration, 0) AS duration
, CASE naive.duration - ISNULL(correction.duration, 0)
WHEN 10 THEN 'ok'
ELSE 'ERROR'
END
FROM (
SELECT cte.userid
, COUNT(DISTINCT cte.sessionid) AS sessioncount
, SUM(cte.duration) AS duration
FROM cte
WHERE cte.courseid = @courseid
GROUP BY cte.userid
) naive
LEFT JOIN (
SELECT errors.userid
, SUM(errors.duration) AS duration
FROM cte errors
WHERE errors.courseid <> @courseid
AND EXISTS (
SELECT *
FROM cte
WHERE cte.start <= errors.start
AND cte.finish >= errors.finish
AND cte.courseid = @courseid
)
GROUP BY errors.userid
) correction
ON naive.userid = correction.userid
;
-- userid sessioncount naiveduration correctionduration duration
-- 1 1 10 NULL 10 ok
-- 2 1 12 2 10 ok
-- 3 1 12 NULL 12 ERROR
-- 4 1 12 NULL 12 ERROR
-- 5 2 10 NULL 10 ok
Actualización: Ed Harpers comment realmente me hizo repensar mi enfoque.
Así que aquí viene la tercera prueba. Aquí primero busco las filas que representan una entrada al curso y que representan a alguien que se va. Luego tomo la suma de todos los tiempos del fin y restamos la suma de todas las begintimes. Creo que es más correcto, aunque no perfecto.
DECLARE @courseid INT;
SET @courseid = 1;
WITH numberedcte (rn, id, userid, courseid, sessionid, requestdate)
AS (
SELECT ROW_NUMBER() OVER (PARTITION BY sessionid, userid ORDER BY id)
, id
, userid
, courseid
, sessionid
, requestdate
FROM PageLogSample
)
, typedcte (rowtype, id, userid, courseid, sessionid, requestdate, nextrequestdate)
AS (
SELECT CASE
WHEN previousrequest.courseid = nextrequest.courseid
THEN 'between'
WHEN previousrequest.courseid IS NULL
OR nextrequest.courseid = numberedcte.courseid
THEN 'begin'
WHEN nextrequest.courseid IS NULL
OR previousrequest.courseid = numberedcte.courseid
THEN 'end'
ELSE 'error?'
END AS rowtype
, numberedcte.id
, numberedcte.userid
, numberedcte.courseid
, numberedcte.sessionid
, numberedcte.requestdate
, nextrequest.requestdate
FROM numberedcte
LEFT JOIN numberedcte previousrequest
ON previousrequest.userid = numberedcte.userid
AND previousrequest.sessionid = numberedcte.sessionid
AND previousrequest.rn = numberedcte.rn - 1
LEFT JOIN numberedcte nextrequest
ON nextrequest.userid = numberedcte.userid
AND nextrequest.sessionid = numberedcte.sessionid
AND nextrequest.rn = numberedcte.rn + 1
WHERE numberedcte.courseid = @courseid
AND (
nextrequest.courseid = @courseid
OR previousrequest.courseid = @courseid
)
)
, beginsum (userid, value)
AS (
SELECT userid, SUM(DATEPART(MINUTE, requestdate))
FROM typedcte
WHERE rowtype = 'begin'
GROUP BY userid
)
, endsum (userid, value)
AS (
SELECT userid, SUM(DATEPART(MINUTE, ISNULL(nextrequestdate, requestdate)))
FROM typedcte
WHERE rowtype = 'end'
GROUP BY userid
)
SELECT beginsum.userid
, endsum.value - beginsum.value AS duration
FROM beginsum
INNER JOIN endsum
ON beginsum.userid = endsum.userid
;
El único problema aquí es que solo me dan salida para el usuario 1 y 5 de mis datos de las muestras originales. El usuario agregado 6 también da salida correcta. El usuario agregado 7 me da un resultado satisfactorio ahora. El usuario 8 es casi perfecto, pierdo un minuto desde la primera fila hasta la segunda.
-- userid duration
-- 1 10
-- 5 10
-- 6 10
-- 7 9
-- 8 9
Siento que estoy a centímetros de conseguir esto completamente bien. Las únicas duraciones que faltan son las páginas solicitadas que no ocurrieron en grupos. ¿Alguien puede ayudarme a encontrar la forma de obtener vistas de página solitarias?
Actualización: Aquí viene una cuarta prueba. Aquí asigno un valor a cada solicitud y las resumo. No me da exactamente el resultado que esperaba, pero parece que podría ser lo suficientemente bueno.
DECLARE @courseid INT;
SET @courseid = 1;
WITH numberedcte (rn, userid, courseid, sessionid, requestdate)
AS (
SELECT ROW_NUMBER() OVER (PARTITION BY sessionid, userid ORDER BY id)
, userid
, courseid
, sessionid
, requestdate
FROM PageLogSample
)
, valuecte (value, userid, courseid, sessionid)
AS (
SELECT CASE
--alone
WHEN (previousrequest.courseid IS NULL
OR previousrequest.courseid <> numberedcte.courseid
)
AND nextrequest.courseid <> numberedcte.courseid
THEN DATEDIFF(MINUTE, numberedcte.requestdate, nextrequest.requestdate)
--between
WHEN previousrequest.courseid = nextrequest.courseid
THEN 0
--begin
WHEN previousrequest.courseid IS NULL
OR nextrequest.courseid = numberedcte.courseid
THEN -1 * DATEPART(MINUTE, numberedcte.requestdate)
--ignored (end with no next request)
WHEN nextrequest.courseid IS NULL
AND previousrequest.courseid <> numberedcte.courseid
THEN 0
--end
WHEN nextrequest.courseid IS NULL
OR previousrequest.courseid = numberedcte.courseid
THEN DATEPART(MINUTE, ISNULL(nextrequest.requestdate, numberedcte.requestdate))
--impossible?
ELSE 0
END
, numberedcte.userid
, numberedcte.courseid
, numberedcte.sessionid
FROM numberedcte
LEFT JOIN numberedcte previousrequest
ON previousrequest.userid = numberedcte.userid
AND previousrequest.sessionid = numberedcte.sessionid
AND previousrequest.rn = numberedcte.rn - 1
LEFT JOIN numberedcte nextrequest
ON nextrequest.userid = numberedcte.userid
AND nextrequest.sessionid = numberedcte.sessionid
AND nextrequest.rn = numberedcte.rn + 1
WHERE numberedcte.courseid = @courseid
)
SELECT userid
, courseid
, COUNT(DISTINCT sessionid) AS sessioncount
, SUM(value) AS duration
FROM valuecte
GROUP BY userid
, courseid
ORDER BY userid
;
Como puede ver, los resultados no son del todo lo que esperaba.
-- userid courseid sessioncount duration
-- 1 1 1 10
-- 2 1 1 3
-- 3 1 1 6
-- 4 1 1 4
-- 5 1 2 10
-- 6 1 1 10
-- 7 1 1 9
-- 8 1 1 10
El rendimiento es horrible en mi copia local de la base de datos real. Entonces, si alguien tiene ideas para escribir esto de una manera más eficiente ... disparar.
Actualización: El rendimiento está arriba. Agregué un índice y funciona un encanto ahora.
Buena pregunta, en términos de secuencia de comandos y muestra para que sea más fácil de resolver. – Andrew
La dificultad con los datos es que requestdate no tiene un significado consistente. A veces es la hora de inicio y algunas veces la hora de finalización de un curso. –
Buen comentario a la buena pregunta. –