Experience on Namenode backup and restore

Hadoop version: Hadoop 2.2.0.2.0.6.0-0009

Well, We can do this by building Secondary Namenode, Checkpoint node or Backup node.

Example:

Assuming you have a Secondary Namenode.

1. Check secondary namenode checkpoint status:

dfs.namenode.secondary.http-address in  %HADOOP_CONF_DIR%/hdfs-site.xml

fs.namenode.checkpoint.dir in  %HADOOP_CONF_DIR%/hdfs-site.xml

dfs.namenode.checkpoint.edits.dir in  %HADOOP_CONF_DIR%/hdfs-site.xml

dfs.namenode.checkpoint.period in  %HADOOP_CONF_DIR%/hdfs-site.xml

2. Backup your real time checkpoint by hand:

On Secondary namenode, Stop Hadoop secondary namenode service.

Run cmd.exe by user hadoop ( or some users have full permission )

[plain] view plaincopy

1. Runas /user:hadoop cmd.exe  

You must have user hadoop password.

Backup real time checkpoint:

[plain] view plaincopy

1. cmd>%hadoop_home%/bin/hadoop secondarynamenode -checkpoint force  

Start Hadoop secondary namenode service. and check secondary namenode checkpoint status ( see step 1)

3. Stop Namenode services or reboot Namenode ( if hadoop service set to booting manual ,the services would all stop after reboot )

As for test, I backup my dfs.namenode.name.dir (i.e C:\hdpdata\hdfs\nn)  first for my next test ( restore from my namenode dir backup ) .

Delete all files in C:\hdpdata\hdfs\nn ,

Open  dfs.namenode.checkpoint.dir (see %HADOOP_CONF_DIR%/hdfs-site.xml ) in secondary namenode (i.e. c:\hdpdata\hdfs\snn )

Copy all secondary checkpoint files( except the lock file) from this folder to your namenode's checkpoint dir (dfs.namenode.checkpoint.dir the same as secondary namenode)

Make sure namenode's checkpoint dir is empty already !

4.  Restore from checkpoint dir

Run cmd.exe by user hadoop ( or some users have full permission )

[plain] view plaincopy

1. Runas /user:hadoop cmd.exe  

You must have user hadoop password.

Use this command to start hadoop service and import checkpoint from checkpoint dir

[plain] view plaincopy

1. cmd>%hadoop_home%/bin/hdfs namenode -importcheckpoint  

Use ctrl+C to stop service which is completed. and Delete your namenode's checkpoint dir (dfs.namenode.checkpoint.dir the same as secondary namenode)

Start service by this command:

[plain] view plaincopy

1. cmd>start_local_hdp_services.cmd  

Levae safemode

[plain] view plaincopy

1. cmd>%hadoop_home%/bin/hdfs dfsadmin -safemode leave  

Balance you HDFS:

[plain] view plaincopy

1. cmd>%hadoop_home%/bin/hdfs balancer -threshold 5  

5. Confirm your Hadoop service is restored successfully.

Open URL  http://namenode:50070/ to check if there are some missing block. If yes. Please kindly check where they are and what they are.

Because restore from secondary namenode isn't a real time restore solution. It may lost the last time what you do in the jobtracker. It doesn't matter. Just delete them.

Tips: If you want to restore a real time backup, please use multiplicate namenode dir mode. see next post... ...