ALM-14019 DataNode Non-heap Memory Usage Exceeds the Threshold

Description

The system checks the non-heap memory usage of the HDFS DataNode every 30 seconds and compares the actual usage with the threshold. The non-heap memory usage of the HDFS DataNode has a default threshold. This alarm is generated when the non-heap memory usage of the HDFS DataNode exceeds the threshold.

Users can choose O&M > Alarm > Thresholds> Name of the desired cluster > HDFS to change the threshold.

This alarm is cleared when the no-heap memory usage of the HDFS DataNode is less than or equal to the threshold.

Attribute

Parameters

Impact on the System

If the memory usage of the HDFS DataNode is too high, data read/write performance of HDFS will be affected.

Possible Causes

Non-heap memory of the HDFS DataNode is insufficient.

Procedure

Delete unnecessary files.

1. Log in to the HDFS client as user root. Run the cd command to go to the client installation directory, and run the source bigdata_env command.

   If the cluster adopts the security mode, perform security authentication.

   Run the kinit hdfs command and enter the password as prompted. Obtain the password from the administrator.

2. Run the hdfs dfs -rm -r file or directory path command to delete unnecessary files.
3. Check whether the alarm is cleared.
   • If yes, no further action is required.
   • If no, go to 4.

Check the DataNode JVM non-heap memory usage and configuration.

4. On the FusionInsight Manager portal, choose Cluster > Name of the desired cluster > Services > HDFS.
5. In the Basic Information area, click NameNode(Active). The HDFS WebUI is displayed.
   

   By default, the admin user does not have the permissions to manage other components. If the page cannot be opened or the displayed content is incomplete when you access the native UI of a component due to insufficient permissions, you can manually create a user with the permissions to manage that component.

6. On the HDFS WebUI, click the Datanodes tab to view the number of blocks of all DataNodes that report alarms.
7. On the FusionInsight Manager portal, choose Cluster > Name of the desired cluster > Services > HDFS > Configurations > All Configurations. In Search, enter GC_OPTS to check the GC_OPTS non-heap memory parameter of HDFS->DataNode.

Adjust system configurations.

8. Check whether the memory is properly configured based on the number of blocks in 6 and the memory parameters configured for DataNode in 7.
   • If yes, go to 9.
   • If no, go to 12.
   

   The mapping between the average number of blocks of a DataNode instance and the DataNode memory is as follows:

   • If the average number of blocks of a DataNode instance reaches 2,000,000, the reference values of the JVM parameters of the DataNode are as follows: -Xms6G -Xmx6G -XX:NewSize=512M -XX:MaxNewSize=512M
   • If the average number of blocks of a DataNode instance reaches 5,000,000, the reference values of the JVM parameters of the DataNode are as follows: -Xms12G -Xmx12G -XX:NewSize=1G -XX:MaxNewSize=1G

9. Modify the GC_OPTS parameter of the DataNode based on the mapping between the number of blocks and memory.
10. Save the configuration and click Dashboard > More > Restart Service.
    

    The service will be unavailable during the restart. In addition, upper-layer services that depend on the service are affected.

11. Check whether the alarm is cleared.
    • If yes, no further action is required.
    • If no, go to 12.

Collect fault information.

12. On the FusionInsight Manager portal, choose O&M > Log > Download.
13. Select the following services in the required cluster from the Service.
    • ZooKeeper
    • HDFS

14. Click in the upper right corner, and set Start Date and End Date for log collection to 10 minutes ahead of and after the alarm generation time, respectively. Then, click Download.
15. Contact the O&M personnel and send the collected logs.

Alarm Clearing

After the fault is rectified, the system automatically clears this alarm.

Related Information

None