Question

我在导出的GA数据上使用BigQuery（请参阅架构here）

查看文档，我看到当我选择记录中的字段时，它会自动展平该记录并复制周围的列。

所以我尝试创建一个非规范化的表，我可以在更像SQL的思维模式中查询

SELECT
  CONCAT( date, " ", if (hits.hour < 10,
      CONCAT("0", STRING(hits.hour)),
      STRING(hits.hour)), ":", IF(hits.minute < 10, CONCAT("0", STRING(hits.minute)), STRING(hits.minute)) ) AS hits.date__STRING,
  CONCAT(fullVisitorId, STRING(visitId)) AS session_id__STRING,
  fullVisitorId AS google_identity__STRING,
  MAX(IF(hits.customDimensions.index=7, hits.customDimensions.value,NULL)) WITHIN RECORD AS customer_id__LONG,
  hits.hitNumber AS hit_number__INT,
  hits.type AS hit_type__STRING,
  hits.isInteraction AS hit_is_interaction__BOOLEAN,
  hits.isEntrance AS hit_is_entrance__BOOLEAN,
  hits.isExit AS hit_is_exit__BOOLEAN,
  hits.promotion.promoId AS promotion_id__STRING,
  hits.promotion.promoName AS promotion_name__STRING,
  hits.promotion.promoCreative AS promotion_creative__STRING,
  hits.promotion.promoPosition AS promotion_position__STRING,
  hits.eventInfo.eventCategory AS event_category__STRING,
  hits.eventInfo.eventAction AS event_action__STRING,
  hits.eventInfo.eventLabel AS event_label__STRING,
  hits.eventInfo.eventValue AS event_value__INT,
  device.language AS device_language__STRING,
  device.screenResolution AS device_resolution__STRING,
  device.deviceCategory AS device_category__STRING,
  device.operatingSystem AS device_os__STRING,
  geoNetwork.country AS geo_country__STRING,
  geoNetwork.region AS geo_region__STRING,
  hits.page.searchKeyword AS hit_search_keyword__STRING,
  hits.page.searchCategory AS hits_search_category__STRING,
  hits.page.pageTitle AS hits_page_title__STRING,
  hits.page.pagePath AS page_path__STRING,
  hits.page.hostname AS page_hostname__STRING,
  hits.eCommerceAction.action_type AS commerce_action_type__INT,
  hits.eCommerceAction.step AS commerce_action_step__INT,
  hits.eCommerceAction.option AS commerce_action_option__STRING,
  hits.product.productSKU AS product_sku__STRING,
  hits.product.v2ProductName AS product_name__STRING,
  hits.product.productRevenue AS product_revenue__INT,
  hits.product.productPrice AS product_price__INT,
  hits.product.productQuantity AS product_quantity__INT,
  hits.product.productRefundAmount AS hits.product.product_refund_amount__INT,
  hits.product.v2ProductCategory AS product_category__STRING,
  hits.transaction.transactionId AS transaction_id__STRING,
  hits.transaction.transactionCoupon AS transaction_coupon__STRING,
  hits.transaction.transactionRevenue AS transaction_revenue__INT,
  hits.transaction.transactionTax AS transaction_tax__INT,
  hits.transaction.transactionShipping AS transaction_shipping__INT,
  hits.transaction.affiliation AS transaction_affiliation__STRING,
  hits.appInfo.screenName AS app_current_name__STRING,
  hits.appInfo.screenDepth AS app_screen_depth__INT,
  hits.appInfo.landingScreenName AS app_landing_screen__STRING,
  hits.appInfo.exitScreenName AS app_exit_screen__STRING,
  hits.exceptionInfo.description AS exception_description__STRING,
  hits.exceptionInfo.isFatal AS exception_is_fatal__BOOLEAN
FROM
  [98513938.ga_sessions_20151112]
 HAVING
  customer_id__LONG IS NOT NULL
  AND customer_id__LONG != 'NA'
  AND customer_id__LONG != ''

我将此表的结果写入另一个表 denorm （展平，大数据集）。

当我使用

子句查询 denorm 时，我得到了不同的结果

WHERE session_id_STRING = "100001897901013346771447300813"

与包含上述查询（产生所需结果）

SELECT * FROM (_above query_) as foo where session_id_STRING = 100001897901013346771447300813

我确定这是设计上的，但如果有人能解释这两种方法之间的差异会非常有用吗？

Answer 1

我相信你说你确实选中了“＃Flat; Flatten Results＆＃34;什么时候创建输出表？我从你的问题中假设session_id_STRING是一个重复的字段？

如果这些是正确的假设，那么您所看到的正是您从上述文档中引用的行为。你让BigQuery要求＆＃34;压扁结果＆＃34;所以它将你重复的字段变成了一个不重复的字段，并复制了它周围的所有字段，这样你就有了一个平面（即没有重复的数据）表。

如果您在查询子查询时看到所需的行为，则应在创建表时取消选中该框。

Answer 2

查看文档，我看到当我选择一个字段时在记录中它会自动压平该记录复制周围的列。

这不正确。顺便说一句，请你指出文档 - 它需要改进。

选择字段不会使该记录变平。所以，如果你有一个表T，只有一个记录{a = 1，b =（2,2,3）}，那么

SELECT * FROM T WHERE b = 2

您仍然可以获得单个记录{a = 1，b =（2,2）}。此子查询中的SELECT COUNT（a）将返回1.

但是一旦用flatten = on编写此查询的结果，就会得到两条记录：{a = 1，b = 2}，{a = 1，b = 2}。来自展平表的SELECT COUNT（a）将返回2.

发生不受欢迎的扁平化

2 个答案: