Hive索引

一、Hive Index具体实现

索引是标准的数据库技术，hive 0.7版本之后支持索引。hive索引采用的不是'one size fites all'的索引实现方式，而是提供插入式接口，并且提供一个具体的索引实现作为参考。Hive的Index接口如下：

/**
 * HiveIndexHandler defines a pluggable interface for adding new index handlers
 * to Hive.
 */
public interface HiveIndexHandler extends Configurable {
  /**
   * Determines whether this handler implements indexes by creating an index
   * table.
   *
   * @return true if index creation implies creation of an index table in Hive;
   *         false if the index representation is not stored in a Hive table
   */
  boolean usesIndexTable();

  /**
   * Requests that the handler validate an index definition and fill in
   * additional information about its stored representation.
   *
   * @param baseTable
   *          the definition of the table being indexed
   *
   * @param index
   *          the definition of the index being created
   *
   * @param indexTable
   *          a partial definition of the index table to be used for storing the
   *          index representation, or null if usesIndexTable() returns false;
   *          the handler can augment the index's storage descriptor (e.g. with
   *          information about input/output format) and/or the index table's
   *          definition (typically with additional columns containing the index
   *          representation, e.g. pointers into HDFS).
   *
   * @throws HiveException if the index definition is invalid with respect to
   *         either the base table or the supplied index table definition
   */
  void analyzeIndexDefinition(
      org.apache.hadoop.hive.metastore.api.Table baseTable,
      org.apache.hadoop.hive.metastore.api.Index index,
      org.apache.hadoop.hive.metastore.api.Table indexTable)
      throws HiveException;

  /**
   * Requests that the handler generate a plan for building the index; the plan
   * should read the base table and write out the index representation.
   *
   * @param baseTbl
   *          the definition of the table being indexed
   *
   * @param index
   *          the definition of the index
   *
   * @param baseTblPartitions
   *          list of base table partitions with each element mirrors to the
   *          corresponding one in indexTblPartitions
   *
   * @param indexTbl
   *          the definition of the index table, or null if usesIndexTable()
   *          returns null
   *
   * @param inputs
   *          inputs for hooks, supplemental outputs going
   *          along with the return value
   *
   * @param outputs
   *          outputs for hooks, supplemental outputs going
   *          along with the return value
   *
   * @return list of tasks to be executed in parallel for building the index
   *
   * @throws HiveException if plan generation fails
   */
  List<Task<?>> generateIndexBuildTaskList(
      org.apache.hadoop.hive.ql.metadata.Table baseTbl,
      org.apache.hadoop.hive.metastore.api.Index index,
      List<Partition> indexTblPartitions, List<Partition> baseTblPartitions,
      org.apache.hadoop.hive.ql.metadata.Table indexTbl,
      Set<ReadEntity> inputs, Set<WriteEntity> outputs)
      throws HiveException;

  /**
   * Generate the list of tasks required to run an index optimized sub-query for the
   * given predicate, using the given indexes. If multiple indexes are
   * provided, it is up to the handler whether to use none, one, some or all of
   * them. The supplied predicate may reference any of the columns from any of
   * the indexes. If the handler decides to use more than one index, it is
   * responsible for generating tasks to combine their search results
   * (e.g. performing a JOIN on the result).
   * @param indexes
   * @param predicate
   * @param pctx
   * @param queryContext contains results, such as query tasks and input configuration
   */
  void generateIndexQuery(List<Index> indexes, ExprNodeDesc predicate,
    ParseContext pctx, HiveIndexQueryContext queryContext);

  /**
   * Check the size of an input query to make sure it fits within the bounds
   *
   * @param inputSize size (in bytes) of the query in question
   * @param conf
   * @return true if query is within the bounds
   */
  boolean checkQuerySize(long inputSize, HiveConf conf);
}

创建索引的时候，Hive首先调用接口的usesIndexTable方法，判断索引是否是已Hive Table的方式存储（默认的实现是存储在Hive中的）。然后调用analyzeIndexDefinition分析索引创建语句是否合法，如果没有问题将在元数据标IDXS中添加索引表，否则抛出异常。如果索引创建语句中使用with deferred rebuild，在执行alter index xxx_index on xxx rebuild时将调用generateIndexBuildTaskList获取Index的MapReduce，并执行为索引填充数据。

 

测试索引例子：

 

1.创建测试数据

 

#! /bin/bash  
#generating 350M raw data.  
i=0  
while [ $i -ne 1000000 ]  
do  
        echo -e "$i\tA decade ago, many were predicting that Cooke, a New York City prodigy, would become a basketball shoe pitchman and would flaunt his wares and skills at All-Star weekends like the recent aerial show in Orlando, Fla. There was a time, however fleeting, when he was more heralded, or perhaps merely hyped, than any other high school player in America."  
        i=$(($i+1))  
done

 

 

2.创建测试表1

 

create table table01( id int, name string)  ROW FORMAT DELIMITED  FIELDS TERMINATED BY '\t'; 

 

load data local inpath '/data/tmp/huzhirong/dual.txt' overwrite into table table01;

 

 

3.创建测试表2，并从表1中select数据

 

create table table02 as select id,name as text from table01;

 table02在hdfs的数据

hive> dfs -ls /user/hive/warehouse/table02; 
Found 5 items
-rw-r--r--   3 hadoop supergroup   88453176 2013-04-26 20:56 /user/hive/warehouse/table02/000000_0
-rw-r--r--   3 hadoop supergroup   67108860 2013-04-26 20:56 /user/hive/warehouse/table02/000001_0
-rw-r--r--   3 hadoop supergroup   67109134 2013-04-26 20:56 /user/hive/warehouse/table02/000002_0
-rw-r--r--   3 hadoop supergroup   67108860 2013-04-26 20:56 /user/hive/warehouse/table02/000003_0
-rw-r--r--   3 hadoop supergroup   67108860 2013-04-26 20:56 /user/hive/warehouse/table02/000004_0

 测试查询：

select * from table02 where id=500000;

 耗时：

OK
500000  A decade ago, many were predicting that Cooke, a New York City prodigy, would become a basketball shoe pitchman and would flaunt his wares and skills at All-Star weekends like the recent aerial show in Orlando, Fla. There was a time, however fleeting, when he was more heralded, or perhaps merely hyped, than any other high school player in America.
Time taken: 35.022 seconds

 

4.创建索引

 

create index table02_index on table table02(id) as 'org.apache.hadoop.hive.ql.index.compact.CompactIndexHandler' with deferred rebuild;

alter index table02_index on table02 rebuild;

Loading data to table default.default__table02_table02_index__
Moved to trash: hdfs://namenode.hadoop.game.yy.com/user/hive/warehouse/default__table02_table02_index__
Table default.default__table02_table02_index__ stats: [num_partitions: 0, num_files: 1, num_rows: 0, total_size: 87733114, raw_data_size: 0]
MapReduce Jobs Launched: 
Job 0: Map: 3  Reduce: 1   Cumulative CPU: 51.28 sec   HDFS Read: 357021261 HDFS Write: 87733114 SUCCESS
Total MapReduce CPU Time Spent: 51 seconds 280 msec
OK
Time taken: 65.6 seconds

hive> dfs -ls /user/hive/warehouse/default__table02_table02_index__;
Found 1 items
-rw-r--r--   3 hadoop supergroup   87733114 2013-04-26 21:04 /user/hive/warehouse/default__table02_table02_index__/000000_0

 

可以看到索引内存储的数据：

hive> select * from default__table02_table02_index__ limit 3;
OK
0       hdfs://namenode.hadoop.game.yy.com/user/hive/warehouse/table02/000002_0 [0]
1       hdfs://namenode.hadoop.game.yy.com/user/hive/warehouse/table02/000002_0 [352]
2       hdfs://namenode.hadoop.game.yy.com/user/hive/warehouse/table02/000002_0 [704]

 应该是{值，HDFS文件位置，偏移量的数组（可能有多个）}

自定义索引文件：

insert overwrite directory "/tmp/table02_index_data" select `_bucketname`, `_offsets` from   default__table02_table02_index__ where id =500000;

 

查询table02数据

直接查询：

hive> select * from table02 where id =500000;
Total MapReduce jobs = 1
Launching Job 1 out of 1
Number of reduce tasks is set to 0 since there's no reduce operator
Starting Job = job_201301301559_29049, Tracking URL = http://namenode.hadoop.game.yy.com:50030/jobdetails.jsp?jobid=job_201301301559_29049
Kill Command = /home/hadoop/hadoop-1.0.3/libexec/../bin/hadoop job  -Dmapred.job.tracker=namenode.hadoop.game.yy.com:8021 -kill job_201301301559_29049
Hadoop job information for Stage-1: number of mappers: 6; number of reducers: 0
2013-04-26 22:34:20,755 Stage-1 map = 0%,  reduce = 0%
2013-04-26 22:34:26,797 Stage-1 map = 17%,  reduce = 0%, Cumulative CPU 2.23 sec
2013-04-26 22:34:27,812 Stage-1 map = 17%,  reduce = 0%, Cumulative CPU 2.23 sec
2013-04-26 22:34:28,859 Stage-1 map = 17%,  reduce = 0%, Cumulative CPU 2.23 sec
2013-04-26 22:34:29,871 Stage-1 map = 17%,  reduce = 0%, Cumulative CPU 2.23 sec
2013-04-26 22:34:30,874 Stage-1 map = 17%,  reduce = 0%, Cumulative CPU 2.23 sec
2013-04-26 22:34:31,877 Stage-1 map = 17%,  reduce = 0%, Cumulative CPU 2.23 sec
2013-04-26 22:34:32,879 Stage-1 map = 83%,  reduce = 0%, Cumulative CPU 11.58 sec
2013-04-26 22:34:33,882 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 12.99 sec
2013-04-26 22:34:34,884 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 12.99 sec
2013-04-26 22:34:35,887 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 12.99 sec
2013-04-26 22:34:36,890 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 12.99 sec
2013-04-26 22:34:37,893 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 12.99 sec
2013-04-26 22:34:38,895 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 12.99 sec
2013-04-26 22:34:39,898 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 12.99 sec
MapReduce Total cumulative CPU time: 12 seconds 990 msec
Ended Job = job_201301301559_29049
MapReduce Jobs Launched: 
Job 0: Map: 6   Cumulative CPU: 12.99 sec   HDFS Read: 357021325 HDFS Write: 357 SUCCESS
Total MapReduce CPU Time Spent: 12 seconds 990 msec
OK
500000  A decade ago, many were predicting that Cooke, a New York City prodigy, would become a basketball shoe pitchman and would flaunt his wares and skills at All-Star weekends like the recent aerial show in Orlando, Fla. There was a time, however fleeting, when he was more heralded, or perhaps merely hyped, than any other high school player in America.
Time taken: 33.189 seconds

 指定索引：

hive> set hive.index.compact.file=/tmp/table02_index_data;                                        
hive> set hive.optimize.index.filter=false; 
hive> set hive.input.format=org.apache.hadoop.hive.ql.index.compact.HiveCompactIndexInputFormat;
hive> select * from table02 where id =500000; 
Total MapReduce jobs = 1
Launching Job 1 out of 1
Number of reduce tasks is set to 0 since there's no reduce operator
Starting Job = job_201301301559_29051, Tracking URL = http://namenode.hadoop.game.yy.com:50030/jobdetails.jsp?jobid=job_201301301559_29051
Kill Command = /home/hadoop/hadoop-1.0.3/libexec/../bin/hadoop job  -Dmapred.job.tracker=namenode.hadoop.game.yy.com:8021 -kill job_201301301559_29051
Hadoop job information for Stage-1: number of mappers: 1; number of reducers: 0
2013-04-26 22:40:06,793 Stage-1 map = 0%,  reduce = 0%
2013-04-26 22:40:12,803 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.69 sec
2013-04-26 22:40:13,806 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.69 sec
2013-04-26 22:40:14,808 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.69 sec
2013-04-26 22:40:15,811 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.69 sec
2013-04-26 22:40:16,813 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.69 sec
2013-04-26 22:40:17,815 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.69 sec
2013-04-26 22:40:18,818 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 1.69 sec
MapReduce Total cumulative CPU time: 1 seconds 690 msec
Ended Job = job_201301301559_29051
MapReduce Jobs Launched: 
Job 0: Map: 1   Cumulative CPU: 1.69 sec   HDFS Read: 33554658 HDFS Write: 357 SUCCESS
Total MapReduce CPU Time Spent: 1 seconds 690 msec
OK
500000  A decade ago, many were predicting that Cooke, a New York City prodigy, would become a basketball shoe pitchman and would flaunt his wares and skills at All-Star weekends like the recent aerial show in Orlando, Fla. There was a time, however fleeting, when he was more heralded, or perhaps merely hyped, than any other high school player in America.
Time taken: 26.776 seconds

 看Map数：变成1个了，不过这里是手工插入id=500000的索引。

总结：索引表的基本包含几列：1. 源表的索引列；2. _bucketname hdfs中文件地址 3. 索引列在hdfs文件中的偏移量。原理是通过记录索引列在HDFS中的偏移量，精准获取数据，避免全表扫描。