Question

想象一下，你有两个事件表。表A和表B都有一个列，称为时间戳，有多行。

现在我想将这两个表组合成一个具有以下属性的表C：

C在A
C有一个时间戳列，完美地反映了A
C有另一个名为near_event的列，如果在该行的时间戳的1s内有B行，则为true，否则为false

我怎样才能有效地做到这一点？

Answer 1

mauro向我指出了这一点，说Vertica可以做得更好 - 事实上，它可以，因为它有一个谓词，可以实现我们所谓的事件系列连接。您需要做的就是运行非内连接（左，右或全外）并智能地使用INTERPOLATE PREVIOUS VALUE作为连接谓词。

您可能想查看我的LinkedIn帖子：

https://www.linkedin.com/pulse/verticas-event-series-join-joining-two-time-tables-marco-gessner/

..这说明了一个更复杂的用例。

使用与该博客相同的表格：

CREATE LOCAL TEMPORARY TABLE oilpressure (
 op_ts,op_psi
) ON COMMIT PRESERVE ROWS AS (
          SELECT TIMESTAMP '2015-04-01 07:00:00', 25.356
UNION ALL SELECT TIMESTAMP '2015-04-01 07:00:10', 35.124
UNION ALL SELECT TIMESTAMP '2015-04-01 07:00:20', 47.056
UNION ALL SELECT TIMESTAMP '2015-04-01 07:00:30', 45.225
)
;

CREATE LOCAL TEMPORARY TABLE revspeed (
 rs_ts,rpm
) ON COMMIT PRESERVE ROWS AS (
          SELECT TIMESTAMP '2015-04-01 07:00:00', 2201
UNION ALL SELECT TIMESTAMP '2015-04-01 07:00:08', 3508
UNION ALL SELECT TIMESTAMP '2015-04-01 07:00:15', 6504
UNION ALL SELECT TIMESTAMP '2015-04-01 07:00:20', 6608
)
;

让oilpressure成为您的表格A，revspeed成为您的表格B.

然后您想要的（如果您只想要时间戳）是：

SELECT
  op_ts
, rs_ts
FROM oilpressure
LEFT JOIN revspeed
ON op_ts INTERPOLATE PREVIOUS VALUE rs_ts;
op_ts              |rs_ts
2015-04-01 07:00:00|2015-04-01 07:00:00
2015-04-01 07:00:10|2015-04-01 07:00:08
2015-04-01 07:00:20|2015-04-01 07:00:20
2015-04-01 07:00:30|2015-04-01 07:00:20

Answer 2

如果您没有太多重复项，您可以这样做。这是一个想法：

select timestamp,
       (case when timestamp < timestamp_add(second, 1, last_b_timestamp) or
                  timestamp > timestamp_add(second, -1, next_b_timestamp)
             then 1 else 0
        end) as flag
from (select timestamp, which,
             last_value(case when which = 'b' then timestamp) over (order by timestamp) as last_b_timestamp,
             last_value(case when which = 'b' then timestamp) over (order by timestamp desc) as next_b_timestamp,
      from ((select a.timestamp, 'a' as which from a) union all
            (select b.timestamp, 'b' as which from b)
           ) ab
     ) ab
where which = 'a';

如何根据时间戳组合表

2 个答案: