Hbase 多版本

Bertha 。 2023-07-25 12:49 7阅读 0赞

hbase支持多版本存储，其一条数据的多版本是以timestamp来标识的。

#### 设置多版本 ####

1.  .新建测试表

hbase(main):032:0* create 'tmp_mutilversion', {
        NAME => 'f', VERSIONS => 5}
    0 row(s) in 2.2860 seconds
    
    hbase(main):006:0> desc 'tmp_mutilversion'
    Table tmp_mutilversion is ENABLED                                                                                                                                                                                                                                             
    tmp_mutilversion                                                                                                                                                                                                                                                              
    COLUMN FAMILIES DESCRIPTION                                                                                                                                                                                                                                                   
    {
        NAME => 'f', BLOOMFILTER => 'ROW', VERSIONS => '5', IN_MEMORY => 'false', KEEP_DELETED_CELLS => 'FALSE', DATA_BLOCK_ENCODING => 'NONE', TTL => 'FOREVER', COMPRESSION => 'NONE', MIN_VERSIONS => '0', BLOCKCACHE => 'true', BLOCKSIZE => '65536', REPLICATION_SCOPE => '0'}  
    1 row(s) in 0.0290 seconds

1.  put 测试数据

put 'tmp_mutilversion','rk1111','f:q1','v10'
    put 'tmp_mutilversion','rk1112','f:q1','v20'
    put 'tmp_mutilversion','rk1113','f:q1','v30'
    put 'tmp_mutilversion','rk1114','f:q1','v40'
    put 'tmp_mutilversion','rk1115','f:q1','v50'
    put 'tmp_mutilversion','rk1116','f:q1','v60'
    put 'tmp_mutilversion','rk1112','f:q1','v21'
    put 'tmp_mutilversion','rk1113','f:q1','v31'
    put 'tmp_mutilversion','rk1114','f:q1','v41'
    put 'tmp_mutilversion','rk1117','f:q1','v70'
    put 'tmp_mutilversion','rk1118','f:q1','v80'
    put 'tmp_mutilversion','rk1119','f:q1','v90'
    put 'tmp_mutilversion','rk1113','f:q1','v32'
    put 'tmp_mutilversion','rk1114','f:q1','v42'

1.  scan表看结果，直接看是看不到多个版本的，直接查看到的是最新的那条数据

hbase(main):047:0> scan 'tmp_mutilversion'
    ROW                                                                  COLUMN+CELL                                                                                                                                                                                              
     rk1111                                                              column=f:q1, timestamp=1586916990543, value=v10                                                                                                                                                          
     rk1112                                                              column=f:q1, timestamp=1586916990644, value=v21                                                                                                                                                          
     rk1113                                                              column=f:q1, timestamp=1586916990718, value=v32                                                                                                                                                          
     rk1114                                                              column=f:q1, timestamp=1586916991906, value=v42                                                                                                                                                          
     rk1115                                                              column=f:q1, timestamp=1586916990615, value=v50                                                                                                                                                          
     rk1116                                                              column=f:q1, timestamp=1586916990630, value=v60                                                                                                                                                          
     rk1117                                                              column=f:q1, timestamp=1586916990683, value=v70                                                                                                                                                          
     rk1118                                                              column=f:q1, timestamp=1586916990694, value=v80                                                                                                                                                          
     rk1119                                                              column=f:q1, timestamp=1586916990707, value=v90                                                                                                                                                          
    9 row(s) in 0.0350 seconds

1.  查看多版本

hbase(main):001:0> get 'tmp_mutilversion','rk1114',{
        COLUMN => 'f:q1', VERSIONS => 2}
    COLUMN                                                               CELL                                                                                                                                                                                                     
     f:q1                                                                timestamp=1586916991906, value=v42                                                                                                                                                                       
     f:q1                                                                timestamp=1586916990671, value=v41                                                                                                                                                                       
    1 row(s) in 0.2300 seconds
    
    hbase(main):002:0> get 'tmp_mutilversion','rk1114',{
        COLUMN => 'f:q1', VERSIONS => 5}
    COLUMN                                                               CELL                                                                                                                                                                                                     
     f:q1                                                                timestamp=1586916991906, value=v42                                                                                                                                                                       
     f:q1                                                                timestamp=1586916990671, value=v41                                                                                                                                                                       
     f:q1                                                                timestamp=1586916990598, value=v40                                                                                                                                                                       
    1 row(s) in 0.0090 seconds

1.  查看wal日志

Sequence=4 , region=0b085a4a81c0fa739c4117970f75ac7c at write timestamp=Wed Apr 15 10:16:30 CST 2020
    row=rk1111, column=f:q1
    Sequence=5 , region=0b085a4a81c0fa739c4117970f75ac7c at write timestamp=Wed Apr 15 10:16:30 CST 2020
    row=rk1112, column=f:q1
    Sequence=6 , region=0b085a4a81c0fa739c4117970f75ac7c at write timestamp=Wed Apr 15 10:16:30 CST 2020
    row=rk1113, column=f:q1
    Sequence=7 , region=0b085a4a81c0fa739c4117970f75ac7c at write timestamp=Wed Apr 15 10:16:30 CST 2020
    row=rk1114, column=f:q1
    Sequence=8 , region=0b085a4a81c0fa739c4117970f75ac7c at write timestamp=Wed Apr 15 10:16:30 CST 2020
    row=rk1115, column=f:q1
    Sequence=9 , region=0b085a4a81c0fa739c4117970f75ac7c at write timestamp=Wed Apr 15 10:16:30 CST 2020
    row=rk1116, column=f:q1
    Sequence=10 , region=0b085a4a81c0fa739c4117970f75ac7c at write timestamp=Wed Apr 15 10:16:30 CST 2020
    row=rk1112, column=f:q1
    Sequence=11 , region=0b085a4a81c0fa739c4117970f75ac7c at write timestamp=Wed Apr 15 10:16:30 CST 2020
    row=rk1113, column=f:q1
    Sequence=12 , region=0b085a4a81c0fa739c4117970f75ac7c at write timestamp=Wed Apr 15 10:16:30 CST 2020
    row=rk1114, column=f:q1
    Sequence=13 , region=0b085a4a81c0fa739c4117970f75ac7c at write timestamp=Wed Apr 15 10:16:30 CST 2020
    row=rk1117, column=f:q1
    Sequence=14 , region=0b085a4a81c0fa739c4117970f75ac7c at write timestamp=Wed Apr 15 10:16:30 CST 2020
    row=rk1118, column=f:q1
    Sequence=15 , region=0b085a4a81c0fa739c4117970f75ac7c at write timestamp=Wed Apr 15 10:16:30 CST 2020
    row=rk1119, column=f:q1
    Sequence=16 , region=0b085a4a81c0fa739c4117970f75ac7c at write timestamp=Wed Apr 15 10:16:30 CST 2020
    row=rk1113, column=f:q1
    Sequence=17 , region=0b085a4a81c0fa739c4117970f75ac7c at write timestamp=Wed Apr 15 10:16:31 CST 2020
    row=rk1114, column=f:q1

1.  查看hfile文件，可能需要手动flush一下这个表

K: rk1111/f:q1/1586916990543/Put/vlen=3/seqid=4 V: v10
    K: rk1112/f:q1/1586916990644/Put/vlen=3/seqid=10 V: v21
    K: rk1112/f:q1/1586916990566/Put/vlen=3/seqid=5 V: v20
    K: rk1113/f:q1/1586916990718/Put/vlen=3/seqid=16 V: v32
    K: rk1113/f:q1/1586916990659/Put/vlen=3/seqid=11 V: v31
    K: rk1113/f:q1/1586916990583/Put/vlen=3/seqid=6 V: v30
    K: rk1114/f:q1/1586916991906/Put/vlen=3/seqid=17 V: v42
    K: rk1114/f:q1/1586916990671/Put/vlen=3/seqid=12 V: v41
    K: rk1114/f:q1/1586916990598/Put/vlen=3/seqid=7 V: v40
    K: rk1115/f:q1/1586916990615/Put/vlen=3/seqid=8 V: v50
    K: rk1116/f:q1/1586916990630/Put/vlen=3/seqid=9 V: v60
    K: rk1117/f:q1/1586916990683/Put/vlen=3/seqid=13 V: v70
    K: rk1118/f:q1/1586916990694/Put/vlen=3/seqid=14 V: v80
    K: rk1119/f:q1/1586916990707/Put/vlen=3/seqid=15 V: v90
    Scanned kv count -> 14

从上述测试来看，设置了多版本\{5\}以后，不同对于同一条数据不同timestamp的数据，是存放在同一个hfile文件中。

那么，假设建表时默认只有一个version，那么同一个rk还能写入多条数据吗？

1.  建测试表

hbase(main):009:0* create 'tmp_singleversion', {
        NAME => 'f', VERSIONS => 1}
    0 row(s) in 2.3990 seconds
    
    => Hbase::Table - tmp_singleversion
    hbase(main):010:0> desc 'tmp_singleversion'
    Table tmp_singleversion is ENABLED                                                                                                                                                                                                                                            
    tmp_singleversion                                                                                                                                                                                                                                                             
    COLUMN FAMILIES DESCRIPTION                                                                                                                                                                                                                                                   
    {
        NAME => 'f', BLOOMFILTER => 'ROW', VERSIONS => '1', IN_MEMORY => 'false', KEEP_DELETED_CELLS => 'FALSE', DATA_BLOCK_ENCODING => 'NONE', TTL => 'FOREVER', COMPRESSION => 'NONE', MIN_VERSIONS => '0', BLOCKCACHE => 'true', BLOCKSIZE => '65536', REPLICATION_SCOPE => '0'}  
    1 row(s) in 0.0210 seconds

1.  输入同样的数据

put 'tmp_singleversion','rk1111','f:q1','v10'
    put 'tmp_singleversion','rk1112','f:q1','v20'
    put 'tmp_singleversion','rk1113','f:q1','v30'
    put 'tmp_singleversion','rk1114','f:q1','v40'
    put 'tmp_singleversion','rk1115','f:q1','v50'
    put 'tmp_singleversion','rk1116','f:q1','v60'
    put 'tmp_singleversion','rk1112','f:q1','v21'
    put 'tmp_singleversion','rk1113','f:q1','v31'
    put 'tmp_singleversion','rk1114','f:q1','v41'
    put 'tmp_singleversion','rk1117','f:q1','v70'
    put 'tmp_singleversion','rk1118','f:q1','v80'
    put 'tmp_singleversion','rk1119','f:q1','v90'
    put 'tmp_singleversion','rk1113','f:q1','v32'
    put 'tmp_singleversion','rk1114','f:q1','v42'

1.  同样scan

hbase(main):025:0> scan 'tmp_singleversion'
    ROW                                                                  COLUMN+CELL                                                                                                                                                                                              
     rk1111                                                              column=f:q1, timestamp=1586918408170, value=v10                                                                                                                                                          
     rk1112                                                              column=f:q1, timestamp=1586918408288, value=v21                                                                                                                                                          
     rk1113                                                              column=f:q1, timestamp=1586918408533, value=v32                                                                                                                                                          
     rk1114                                                              column=f:q1, timestamp=1586918409641, value=v42                                                                                                                                                          
     rk1115                                                              column=f:q1, timestamp=1586918408257, value=v50                                                                                                                                                          
     rk1116                                                              column=f:q1, timestamp=1586918408271, value=v60                                                                                                                                                          
     rk1117                                                              column=f:q1, timestamp=1586918408350, value=v70                                                                                                                                                          
     rk1118                                                              column=f:q1, timestamp=1586918408367, value=v80                                                                                                                                                          
     rk1119                                                              column=f:q1, timestamp=1586918408505, value=v90                                                                                                                                                          
    9 row(s) in 0.0150 seconds

1.  查看多版本看看

hbase(main):028:0> get 'tmp_singleversion','rk1114',{
        COLUMN => 'f:q1', VERSIONS => 5}
    COLUMN                                                               CELL                                                                                                                                                                                                     
     f:q1                                                                timestamp=1586918409641, value=v42                                                                                                                                                                       
    1 row(s) in 0.0040 seconds
    
    hbase(main):029:0> get 'tmp_singleversion','rk1114',{
        COLUMN => 'f:q1', VERSIONS => 2}
    COLUMN                                                               CELL                                                                                                                                                                                                     
     f:q1                                                                timestamp=1586918409641, value=v42                                                                                                                                                                       
    1 row(s) in 0.0040 seconds

1.  wal

Sequence=4 , region=76c57383a915f168383bff0c26d59251 at write timestamp=Wed Apr 15 10:40:08 CST 2020
    row=rk1111, column=f:q1
    Sequence=5 , region=76c57383a915f168383bff0c26d59251 at write timestamp=Wed Apr 15 10:40:08 CST 2020
    row=rk1112, column=f:q1
    Sequence=6 , region=76c57383a915f168383bff0c26d59251 at write timestamp=Wed Apr 15 10:40:08 CST 2020
    row=rk1113, column=f:q1
    Sequence=7 , region=76c57383a915f168383bff0c26d59251 at write timestamp=Wed Apr 15 10:40:08 CST 2020
    row=rk1114, column=f:q1
    Sequence=8 , region=76c57383a915f168383bff0c26d59251 at write timestamp=Wed Apr 15 10:40:08 CST 2020
    row=rk1115, column=f:q1
    Sequence=9 , region=76c57383a915f168383bff0c26d59251 at write timestamp=Wed Apr 15 10:40:08 CST 2020
    row=rk1116, column=f:q1
    Sequence=10 , region=76c57383a915f168383bff0c26d59251 at write timestamp=Wed Apr 15 10:40:08 CST 2020
    row=rk1112, column=f:q1
    Sequence=11 , region=76c57383a915f168383bff0c26d59251 at write timestamp=Wed Apr 15 10:40:08 CST 2020
    row=rk1113, column=f:q1
    Sequence=12 , region=76c57383a915f168383bff0c26d59251 at write timestamp=Wed Apr 15 10:40:08 CST 2020
    row=rk1114, column=f:q1
    Sequence=13 , region=76c57383a915f168383bff0c26d59251 at write timestamp=Wed Apr 15 10:40:08 CST 2020
    row=rk1117, column=f:q1
    Sequence=14 , region=76c57383a915f168383bff0c26d59251 at write timestamp=Wed Apr 15 10:40:08 CST 2020
    row=rk1118, column=f:q1
    Sequence=15 , region=76c57383a915f168383bff0c26d59251 at write timestamp=Wed Apr 15 10:40:08 CST 2020
    row=rk1119, column=f:q1
    Sequence=16 , region=76c57383a915f168383bff0c26d59251 at write timestamp=Wed Apr 15 10:40:08 CST 2020
    row=rk1113, column=f:q1
    Sequence=17 , region=76c57383a915f168383bff0c26d59251 at write timestamp=Wed Apr 15 10:40:09 CST 2020
    row=rk1114, column=f:q1

6.hfile

K: rk1111/f:q1/1586918408170/Put/vlen=3/seqid=4 V: v10
    K: rk1112/f:q1/1586918408288/Put/vlen=3/seqid=10 V: v21
    K: rk1113/f:q1/1586918408533/Put/vlen=3/seqid=16 V: v32
    K: rk1114/f:q1/1586918409641/Put/vlen=3/seqid=17 V: v42
    K: rk1115/f:q1/1586918408257/Put/vlen=3/seqid=8 V: v50
    K: rk1116/f:q1/1586918408271/Put/vlen=3/seqid=9 V: v60
    K: rk1117/f:q1/1586918408350/Put/vlen=3/seqid=13 V: v70
    K: rk1118/f:q1/1586918408367/Put/vlen=3/seqid=14 V: v80
    K: rk1119/f:q1/1586918408505/Put/vlen=3/seqid=15 V: v90
    Scanned kv count -> 9

所以多版本这部分数据发生了什么？

查看日志，可以看出最后正式提交后就是9条数据了。所以wal中写入后，在memstore到在flush的过程中，就对多余的版本做出处理了，生成的hfile就是这9条数据。具体还有待研究源码拉。

2020-04-15 10:50:05,940 DEBUG org.apache.hadoop.hbase.regionserver.HRegionFileSystem: Committing store file hdfs://hbase/data/default/tmp_singleversion/76c57383a915f168383bff0c26d59251/.tmp/360a0fbdaa5a46c4bf94da028c6a0679 as hdfs://hbase/data/default/tmp_singleversion/76c57383a915f168383bff0c26d59251/f/360a0fbdaa5a46c4bf94da028c6a0679
    2020-04-15 10:50:05,962 INFO org.apache.hadoop.hbase.regionserver.HStore: Added hdfs://hbase/data/default/tmp_singleversion/76c57383a915f168383bff0c26d59251/f/360a0fbdaa5a46c4bf94da028c6a0679, entries=9, sequenceid=19, filesize=5.1 K

另一种情况，也是实际使用的情况

1.  建测试表

hbase(main):037:0* create 'tmp_single', {
        NAME => 'f', VERSIONS => 1}
    0 row(s) in 2.3840 seconds
    
    => Hbase::Table - tmp_single
    hbase(main):038:0> desc 'tmp_single'
    Table tmp_single is ENABLED                                                                                           
    tmp_single                                                                                                            
    COLUMN FAMILIES DESCRIPTION                                                                                           
    {
        NAME => 'f', BLOOMFILTER => 'ROW', VERSIONS => '1', IN_MEMORY => 'false', KEEP_DELETED_CELLS => 'FALSE', DATA_BLOCK_E
    NCODING => 'NONE', TTL => 'FOREVER', COMPRESSION => 'NONE', MIN_VERSIONS => '0', BLOCKCACHE => 'true', BLOCKSIZE => '6
    5536', REPLICATION_SCOPE => '0'}                                                                                      
    1 row(s) in 0.0370 seconds

1.  输入第一部分数据

put 'tmp_single','rk1111','f:q1','v10'
    put 'tmp_single','rk1112','f:q1','v20'
    put 'tmp_single','rk1113','f:q1','v30'
    put 'tmp_single','rk1114','f:q1','v40'
    put 'tmp_single','rk1115','f:q1','v50'
    put 'tmp_single','rk1116','f:q1','v60'
    put 'tmp_single','rk1117','f:q1','v70'
    put 'tmp_single','rk1118','f:q1','v80'
    put 'tmp_single','rk1119','f:q1','v90'

1.  flush以生成hfile

K: rk1111/f:q1/1586920978622/Put/vlen=3/seqid=4 V: v10
    K: rk1112/f:q1/1586920978659/Put/vlen=3/seqid=5 V: v20
    K: rk1113/f:q1/1586920978674/Put/vlen=3/seqid=6 V: v30
    K: rk1114/f:q1/1586920978688/Put/vlen=3/seqid=7 V: v40
    K: rk1115/f:q1/1586920978701/Put/vlen=3/seqid=8 V: v50
    K: rk1116/f:q1/1586920978717/Put/vlen=3/seqid=9 V: v60
    K: rk1117/f:q1/1586920978733/Put/vlen=3/seqid=10 V: v70
    K: rk1118/f:q1/1586920978748/Put/vlen=3/seqid=11 V: v80
    K: rk1119/f:q1/1586920980686/Put/vlen=3/seqid=12 V: v90
    Scanned kv count -> 9

1.  再输入多版本部分的数据

put 'tmp_single','rk1112','f:q1','v21'
    put 'tmp_single','rk1113','f:q1','v31'
    put 'tmp_single','rk1114','f:q1','v41'
    put 'tmp_single','rk1113','f:q1','v32'
    put 'tmp_single','rk1114','f:q1','v42'

5.查看表

hbase(main):065:0> scan 'tmp_single'
    ROW                                                                  COLUMN+CELL                                                                                                                                                                                              
     rk1111                                                              column=f:q1, timestamp=1586920978622, value=v10                                                                                                                                                          
     rk1112                                                              column=f:q1, timestamp=1586921446825, value=v21                                                                                                                                                          
     rk1113                                                              column=f:q1, timestamp=1586921446867, value=v32                                                                                                                                                          
     rk1114                                                              column=f:q1, timestamp=1586921448047, value=v42                                                                                                                                                          
     rk1115                                                              column=f:q1, timestamp=1586920978701, value=v50                                                                                                                                                          
     rk1116                                                              column=f:q1, timestamp=1586920978717, value=v60                                                                                                                                                          
     rk1117                                                              column=f:q1, timestamp=1586920978733, value=v70                                                                                                                                                          
     rk1118                                                              column=f:q1, timestamp=1586920978748, value=v80                                                                                                                                                          
     rk1119                                                              column=f:q1, timestamp=1586920980686, value=v90                                                                                                                                                          
    9 row(s) in 0.0420 seconds

6.flush 会生成一个新的hfile，查看

K: rk1112/f:q1/1586921446825/Put/vlen=3/seqid=17 V: v21
    K: rk1113/f:q1/1586921446867/Put/vlen=3/seqid=20 V: v32
    K: rk1114/f:q1/1586921448047/Put/vlen=3/seqid=21 V: v42
    Scanned kv count -> 3

那么应该是这样，对于一个表的一个rk，同一次flush中，会对超出版本数的数据做出处理。不同次flush写入同一rk数据的多版本，会在当时生成的hfile中保存；多个storfile中的多版本数据，就会在majorcompact后进行合并和丢弃，只留下最新的一条。