hdfs使用手册balancer（数据平衡）命令详解

2019/1/21 星期一

2.3.1. balancer
运行一个集群平衡工具。管理员可以通过按Ctrl+C键停止再平衡过程。
数据平衡的需求
HDFS数据不平衡的原因
1.某个DataNode机器内硬盘存储达到饱和值。
2.集群内新增、删除节点。
数据不平衡的影响
1.Map任务可能会被分配给没有存储数据的机器，结果是不能实现本地计算，最终会导致网络
带宽的消耗。
2.当一些数据节点数据完全满载时，新的数据块只会被存放在有空余数据的节点机器上，造成了并行读取的可能性。
数据平衡过程的要求
1.数据平衡不会导致数据块减少、数据块备份丢失。
2.管理员可以中止数据平衡进程。
3.每次数据块移动的大小应该是可控的，这样可以放置阻塞网络。
4.namenode不会因为数据平衡服务而导致过于繁忙。
数据自动平衡原理
由于使用了平衡算法，导致数据平衡是一个迭代的、周而复始的过程。每一次迭代的最终目的
是让高负载的机器能够降低数据负载，所以数据平衡会最大程度上地使用网络带宽。

数据平衡流程交互图

步骤分析如下：

1、数据均衡服务（Rebalancing Server）首先要求 NameNode 生成 DataNode 数据分布分析报告,获取每个DataNode磁盘使用情况2、Rebalancing Server汇总需要移动的数据分布情况，计算具体数据块迁移路线图。数据块迁移路线图，确保网络内最短路径3、开始数据块迁移任务，Proxy Source Data Node复制一块需要移动数据块4、将复制的数据块复制到目标DataNode上5、删除原始数据块6、目标DataNode向Proxy Source Data Node确认该数据块迁移完成7、Proxy Source Data Node向Rebalancing Server确认本次数据块迁移完成。然后继续执行这个过程，直至集群达到数据均衡标准

实际操作
1.切换到hdfs用户

[root@hadoop-master ~]# su - hdfs2.查看当前的数据分布情况[hdfs@hadoop-master ~]$ hdfs dfsadmin -report > /tmp/bq[hdfs@hadoop-master ~]$ cat /tmp/bq Configured Capacity: 273287419086 (254.52 GB)Present Capacity: 209643254756 (195.25 GB)DFS Remaining: 199579415524 (185.87 GB)DFS Used: 10063839232 (9.37 GB)DFS Used%: 4.80%Under replicated blocks: 0Blocks with corrupt replicas: 0Missing blocks: 0Missing blocks (with replication factor 1): 0配置容量：273287419086（254.52 GB）现有容量：209643254756（195.25 GB）DFS剩余：199579415524（185.87 GB）使用的DFS：10063839232（9.37 GB）使用DFS％：4.80％在复制块下：0具有损坏副本的块：0缺少块：0缺少块（复制因子1）：0-------------------------------------------------Live datanodes (3):  实时数据节点（3）：Name: 192.168.0.117:50010 (hadoop-node01)Hostname: hadoop-node01Rack: /defaultDecommission Status : NormalConfigured Capacity: 91095806362 (84.84 GB)DFS Used: 3354603520 (3.12 GB)Non DFS Used: 12246245786 (11.41 GB)DFS Remaining: 69809631564 (65.02 GB)DFS Used%: 3.68%DFS Remaining%: 76.63%Configured Cache Capacity: 4294967296 (4 GB)Cache Used: 0 (0 B)Cache Remaining: 4294967296 (4 GB)Cache Used%: 0.00%Cache Remaining%: 100.00%Xceivers: 10Last contact: Mon Jan 21 10:45:24 CST 2019名称：192.168.0.117：50010（hadoop-node01）主机名：hadoop-node01机架：/默认退役状态：正常配置容量：91095806362（84.84 GB）使用的DFS：3354603520（3.12 GB）非DFS使用：12246245786（11.41 GB）剩余DFS：69809631564（65.02 GB）使用DFS％：3.68％DFS剩余％：76.63％配置的缓存容量：4294967296（4 GB）使用的缓存：0（0 B）剩余高速缓存：4294967296（4 GB）使用的缓存％：0.00％剩余高速缓存％：100.00％Xceivers：10最后联系人：Mon Jan 21 10:45:24 CST 2019Name: 192.168.0.118:50010 (hadoop-master)Hostname: hadoop-masterRack: /defaultDecommission Status : NormalConfigured Capacity: 91095806362 (84.84 GB)DFS Used: 3354632192 (3.12 GB)Non DFS Used: 29517959578 (27.49 GB)DFS Remaining: 52537889100 (48.93 GB)DFS Used%: 3.68%DFS Remaining%: 57.67%Configured Cache Capacity: 4294967296 (4 GB)Cache Used: 0 (0 B)Cache Remaining: 4294967296 (4 GB)Cache Used%: 0.00%Cache Remaining%: 100.00%Xceivers: 10Last contact: Mon Jan 21 10:45:24 CST 2019Name: 192.168.0.121:50010 (hadoop-node02)Hostname: hadoop-node02Rack: /defaultDecommission Status : NormalConfigured Capacity: 91095806362 (84.84 GB)DFS Used: 3354603520 (3.12 GB)Non DFS Used: 4823982490 (4.49 GB)DFS Remaining: 77231894860 (71.93 GB)DFS Used%: 3.68%DFS Remaining%: 84.78%Configured Cache Capacity: 4294967296 (4 GB)Cache Used: 0 (0 B)Cache Remaining: 4294967296 (4 GB)Cache Used%: 0.00%Cache Remaining%: 100.00%Xceivers: 10Last contact: Mon Jan 21 10:45:24 CST 20193.使用命令平衡数据[hdfs@hadoop-master ~]$ hdfs balancer19/01/21 10:49:19 INFO balancer.Balancer: namenodes  = [hdfs://vg-cdh-test]19/01/21 10:49:19 INFO balancer.Balancer: parameters = Balancer.Parameters [BalancingPolicy.Node, threshold = 10.0, max idle iteration = 5, number of nodes to be excluded = 0, number of nodes to be included = 0, run during upgrade = false]Time Stamp               Iteration#  Bytes Already Moved  Bytes Left To Move  Bytes Being Moved19/01/21 10:49:21 INFO net.NetworkTopology: Adding a new node: /default/192.168.0.117:5001019/01/21 10:49:21 INFO net.NetworkTopology: Adding a new node: /default/192.168.0.118:5001019/01/21 10:49:21 INFO net.NetworkTopology: Adding a new node: /default/192.168.0.121:5001019/01/21 10:49:21 INFO balancer.Balancer: 0 over-utilized: []19/01/21 10:49:21 INFO balancer.Balancer: 0 underutilized: []The cluster is balanced. Exiting...2019-1-21 10:49:21                0                  0 B                 0 B               -1 B2019-1-21 10:49:21       Balancing took 2.738 seconds4.查看数据平衡后的数据分布情况[hdfs@hadoop-master ~]$ hdfs dfsadmin -report > /tmp/bh[hdfs@hadoop-master ~]$ cat /tmp/bhConfigured Capacity: 273287419086 (254.52 GB)Present Capacity: 209660106924 (195.26 GB)DFS Remaining: 199596266468 (185.89 GB)DFS Used: 10063840456 (9.37 GB)DFS Used%: 4.80%Under replicated blocks: 0Blocks with corrupt replicas: 0Missing blocks: 0Missing blocks (with replication factor 1): 0-------------------------------------------------Live datanodes (3):Name: 192.168.0.117:50010 (hadoop-node01)Hostname: hadoop-node01Rack: /defaultDecommission Status : NormalConfigured Capacity: 91095806362 (84.84 GB)DFS Used: 3354603928 (3.12 GB)Non DFS Used: 12246663170 (11.41 GB)DFS Remaining: 69809213772 (65.01 GB)DFS Used%: 3.68%DFS Remaining%: 76.63%Configured Cache Capacity: 4294967296 (4 GB)Cache Used: 0 (0 B)Cache Remaining: 4294967296 (4 GB)Cache Used%: 0.00%Cache Remaining%: 100.00%Xceivers: 10Last contact: Mon Jan 21 10:50:10 CST 2019Name: 192.168.0.118:50010 (hadoop-master)Hostname: hadoop-masterRack: /defaultDecommission Status : NormalConfigured Capacity: 91095806362 (84.84 GB)DFS Used: 3354632600 (3.12 GB)Non DFS Used: 29501419522 (27.48 GB)DFS Remaining: 52554428748 (48.95 GB)DFS Used%: 3.68%DFS Remaining%: 57.69%Configured Cache Capacity: 4294967296 (4 GB)Cache Used: 0 (0 B)Cache Remaining: 4294967296 (4 GB)Cache Used%: 0.00%Cache Remaining%: 100.00%Xceivers: 10Last contact: Mon Jan 21 10:50:10 CST 2019Name: 192.168.0.121:50010 (hadoop-node02)Hostname: hadoop-node02Rack: /defaultDecommission Status : NormalConfigured Capacity: 91095806362 (84.84 GB)DFS Used: 3354603928 (3.12 GB)Non DFS Used: 4823252994 (4.49 GB)DFS Remaining: 77232623948 (71.93 GB)DFS Used%: 3.68%DFS Remaining%: 84.78%Configured Cache Capacity: 4294967296 (4 GB)Cache Used: 0 (0 B)Cache Remaining: 4294967296 (4 GB)Cache Used%: 0.00%Cache Remaining%: 100.00%Xceivers: 10Last contact: Mon Jan 21 10:50:10 CST 2019--------------------------------------------------------------------------------5.对比数据平衡前后的报告信息[hdfs@hadoop-master ~]$ diff /tmp/bq /tmp/bh2,4c2,4< Present Capacity: 209643254756 (195.25 GB)< DFS Remaining: 199579415524 (185.87 GB)< DFS Used: 10063839232 (9.37 GB)---> Present Capacity: 209660106924 (195.26 GB)> DFS Remaining: 199596266468 (185.89 GB)> DFS Used: 10063840456 (9.37 GB)19,21c19,21< DFS Used: 3354603520 (3.12 GB)< Non DFS Used: 12246245786 (11.41 GB)< DFS Remaining: 69809631564 (65.02 GB)---> DFS Used: 3354603928 (3.12 GB)> Non DFS Used: 12246663170 (11.41 GB)> DFS Remaining: 69809213772 (65.01 GB)30c30< Last contact: Mon Jan 21 10:45:24 CST 2019---> Last contact: Mon Jan 21 10:50:10 CST 201938,40c38,40< DFS Used: 3354632192 (3.12 GB)< Non DFS Used: 29517959578 (27.49 GB)< DFS Remaining: 52537889100 (48.93 GB)---> DFS Used: 3354632600 (3.12 GB)> Non DFS Used: 29501419522 (27.48 GB)> DFS Remaining: 52554428748 (48.95 GB)42c42< DFS Remaining%: 57.67%---> DFS Remaining%: 57.69%49c49< Last contact: Mon Jan 21 10:45:24 CST 2019---> Last contact: Mon Jan 21 10:50:10 CST 201957,59c57,59< DFS Used: 3354603520 (3.12 GB)< Non DFS Used: 4823982490 (4.49 GB)< DFS Remaining: 77231894860 (71.93 GB)---> DFS Used: 3354603928 (3.12 GB)> Non DFS Used: 4823252994 (4.49 GB)> DFS Remaining: 77232623948 (71.93 GB)68c68< Last contact: Mon Jan 21 10:45:24 CST 2019---> Last contact: Mon Jan 21 10:50:10 CST 2019

在生产上实际的操作如下：hdfs dfsadmin -fs hdfs://uhadoop-mzwc2w-master2:8020 -setBalancerBandwidth 3145728000[hadoop@uhadoop-mzwc2w-master1 ~]$ hdfs dfsadmin -fs hdfs://uhadoop-mzwc2w-master2:8020 -setBalancerBandwidth 3145728000Balancer bandwidth is set to 3145728000[hadoop@uhadoop-mzwc2w-master1 ~]$ hdfs dfsadmin -fs hdfs://uhadoop-mzwc2w-master1:8020 -setBalancerBandwidth 3145728000Balancer bandwidth is set to 3145728000//一定要2个节点namenode节点 master1 master2 都要执行 在生产上2019/7/22 星期一 我在master1 和master2 都执行balancer [hadoop@uhadoop-mzwc2w-master1 majihui0718]$ nohup hdfs balancer > balancer.log & //生产上这样处理[hadoop@uhadoop-mzwc2w-master2 majihui0722]$ pwd/home/hadoop/majihui0722[hadoop@uhadoop-mzwc2w-master2 majihui0722]$ lltotal 596-rw-r--r-- 1 hadoop wheel 609794 Jul 22 14:11 balancer.log我们是1000M的网，给300M的带宽用于datanode的数据balance

参考链接
https://www.cnblogs.com/qingyunzong/p/8535995.html
参考链接
HDFS balance策略详解 https://www.jianshu.com/p/f7c1cd476601