Configuring Performance Tuning for Cluster Backup and Restoration

Modifying YARN and BackupPlugin Parameters

1. Log in to FusionInsight Manager of the active cluster, choose Cluster > Services > Yarn, click Configurations > All Configurations, and modify the YARN service parameters in Table 1 as required.

   For details about how to log in to FusionInsight Manager, see Accessing MRS Manager.

   Table 1 YARN component performance parameters

   Parameter	Default Value	Description
   mapreduce.task.stuck.timeout-ms	1200000	Maximum timeout (in milliseconds) before YARN receives remote task's first heartbeat. This parameter is in order to avoid waiting for the container to start indefinitely, which made task stuck in the NEW state.
   mapreduce.task.timeout	600000	Number of milliseconds before a task will be terminated if it neither reads an input, writes an output, nor updates its status string. The unit is milliseconds.
   yarn.distcp.fs-limits.max-directory-items	10000000	Maximum number of files that can be backed up per HDFS task when DistCp is used for HDFS backup. For Hive database-level backups, modify this parameter in advance if there are many files.
   distcp.copy.attributes	-	File attributes to be copied to the remote HDFS during backup. If this parameter is left blank, all attributes are copied. It is recommended that you select attributes based on your site requirements. More attributes increase the impact on performance. For example, QUOTA and ACL significantly affect the performance.
   distcp.liststatus.threads	20	Number of threads used to calculate the total files for copy when a DistCp task starts.
   distcp.pool.size	50	Number of threads for modifying permissions and deleting redundant files during a DistCp task.
   distcp.skip.crc	false	Whether to perform CRC verification after the DistCp task is complete.
   yarn.app.mapreduce.am.resource.mb	4096	Memory limit (in MB) for the DistCp ApplicationMaster. If a single backup directory contains many files, you must increase the ApplicationMaster memory. For details, refer to the backup.xms and backup.xmx values in the BackupPlugin. Note: The backup process memory size is determined by the total number of files scanned by all concurrent backup tasks, whereas yarn.app.mapreduce.am.resource.mb is the memory required for the files scanned by a single task. If a single task scans many files, you must modify the parameters. If multiple concurrent tasks scan a large total volume of files, you must also adjust each parameter based on its specifications.
   mapreduce.reduce.memory.mb	4096	Memory limit (in MB) for the DistCp Reduce task.

2. Click Save to save the settings.
3. (Skip this step if only the parameters that take effect dynamically are modified.) Click Dashboard. In the upper right corner of the displayed page, choose More > Restart Service, enter the password of the current user, and click OK to restart the YARN service.
4. Log in to the active and standby OMS nodes of the active cluster. Switch to user omm and edit the ${BIGDATA_HOME}/om-server/om/etc/om/backupconfig/backup.properties file. Based on Table 2, update backup.xms and backup.xmx, ensuring that both parameters are set to the same value. Refer to Table 3 for recommended values.

   Table 2 Modifying BackupPlugin parameters

   Parameter	Default Value	Description
   backup.xms	1024	Initial heap memory size (in MB) for the backup management process. This value depends on the number of files scanned concurrently.
   backup.xmx	8192	Maximum heap memory size (in MB) for the backup management process. This value depends on the number of files scanned concurrently.

   Table 3 Reference values for backup.xms and backup.xmx (in MB).

   Number of File Objects	Recommended Value (backup.xms and backup.xmx)
   10,000,000	12288
   20,000,000	24576
   30,000,000	32768

5. Log in to the active OMS node as user omm and run the following command to restart the Controller process of the active OMS node for the settings to take effect:

   sh ${BIGDATA_HOME}/om-server/om/sbin/restart-controller.sh