Google cloud platform BigQuery：让脱机用户在某个时间点和他们已脱机的时间点进行访问

假设BigQuery中有一个表，其中包含以下数据：

time  | user | status
 0:00 | 1    | start    <-- user 1 is online
 3:00 | 1    | stop     <-- user 1 is disconnected
 3:10 | 1    | stop     <-- a user can have multiple stops
 3:20 | 1    | stop     <-- a user can have multiple stops
 3:15 | 2    | start    <-- user 2 connected
 4:00 | 2    | stop     <-- user 2 disconnected
 5:00 | 1    | start    <-- user 1 is online again
 5:10 | 2    | stop     <-- user 2 still offline
 9:00 | 1    | start    <-- user 1 connected
10:00 | 1    | stop     <-- user 1 disconnected

---

我尝试过聚合、最后一个值结束、延迟结束，但我只能得到上一个事件。

这是一个很酷的挑战！下面是我为在特定日期连接的用户提出的建议

    WITH
  #Select all the previous state for each user
  select_previous AS (
  SELECT
    time,
    user,
    state,
    LAG(state) OVER (PARTITION BY user ORDER BY time) AS previous_state
  FROM
    `gbl-imt-homerider-basguillaueb.bqml_test.user_up_sof`),
  #Keep only the state change date
  state_change AS (
  SELECT
    *
  FROM
    select_previous
  WHERE
    previous_state != state
    OR previous_state IS NULL ),
  #Select the latest time registered before the wanted timeline
  max_time AS(
  SELECT
    MAX(time) AS time,
    user
  FROM
    state_change
  WHERE
    time < TIMESTAMP("2019-08-01 00:08:00")
  GROUP BY
    user)
  #Join the state change list with the latest time registered per user. Keep only the state stop
SELECT
  state_change.*
FROM
  max_time
JOIN
  state_change
ON
  max_time.time = state_change.time
  AND max_time.user = state_change.user
WHERE
  state = "stop"

不确定是否理解您提供的示例。 如果需要，我可以调整这些查询

    WITH
  #Select all the previous state for each user
  select_previous AS (
  SELECT
    time,
    user,
    state,
    LAG(state) OVER (PARTITION BY user ORDER BY time) AS previous_state
  FROM
    `gbl-imt-homerider-basguillaueb.bqml_test.user_up_sof`),
  #Keep only the state change date
  state_change AS (
  SELECT
    *
  FROM
    select_previous
  WHERE
    previous_state != state
    OR previous_state IS NULL ),
  #Select the latest time registered before the wanted timeline
  max_time AS(
  SELECT
    MAX(time) AS time,
    user
  FROM
    state_change
  WHERE
    time < TIMESTAMP("2019-08-01 00:08:00")
  GROUP BY
    user)
  #Join the state change list with the latest time registered per user. Keep only the state stop
SELECT
  state_change.*
FROM
  max_time
JOIN
  state_change
ON
  max_time.time = state_change.time
  AND max_time.user = state_change.user
WHERE
  state = "stop"

WITH
  #Select all the previous state for each user
  select_previous AS (
  SELECT
    time,
    user,
    state,
    LAG(state) OVER (PARTITION BY user ORDER BY time) AS previous_state
  FROM
    `gbl-imt-homerider-basguillaueb.bqml_test.user_up_sof`),
  #Keep only the state change date
  state_change AS (
  SELECT
    *
  FROM
    select_previous
  WHERE
    previous_state != state
    OR previous_state IS NULL ),
  #get the previous date change
  time_previous_change AS (
  SELECT
    *,
    LAG(time) OVER (PARTITION BY user ORDER BY time) AS previous_time
  FROM
    state_change )
  #perform a sum of the timeDiff, only when the state is start and the previous is stop (offline time)
SELECT
  user,
  SUM(timestamp_diff (time,
      previous_time,
      second)) AS elapsed_in_second
FROM
  time_previous_change
WHERE
  state="start"
  AND previous_state="stop"
GROUP BY
  user