4

我正在尝试利用镶木地板的最小/最大索引。我在这里关注问题/答案：Spark Parquet Statistics(min/max) integration

scala> val foo = spark.sql("select id, cast(id as string) text from range(1000)").sort("id") 

scala> foo.printSchema

root
 |-- id: long (nullable = false)
 |-- text: string (nullable = false)

当我查看单个镶木地板文件时，我看不到任何最小值/最大值

> parquet-tools meta part-00000-tid-5174196010762120422-9

5fb2e22-0dfb-4597-bdca-4fb573873959-0-c000.gz.parquet
file:        file:.../part-00000-tid-5174196010762120422-95fb2e22-0dfb-4597-bdca-4fb573873959-0-c000.gz.parquet
creator:     parquet-mr version 1.8.1 (build 4aba4dae7bb0d4edbcf7923ae1339f28fd3f7fcf)
extra:       org.apache.spark.sql.parquet.row.metadata = {"type":"struct","fields":[{"name":"id","type":"long","nullable":false,"metadata":{}},{"name":"text","type":"string","nullable":false,"metadata":{}}]}

file schema: spark_schema
--------------------------------------------------------------------------------
id:          REQUIRED INT64 R:0 D:0
text:        REQUIRED BINARY O:UTF8 R:0 D:0

row group 1: RC:125 TS:1840 OFFSET:4
--------------------------------------------------------------------------------
id:           INT64 GZIP DO:0 FPO:4 SZ:259/1044/4.03 VC:125 ENC:PLAIN,BIT_PACKED
text:         BINARY GZIP DO:0 FPO:263 SZ:263/796/3.03 VC:125 ENC:PLAIN,BIT_PACKED

我试过 .sortWithinPartitions("id") 得到相同的结果。

4

2 回答 2

5

您可以使用 parquet-tools 查看统计信息。在你的情况下，你会跑

parquet-tools dump -d -n part-00000-tid-5174196010762120422-95fb2e22-0dfb-4597-bdca-4fb573873959-0-c000.gz.parquet

截至今天（2017 年 6 月 9 日），带有 Parquet 1.8.1 的 Spark 2.1.1 不会为字符串等二进制列生成统计信息。

于 2017-06-09T18:42:06.447 回答
0

你想要来自https://github.com/apache/arrow/tree/master/cpp的 parquet-reader

祝你编译好运

然后你会得到一些像这样的元数据，你的最小值/最大值

parquet-reader  --only-metadata --json --columns=0,1,2 widey_event_visit_start_datetime_sorted.pq
{
  "FileName": "widey_event_visit_start_datetime_sorted.pq",
  "Version": "0",
  "CreatedBy": "parquet-mr version 1.10.0 (build 031a6654009e3b82020012a18434c582bd74c73a)",
  "TotalRows": "732999",
  "NumberOfRowGroups": "1",
  "NumberOfRealColumns": "88",
  "NumberOfColumns": "88",
  "Columns": [
     { "Id": "0", "Name": "destination", "PhysicalType": "BYTE_ARRAY", "LogicalType": "UTF8" },
     { "Id": "1", "Name": "visit_id", "PhysicalType": "INT32", "LogicalType": "NONE" },
     { "Id": "2", "Name": "visit_start_datetime", "PhysicalType": "INT64", "LogicalType": "NONE" }
  ],
  "RowGroups": [
     {
       "Id": "0",  "TotalBytes": "125009099",  "Rows": "732999",
       "ColumnChunks": [
          {"Id": "0", "Values": "732999", "StatsSet": "True", "Stats": {"NumNulls": "0", "DistinctValues": "0", "Max": "WS_Programmes_TEST", "Min": "GNL_News_TEST" },
           "Compression": "SNAPPY", "Encodings": "PLAIN_DICTIONARY RLE BIT_PACKED ", "UncompressedSize": "166512", "CompressedSize": "134481" },
          {"Id": "1", "Values": "732999", "StatsSet": "True", "Stats": {"NumNulls": "0", "DistinctValues": "0", "Max": "60419931", "Min": "1072" },
           "Compression": "SNAPPY", "Encodings": "PLAIN_DICTIONARY RLE BIT_PACKED ", "UncompressedSize": "860549", "CompressedSize": "786120" },
          {"Id": "2", "Values": "732999", "StatsSet": "True", "Stats": {"NumNulls": "0", "DistinctValues": "0", "Max": "1548892673", "Min": "1548806803" },
           "Compression": "SNAPPY", "Encodings": "PLAIN_DICTIONARY RLE BIT_PACKED ", "UncompressedSize": "5413", "CompressedSize": "3965" }
        ]
     }
  ]
}
于 2019-02-13T16:38:07.337 回答