Hadoop集群日常运维

一）备份namenode的元数据
namenode中的元数据非常重要，如丢失或者损坏，则整个系统无法使用。因此应该经常对元数据进行备份，最好是异地备份。
1、将元数据复制到远程站点
（1）以下代码将secondary namenode中的元数据复制到一个时间命名的目录下，然后通过scp命令远程发送到其它机器

1. #!/bin/bash
2. export dirname=/mnt/tmphadoop/dfs/namesecondary/current/`date +%y%m%d%H`
3. if [ ! -d ${dirname} ]
4. then
5. mkdir ${dirname}
6. cp /mnt/tmphadoop/dfs/namesecondary/current/* ${dirname}
7. fi
8. scp -r ${dirname} slave1:/mnt/namenode_backup/
9. rm -r ${dirname}

（2）配置crontab，定时执行此项工作
0 0,8,14,20 * * * bash /mnt/scripts/namenode_backup_script.sh

2、在远程站点中启动一个本地namenode守护进程，尝试加载这些备份文件，以确定是否已经进行了正确备份。

（二）数据备份
对于重要的数据，不能完全依赖HDFS，而是需要进行备份，注意以下几点
（1）尽量异地备份
（2）如果使用distcp备份至另一个hdfs集群，则不要使用同一版本的hadoop，避免hadoop自身导致数据出错。

（三）文件系统检查
定期在整个文件系统上运行HDFS的fsck工具，主动查找丢失或者损坏的块。
建议每天执行一次。

1. [jediael@master ~]$ hadoop fsck /
2. ……省略输出（若有错误，则在此外出现，否则只会出现点，一个点表示一个文件）……
3. .........Status: HEALTHY
4. Total size: 14466494870 B
5. Total dirs: 502
6. Total files: 1592 (Files currently being written: 2)
7. Total blocks (validated): 1725 (avg. block size 8386373 B)
8. Minimally replicated blocks: 1725 ( 100.0 %)
9. Over-replicated blocks: 0 ( 0.0 %)
10. Under-replicated blocks: 648 ( 37.565216 %)
11. Mis-replicated blocks: 0 ( 0.0 %)
12. Default replication factor: 2
13. Average block replication: 2.0
14. Corrupt blocks: 0
15. Missing replicas: 760 ( 22.028986 %)
16. Number of data-nodes: 2
17. Number of racks: 1
18. FSCK ended at Sun Mar 01 20: 17: 57 CST 2015 in 608 milliseconds
20. The filesystem under path '/' is HEALTHY

（1）若hdfs-site.xml中的dfs.replication设置为3，而实现上只有2个datanode，则在执行fsck时会出现以下错误；
/hbase/Mar0109_webpage/59ad1be6884739c29d0624d1d31a56d9/il/43e6cd4dc61b49e2a57adf0c63921c09:  Under replicated blk_-4711857142889323098_6221. Target Replicas is 3 but found 2 replica(s).
注意，由于原来的dfs.replication为3，后来下线了一台datanode，并将dfs.replication改为2，但原来已创建的文件也会记录dfs.replication为3，从而出现以上错误，并导致 Under-replicated blocks:       648 (37.565216 %)。

（2）fsck工具还可以用来检查一个文件包括哪些块，以及这些块分别在哪等

1. [jediael@master conf]$ hadoop fsck /hbase/Feb2621_webpage/c23aa183c7cb86af27f15d4c2aee2795/ s/30bee5fb620b4cd184412c69f70d24a7 -files -blocks -racks
3. FSCK started by jediael from /10.171.29.191 for path /hbase /Feb2621_webpage/c23aa183c7cb86af27f15d4c2aee2795/ s/30bee5fb620b4cd184412c69f70d24a7 at Sun Mar 01 20:39:35 CST 2015
4. /hbase/Feb2621_webpage/c23aa183c7cb86af27f15d4c2aee2795/ s/30bee5fb620b4cd184412c69f70d24a7 21507169 bytes, 1 block(s): Under replicated blk_7117944555454804881_3655. Target Replicas is 3 but found 2 replica(s).
5. 0. blk_7117944555454804881_3655 len=21507169 repl=2 [/default-rack/ 10.171. 94.155: 50010, /default-rack/ 10.251. 0. 197: 50010]
7. Status: HEALTHY
8. Total size: 21507169 B
9. Total dirs: 0
10. Total files: 1
11. Total blocks (validated): 1 (avg. block size 21507169 B)
12. Minimally replicated blocks: 1 ( 100.0 %)
13. Over-replicated blocks: 0 ( 0. 0 %)
14. Under-replicated blocks: 1 ( 100.0 %)
15. Mis-replicated blocks: 0 ( 0. 0 %)
16. Default replication factor: 2
17. Average block replication: 2.0
18. Corrupt blocks: 0
19. Missing replicas: 1 ( 50.0 %)
20. Number of data-nodes: 2
21. Number of racks: 1
22. FSCK ended at Sun Mar 01 20: 39: 35 CST 2015 in 0 milliseconds
25. The filesystem under path '/hbase/Feb2621_webpage/c23aa183c7cb86af27f15d4c2aee2795/s/30bee5fb620b4cd184412c69f70d24a7' is HEALTHY

此命令的用法如下：

1. [jediael@master ~]$ hadoop fsck -files
2. Usage: DFSck & lt;path& gt; [-move | - delete | -openforwrite] [-files [-blocks [-locations | -racks]]]
3. & lt;path& gt; start checking from this path
4. -move move corrupted files to /lost+found
5. - delete delete corrupted files
6. -files print out files being checked
7. -openforwrite print out files opened for write
8. -blocks print out block report
9. -locations print out locations for every block
10. -racks print out network topology for data-node locations
11. By default fsck ignores files opened for write, use -openforwrite to report such files. They are usually tagged CORRUPT or HEALTHY depending on their block allocation status
12. Generic options supported are
13. -conf & lt;configuration file& gt; specify an application configuration file
14. -D & lt;property=value& gt; use value for given property
15. -fs & lt; local|namenode:port& gt; specify a namenode
16. -jt & lt; local|jobtracker:port& gt; specify a job tracker
17. -files & lt;comma separated list of files& gt; specify comma separated files to be copied to the map reduce cluster
18. -libjars & lt;comma separated list of jars& gt; specify comma separated jar files to include in the classpath.
19. -archives & lt;comma separated list of archives& gt; specify comma separated archives to be unarchived on the compute machines.
21. The general command line syntax is
22. bin/hadoop command [genericOptions] [commandOptions]

详细解释请见《hadoop权威指南》P376

（四）均衡器
随时时间推移，各个datanode上的块分布来越来越不均衡，这将降低MR的本地性，导致部分datanode相对更加繁忙。

均衡器是一个hadoop守护进程，它将块从忙碌的DN移动相对空闲的DN，同时坚持块复本放置策略，将复本分散到不同的机器、机架。

建议定期执行均衡器，如每天或者每周。

(1)通过以下命令运行均衡器

1. [jediael@master log]$ start-balancer.sh
2. starting balancer, logging to / var/ log/hadoop/hadoop-jediael-balancer-master.out

查看日志如下：

1. [jediael@master hadoop]$ pwd
2. /var/ log/hadoop
3. [jediael@master hadoop]$ ls
4. hadoop-jediael-balancer-master. log hadoop-jediael-balancer-master.out
5. [jediael@master hadoop]$ cat hadoop-jediael-balancer-master. log
6. 2015 -03 -01 21: 08: 08, 027 INFO org.apache.hadoop.net.NetworkTopology: Adding a new node: / default-rack/ 10.251 .0 .197: 50010
7. 2015 -03 -01 21: 08: 08, 028 INFO org.apache.hadoop.net.NetworkTopology: Adding a new node: / default-rack/ 10.171 .94 .155: 50010
8. 2015 -03 -01 21: 08: 08, 028 INFO org.apache.hadoop.hdfs. server.balancer.Balancer: 0 over utilized nodes:
9. 2015 -03 -01 21: 08: 08, 028 INFO org.apache.hadoop.hdfs. server.balancer.Balancer: 0 under utilized nodes:

（2）均衡器将每个DN的使用率与整个集群的使用率接近，这个“接近”是通过-threashold参数指定的，默认是10%。
（3）不同节点之间复制数据的带宽是受限的，默认是1MB/s，可以通过hdfs-site.xml文件中的dfs.balance.bandwithPerSec属性指定（单位是字节）。

（五）datanode块扫描器
每个datanode均会运行一个块扫描器，定期检测本节点上的所有块，若发现存在错误（如检验和错误），则通知namenode，然后由namenode发起数据重新创建复本或者修复。
扫描周期由dfs.datanode.scan.period.hours指定，默认为三周（504小时）。
通过地址以下地址查看扫描信息：
（1）http://datanote:50075/blockScannerReport
列出总体的检测情况

1. Total Blocks : 1919
2. Verified in last hour : 4
3. Verified in last day : 170
4. Verified in last week : 535
5. Verified in last four weeks : 535
6. Verified in SCAN_PERIOD : 535
7. Not yet verified : 1384
8. Verified since restart : 559
9. Scans since restart : 91
10. Scan errors since restart : 0
11. Transient scan errors : 0
12. Current scan rate limit KBps : 1024
13. Progress this period : 113%
14. Time left in cur period : 97.14%

（2）http://123.56.92.95:50075/blockScannerReport?listblocks
列出所有的块及最新验证状态
blk_8482244195562050998_3796 : status : ok     type : none   scan time : 0               not yet verified
blk_3985450615149803606_7952 : status : ok     type : none   scan time : 0               not yet verified
尚未验证的情况如上。更多学习到：http://www.dajiangtai.com/