如何使用kite-dataset分区模式正确导入csv数据集？

Question

我正在使用来自MovieLens的公开可用的csv数据集，我已经为ratings.csv创建了一个分区数据集：

kite-dataset create ratings --schema rating.avsc --partition-by year-month.json --format parquet

下面是我的year-month.json：

[ {
  "name" : "year",
  "source" : "timestamp",
  "type" : "year"
}, {
  "name" : "month",
  "source" : "timestamp",
  "type" : "month"
} ]

下面是我的csv导入命令：

mkite-dataset csv-import ratings.csv ratings

导入完成后，我运行此命令来查看是否实际创建了年和月分区：

hadoop fs -ls /user/hive/warehouse/ratings/

我注意到的是，只创建了一个年份分区，并且在其中创建了一个单月分区：

[cloudera@quickstart ml-20m]$ hadoop fs -ls /user/hive/warehouse/ratings/
Found 3 items
drwxr-xr-x   - cloudera supergroup          0 2016-06-12 18:49 /user/hive/warehouse/ratings/.metadata
drwxr-xr-x   - cloudera supergroup          0 2016-06-12 18:59 /user/hive/warehouse/ratings/.signals
drwxrwxrwx   - cloudera supergroup          0 2016-06-12 18:59 /user/hive/warehouse/ratings/year=1970

[cloudera@quickstart ml-20m]$ hadoop fs -ls /user/hive/warehouse/ratings/year=1970/
Found 1 items
drwxrwxrwx   - cloudera supergroup          0 2016-06-12 18:59 /user/hive/warehouse/ratings/year=1970/month=01

执行这种分区导入的正确方式是什么，这将导致创建所有年份和所有月份的分区？

Answer 1

在时间戳的末尾添加三个零。

使用下面的shell脚本来完成此操作

#!/bin/bash

# add the CSV header to both files
head -n 1 ratings.csv > ratings_1.csv
head -n 1 ratings.csv > ratings_2.csv

# output the first 10,000,000 rows to ratings_1.csv
# this includes the header, and uses tail to remove it
head -n 10000001 ratings.csv | tail -n +2 | awk '{print "000" $1 }' >> ratings_1.csv

    enter code here

# output the rest of the file to ratings_2.csv
# this starts at the line after the ratings_1 file stopped
tail -n +10000002 ratings.csv | awk '{print "000" $1 }' >> ratings_2.csv

就连我也有这个问题，加了3个零就解决了。