我正在比较 SQL Server 2016(星型模式并使用基于列的聚集索引)和 BigQuery(一个表)之间的数据集市中的查询执行时间。我有大约 2000 万个条目。这是我的查询 - 我需要按月计算石油、天然气、水量的总和。10 年来,我每天都有单次条目。我有 6500 个实体,每个实体都有 10 年的石油、天然气、水量的单一条目……所以总行数……6500(实体总数)* 10(总年数)* 365(总天数)=23725000。 .暂时忽略闰年
BigQuery - 旧版 SQL(需要 5 秒)
SELECT [ASSET] AS [ASSET],
SUM([Measurements.GAS]) AS[sum_Measurements_GAS_ok],
SUM([Measurements.OIL]) AS[sum_Measurements_OIL_ok],
SUM([Measurements.WATER]) AS[sum_Measurements_WATER_ok],
STRFTIME_UTC_USEC(TIMESTAMP(TIMESTAMP_TO_MSEC(Measurements.DATE)), '%Y-%m') as [month]
FROM [datamanager-dashboard:bigquerysample.initial_va_schema_v3][initial_va_schema_v3]
GROUP BY 1, 5
SQL Server 2016 - 星型架构(耗时 2 秒) - 在 Google Compute Engine 中的 VM 内运行 - n1-standard-4
SELECT [dim_asset_types].[asset_name] AS Asset,
SUM([fact_well_volume_events].[gas]) AS [sum:gas:ok],
SUM([fact_well_volume_events].[oil]) AS [sum:oil:ok],
SUM([fact_well_volume_events].[water]) AS [sum:water:ok],
DATEADD(month, DATEDIFF(month, 0, [fact_well_volume_events].[measurement_date]), 0) AS [tmn:measurement_date:ok]
FROM [dbo].[dim_asset_types] [dim_asset_types]
INNER JOIN [dbo].[xref_well_to_asset_type] [xref_well_to_asset_type] ON ([dim_asset_types].[dim_asset_type_key] = [xref_well_to_asset_type].[dim_asset_type_key])
INNER JOIN [dbo].[dim_wells] [dim_wells] ON ([xref_well_to_asset_type].[dim_well_key] = [dim_wells].[dim_well_key])
INNER JOIN [dbo].[fact_well_volume_events_with_calculations] [fact_well_volume_events] ON ([dim_wells].[dim_well_key] = [fact_well_volume_events].[dim_well_key])
GROUP BY [dim_asset_types].[asset_name],DATEADD(month, DATEDIFF(month, 0, [fact_well_volume_events].[measurement_date]), 0)
我只举了一个例子,但它发生在各种不同的查询中。我错过了什么吗?为什么 BigQuery 这么慢?
编辑:我正在附加示例模式......它不完整..
[
{
"name": "ASSET",
"type": "STRING"
},
{
"name": "Measurements",
"type": "record",
"mode": "repeated",
"fields": [
{
"name": "DATE",
"type": "TIMESTAMP"
},
{
"name": "OIL",
"type": "FLOAT"
},
{
"name": "WATER",
"type": "FLOAT"
},
{
"name": "GAS",
"type": "FLOAT"
}
]
}
]