如何设计一个数据库来存储不同分辨率的OHLC时间序列？

Question

如果我可能对股票有不同的频率，那么存储各种股票的 OHLC 数据的最佳方法是什么？例如，我可能有：

* OHLC for 5-minute bars for APPL
* OHLC for 1-minute bars for APPL
* OHLC for 5-minute bars for IBM

我正在考虑将所有内容存储在同一个表中，然后添加一个指定分辨率的列，因此它可能如下所示：

symbol, date,       time, resolution, open,  high,   low,   close
AAPL,   2017-06-19, 9:30, 5 min,      99.12, 102.52, 94.22, 98.34   
AAPL,   2017-06-19, 9:30, 1 min,      99.12, 100.11, 99.01, 100.34
IBM,    2017-06-19, 9:30, 5 min,      40.15, 45.78,  39.18, 44.22

看起来不错吗？

Answer 1

看起来确实不错。正如您的另一种可能性一样，您还可以将每个新分辨率存储为（重复字段）STRUCT内的单独（记录）ARRAY，如下所示：

WITH data AS(
  select 'APPL' as symbol, ARRAY<STRUCT<date string, time string, resolution INT64, open FLOAT64, high FLOAT64, low FLOAT64, close FLOAT64>> [STRUCT('2017-06-19' as date, '9:30' as time, 5 as resolution, 99.12 as open, 102.52 as high, 94.22 as low, 98.32 as close), STRUCT('2017-06-19' as date, '9:30' as time, 1 as resolution, 99.12 as open, 100.11 as high, 99.01 as low, 100.34 as close)] stock union all
  select 'IBM' as symbol, ARRAY<STRUCT<date string, time string, resolution INT64, open FLOAT64, high FLOAT64, low FLOAT64, close FLOAT64>> [STRUCT('2017-06-19' as date, '9:30' as time, 5 as resolution, 40.15 as open, 45.78 as high, 39.18 as low, 44.22 as close)]
)

SELECT * FROM data

结果是：

请注意，当您存储新的分辨率值时，它会在为每个股票定义的 ARRAY 中添加另一行。

您还可以在日期级别聚合数组，如下所示：

WITH data AS(
  select 'APPL' as symbol, STRUCT<date string, time string, hit ARRAY<STRUCT<resolution INT64, open FLOAT64, high FLOAT64, low FLOAT64, close FLOAT64>>> ('2017-06-19', '9:30', [STRUCT(1 as resolution, 99.12 as open, 102.52 as high, 94.22 as low, 98.32 as close), STRUCT(5 as resolution, 99.12 as open, 100.11 as high, 99.01 as low, 100.34 as close)]) stock union all
  select 'IBM' as symbol, STRUCT<date string, time string, hit ARRAY<STRUCT<resolution INT64, open FLOAT64, high FLOAT64, low FLOAT64, close FLOAT64>>> ('2017-06-19', '9:30', [STRUCT(1 as resolution, 40.15 as open, 45.78 as high, 39.18 as low, 44.22 as close)])
)
SELECT * FROM data

结果是：

这种类型的模式可能会给您带来一些优势，具体取决于您正在处理的数据量，例如更便宜、更有效的存储以及更快的查询（有时您可能会发现返回错误的查询与其Resources Exceeded工作之间的区别在于明智的使用结构体和数组）。