Hive query generating identifiers for a sequence of row matching a condition

问题

Let's say I have the following hive table as input, let's call it connections:

userid  | timestamp   
--------|-------------
1       | 1433258019  
1       | 1433258020
2       | 1433258080
2       | 1433258083
2       | 1433258088
2       | 1433258170
[...]   | [...]

With the following query:

SELECT
    userid,
    timestamp,
    timestamp - LAG(timestamp, 1, 0) OVER w AS timediff
    CASE
      WHEN timediff > 60
      THEN 'new_session'
      ELSE 'same_session'
    END AS session_state
FROM connections
WINDOW w PARTITION BY userid ORDER BY timestamp ASC;

I'm generating the following output:

userid  | timestamp   | timediff   | session_state
--------|-------------|------------|---------------
1       | 1433258019  | 1433258019 | new_session
1       | 1433258020  | 1          | same_session
2       | 1433258080  | 1433258080 | new_session
2       | 1433258083  | 3          | same_session
2       | 1433258088  | 5          | same_session
2       | 1433258170  | 82         | new_session
[...]   | [...]       | [...]      | [...]

How would I do to generate that:

userid  | timestamp   | timediff   | sessionid
--------|-------------|------------------------------
1       | 1433258019  | 1433258019 | user1-session-1
1       | 1433258020  | 1          | user1-session-1
2       | 1433258080  | 1433258080 | user2-session-1
2       | 1433258083  | 3          | user2-session-1
2       | 1433258088  | 5          | user2-session-1
2       | 1433258170  | 82         | user2-session-2
[...]   | [...]       | [...]      | [...]

Is that possible using only HQL and "famous" UDFs (I'd rather not use custom UDFs or reducer scripts) ?

回答1:

Interesting question. Per your comment to @Madhu, I added the line 2 1433258172 to your example. What you need is to increment every time timediff > 60 is satisfied. The easiest way to do this is to flag it and then cumulatively sum over the window.

Query:

select userid
  , timestamp
  , concat('user', userid, '-session-', s_sum) sessionid
from (
  select *
    , sum( counter ) over (partition by userid
                           order by timestamp asc
                           rows between unbounded preceding and current row) s_sum
  from (
    select *
      , case when timediff > 60 then 1 else 0 end as counter
    from (
      select userid
        , timestamp
        , timestamp - lag(timestamp, 1, 0) over (partition by userid
                                                 order by timestamp asc) timediff
      from connections ) x ) y ) z

Output:

1   1433258019  user1-session-1
1   1433258020  user1-session-1
2   1433258080  user2-session-1
2   1433258083  user2-session-1
2   1433258088  user2-session-1
2   1433258170  user2-session-2
2   1433258172  user2-session-2

回答2:

Use the following select concat_ws('-',name, city) from employee; the first parameter of concat_ws is separator. name and city are column names for employee table. See that they are of type strings. You can look here for more

回答3:

This works:

SELECT 
  userid,
  timestamp,
  timediff,
  CONCAT(
    'user',
     userid,
     '-',
     'session-',
     CAST(timediff / 60 AS INT) + 1
  ) AS session_id
  FROM (
    SELECT   
      userid,
      timestamp,
      timestamp - LAG(timestamp, 1, timestamp) OVER w AS timediff
    FROM connections
    WINDOW w AS (
      PARTITION BY userid
      ORDER BY timestamp ASC
    )
) a;

OUTPUT:

userid  timestamp   timediff    session_state
1       1433258019  0.0         user1-session-1
1       1433258020  1.0         user1-session-1
2       1433258080  0.0         user2-session-1
2       1433258083  3.0         user2-session-1
2       1433258088  5.0         user2-session-1
2       1433258170  82.0        user2-session-2
3       1433258270  0.0         user3-session-1

you can try something like this if timediff is not required:

select userid,timestamp ,session_count+ concat('user',userid,'-','session-',cast(LAG(session_count-1,1,0) over w1 as string)) AS session_state
--LAG(session_count-1,1,0) over w1 AS session_count_new FROM (select userid, timestamp, timediff, cast (timediff/60 as int)+1 as session_count

来源：https://stackoverflow.com/questions/30600543/hive-query-generating-identifiers-for-a-sequence-of-row-matching-a-condition