Consulta SQL para calcular parte de la duración de la visita desde la tabla de registro
https://stackoverflow.com/questions/1613442
-
06-07-2019 - |
italiano
english
français
española
中国
日本の
العربية
Deutsch
한국어
Português
RussianPregunta
Tengo una tabla que registra el ID de usuario, el curso, el ID de sesión y la fecha de solicitud cada vez que se carga una página web. Quiero calcular la duración por ID de usuario para un ID de curso determinado. Es problemático hacer esto debido a la superposición de los intervalos de tiempo.
Los datos proporcionados aquí deben resultar en una duración de 10 minutos por usuario para el curso 1. Parece que no puedo hacer esto 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'
;
primero probando el enfoque ingenuo. Esto da errores con partes superpuestas 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
Segundo intento. Evite 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: comentario de Ed Harpers Realmente me hizo reconsiderar mi enfoque.
Así que aquí viene la tercera prueba. Aquí primero busco qué filas representan una entrada al curso y cuáles representan a alguien que se va. Luego tomo la suma de todos los tiempos finales y resto la suma de todos los comienzos. 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 obtengo resultados para los usuarios 1 y 5 de mis datos de muestra originales. El usuario añadido 6 también da salida correcta. El usuario agregado 7 me da una salida satisfactoria ahora. El usuario 8 es casi perfecto, pierdo un minuto de la primera fila a la segunda.
-- userid duration
-- 1 10
-- 5 10
-- 6 10
-- 7 9
-- 8 9
Siento que estoy a centímetros de tener esto completamente bien. Las únicas duraciones que faltan son de las solicitudes de páginas que no ocurrieron en grupos. ¿Puede alguien ayudarme a encontrar una manera de obtener las vistas de página solitarias?
Actualización: Aquí viene una cuarta prueba. Aquí asigno un valor a cada solicitud y los sumo. 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 eficaz ... dispara.
Actualización: El rendimiento está arriba. Agregué un índice y ahora funciona un hechizo.
Solución 4
Más datos de muestra y una suposición lógicamente esperada de cuánto tiempo pasó cada usuario en cada curso.
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, 3] = 3 minutes
-- there is no way to know how long the user was on that last page
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, 3] + [12, 15] = 6 minutes
-- the [5, 12] part was spent on a page of course 2
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, 3] + [13, 15] = 4 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, 5] + [7, 11] = 9 minutes
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'
-- there is nothing we can say about either of there requests
-- 0 minutes
UNION ALL SELECT 9, 1, 10, '00:10:00'
UNION ALL SELECT 9, 1, 11, '00:20:00'
;
Ahora obtenemos nuestros datos de esta manera:
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 in session
WHEN previousrequest.courseid IS NULL
AND nextrequest.courseid IS NULL
THEN 0
--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
;
Este es el resultado que obtengo. Estoy bastante contento con eso. Observe cómo el recuento de sesiones sigue siendo correcto para el usuario 9.
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
9 1 2 0
Otros consejos
Lo siento, pero creo que tienes un problema de datos. Observando los datos de muestra proporcionados, el usuario 2 está en courseid 1 durante 12 minutos y courseid 2 durante 2 minutos.
¿Seguro que ha proporcionado los datos correctos?
Esto es lo más cerca que puedo conseguir. Falla para el ID de usuario 4.
Como dije en mi comentario, requestdate es a veces un comienzo y otras veces un final de un curso, y no puedo ver una regla general simple para derivar qué rol desempeña en una fila dada .
DECLARE @courseid INT;
SET @courseid = 1;
WITH orderCTE
AS
(
SELECT *
,ROW_NUMBER() OVER (PARTITION BY sessionid
ORDER BY id
) AS rn
FROM PageLogSample
--order by rn
)
,startendCTE
AS
(
SELECT CASE WHEN start1.rn = 1
THEN start1.courseid
ELSE end1.courseid
END courseid
,start1.sessionid
,start1.userid
,DATEDIFF(mi,start1.requestdate,end1.requestdate) duration
FROM orderCTE AS start1
JOIN orderCTE AS end1
ON end1.rn = start1.rn + 1
AND end1.sessionid = start1.sessionid
)
SELECT courseid
,COUNT(1) sessionCount
,userid
,SUM(duration) totalDuration
FROM startendCTE
WHERE courseid = @courseid
GROUP BY courseid
,userid;
Esto es bastante complicado, pero parece que funciona para CourseID 1. No lo probé con otros cursos, ¡así que es posible que desees probarlo! : D
La premisa básica es que obtengo la duración del tiempo entre la primera y la última sesión del CourseID de destino y luego estoy restando la duración de cualquier sesión que no fuera del CourseID especificado, pero donde se solicita la sesión el tiempo cayó dentro de los tiempos de solicitud mínimo y máximo del CourseID objetivo. Espero que tenga sentido.
La consulta definitivamente se podría limpiar, posiblemente con un CTE o algo así. Pregunta interesante por cierto! :)
DECLARE @courseid INT;
SET @courseid = 1;
SELECT
TargetCourse.UserID,
COUNT(Distinct(TargetCourse.SessionID)) as SessionCount,
SUM(TargetCourse.Duration - Coalesce(OtherCourses.Duration,0)) as Duration
FROM
(
SELECT
TargetCourse.UserID, TargetCourse.SessionID,
MIN(TargetCourse.RequestDate) FirstRequest, MAX(TargetCourse.RequestDate) LastRequest,
DATEDIFF(MINUTE, MIN(TargetCourse.RequestDate), MAX(TargetCourse.RequestDate)) AS duration
FROM
PageLogSample TargetCourse
WHERE
TargetCourse.CourseID = @courseid
GROUP BY
TargetCourse.UserID, TargetCourse.SessionID
) as TargetCourse
LEFT OUTER JOIN
(
SELECT
OtherCourses.UserID, OtherCourses.SessionID,
MIN(OtherCourses.RequestDate) AS FirstRequest, MAX(OtherCourses.RequestDate) AS LastRequest,
DATEDIFF(MINUTE, MIN(OtherCourses.RequestDate), MAX(OtherCourses.RequestDate)) AS duration
FROM
PageLogSample OtherCourses
WHERE
OtherCourses.CourseID <> @courseid AND
OtherCourses.RequestDate between
(Select MIN(RequestDate) From PageLogSample T Where T.UserID = OtherCourses.UserID and T.CourseID = @courseid) AND
(Select MAX(RequestDate) From PageLogSample T Where T.UserID = OtherCourses.UserID and T.CourseID = @courseid)
GROUP BY
OtherCourses.UserID, OtherCourses.SessionID
) as OtherCourses ON
OtherCourses.UserID = TargetCourse.UserID AND
OtherCourses.FirstRequest BETWEEN TargetCourse.FirstRequest and TargetCourse.LastRequest
Group By TargetCourse.UserID
" Los datos son correctos, pero es difícil obtener un significado relevante de ellos. "
Estoy presionado para responder que esto es una contradicción de términos. Los datos de los que no sabes lo que significa no son datos.
En cuanto a su pregunta original:
Lo que necesita es un DBMS que ofrezca un soporte decente para los tipos de INTERVAL. Ningún sistema SQL juega en esa liga. Aparte de unos pocos sistemas de tutoriales, mi propio DBMS (no hay que insistir más en este contexto, por lo que no hay enlace) es el único que conozco que ofrece el tipo de soporte que realmente se necesita para tales problemas.
Si está interesado, busque en Google " tipos de intervalo " ;, " forma normal empaquetada " ;, " datos temporales " y eventualmente te encontrarás con él.
