Question

我用过的代码如下所示:(只是一个简单的连接）

tbl(sc, 'dez') %>% inner_join(tbl(sc, 'deg'), by = c("timefrom" = "timefromdeg", "elemuid")) %>% 
    filter(number.x > 2500) %>% glimpse()

单个数据帧的内容无关紧要。连接本身会起作用。为了节省计算能力，我想在加入后直接过滤（或其他）。

但现在我收到错误消息， Spark 无法解析变量 number.x 。

我不明白，因为变量是错误信息的一部分：

Error: org.apache.spark.sql.AnalysisException: cannot resolve '`number.x`' given input columns: [elemname.x, kind.y, timefrom, timetodeg, timeto, kind.x, elemuid, elemname.y, number.y, number.x]; line 7 pos 7;
'Project [*]
+- 'Filter ('number.x > 2500.0)
   +- SubqueryAlias yoxgbdyqlw
      +- Project [elemuid#7505 AS elemuid#7495, elemname#7506 AS elemname.x#7496, kind#7507 AS kind.x#7497, number#7508 AS number.x#7498, timefrom#7509 AS timefrom#7499, timeto#7510 AS timeto#7500, elemname#7512 AS elemname.y#7501, kind#7513 AS kind.y#7502, number#7514 AS number.y#7503, timetodeg#7516 AS timetodeg#7504]
         +- Join Inner, ((timefrom#7509 = timefromdeg#7515) && (elemuid#7505 = elemuid#7511))
            :- SubqueryAlias TBL_LEFT
            :  +- SubqueryAlias dez
            :     +- HiveTableRelation `default`.`dez`, org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe, [elemuid#7505, elemname#7506, kind#7507, number#7508, timefrom#7509, timeto#7510]
            +- SubqueryAlias TBL_RIGHT
               +- SubqueryAlias deg
                  +- HiveTableRelation `default`.`deg`, org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe, [elemuid#7511, elemname#7512, kind#7513, number#7514, timefromdeg#7515, timetodeg#7516]

加入后collect()不是一个选项，因为那时我的内存不足。是否有任何可能使事情发生的可能性。

我会很乐意帮助你！

Answer 1

TL; DR 请勿使用默认suffix（c(".x", ".y")）：

set.seed(1)

df1 <- copy_to(sc, tibble(id = 1:3, value = rnorm(3)))
df2 <- copy_to(sc, tibble(id = 1:3, value = rnorm(3)))

df1 %>% 
  inner_join(df2, by = c("id"), suffix=c("_x", "_y")) %>% 
  filter(value_y > -0.836)

# # Source:   lazy query [?? x 3]
# # Database: spark_connection
#      id value_x value_y
#   <dbl>   <dbl>   <dbl>
# 1    1.  -0.626   1.60 
# 2    2.   0.184   0.330
# 3    3.  -0.836  -0.820

问题：

Spark允许深层嵌套的结构和复杂的类型。使用点语法（记住window访问权限？）访问struct，并在字段上显示完整路径。这就是为什么，像number.x这样的名字含糊不清。

通常我们使用反引号转义查询

`number.x`

但据我所知，不可能用dplyr API来表达这一点（可能会有一些rlang技巧，但我现在也想不到。）

问题并不是特定于连接。您应避免在名称中使用.。如果由于某种原因，您可以随时使用本机Spark API并解决问题：

df3 <- copy_to(sc, tibble(value.x = rnorm(42)))

df3 %>% 
  spark_dataframe() %>% 
  invoke("withColumnRenamed", "`value.x`", "value_x") %>%
  sdf_register()

# # Source:   table<sparklyr_tmp_61acdbbc592> [?? x 1]
# # Database: spark_connection
#    value_x
#      <dbl>
#  1 -0.0162
#  2  0.944 
#  3  0.821 
#  4  0.594 
#  5  0.919 
#  6  0.782 
#  7  0.0746
#  8 -1.99  
#  9  0.620 
# 10 -0.0561
# # ... with more r

加入sparklyr后过滤时出错

1 个答案: