ALM-14028 Number of Blocks to Be Supplemented Exceeds the Threshold

Alarm Description

The system checks the number of blocks to be supplemented every 30 seconds and compares the number with the threshold. The number of blocks to be supplemented has a default threshold. This alarm is generated when the number of blocks to be supplemented exceeds the threshold.

You can change the threshold specified by Blocks Under Replicated (NameNode) by choosing O&M > Alarm > Thresholds > Name of the desired cluster > HDFS > File and Block.

If Trigger Count is set to 1 and the number of blocks to be supplemented is less than or equal to the threshold, this alarm is cleared. If Trigger Count is greater than 1 and the number of blocks to be supplemented is less than or equal to the threshold, this alarm is cleared.

Alarm Attributes

Alarm Parameters

Impact on the System

Data stored in HDFS is lost. HDFS may enter the security mode and cannot provide write services. Lost block data cannot be restored.

Possible Causes

• The DataNode instance is abnormal.
• Data is deleted.
• The number of replicas written into the file is greater than the number of DataNodes.

Handling Procedure

1. On FusionInsight Manager, choose O&M. In the navigation pane on the left, choose Alarm > Alarms. On the page that is displayed, check whether alarm ALM-14003 Number of Lost HDFS Blocks Exceeds the Threshold is generated.
   • If yes, go to 2.
   • If no, go to 3.

2. Rectify the fault according to the handling procedure of ALM-14003 Number of Lost HDFS Blocks Exceeds the Threshold. Five minutes later, check whether the alarm is cleared.
   • If yes, no further action is required.
   • If no, go to 3.

3. Log in to the HDFS client as user root. The user password is defined by the user before the installation. Contact the MRS cluster administrator to obtain the password. Run the following commands:
   • Security mode:

     cd Client installation directory

     source bigdata_env

     kinit hdfs

   • Normal mode:

     su - omm

     cd Client installation directory

     source bigdata_env

4. Run the hdfs fsck / >> fsck.log command to obtain the status of the current cluster.
5. Run the following command to count the number (M) of blocks to be replicated:

   cat fsck.log | grep "Under-replicated"

6. Run the following command to count the number (N) of blocks to be replicated in the /tmp/hadoop-yarn/staging/ directory:

   cat fsck.log | grep "Under replicated" | grep "/tmp/hadoop-yarn/staging/" | wc -l

   

   /tmp/hadoop-yarn/staging/ is the default directory. If the directory is modified, obtain it from the configuration item yarn.app.mapreduce.am.staging-dir in the mapred-site.xml file.

7. Check whether the percentage of N is greater than 50% (N/M > 50%).
   • If yes, go to 8.
   • If no, go to 9.

8. Run the following command to reconfigure the number of file replicas in the directory (set the number of file replicas to the number of DataNodes or the default number of file replicas):

   hdfs dfs -setrep -w Number of file replicas/tmp/hadoop-yarn/staging/

   

   To obtain the default number of file replicas:

   Log in to FusionInsight Manager, choose Cluster > Services > HDFS > Configurations > All Configurations, and search for the dfs.replication parameter. The value of this parameter is the default number of file replicas.

   Check whether the alarm is cleared 5 minutes later.

   • If yes, no further action is required.
   • If no, go to 9.

Collect the fault information.

9. On FusionInsight Manager, choose O&M. In the navigation pane on the left, choose Log > Download.
10. Expand the drop-down list next to the Service field. In the Services dialog box that is displayed, select HDFS for the target cluster.
11. 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.
12. Contact O&M engineers and provide the collected logs.

Alarm Clearance

This alarm is automatically cleared after the fault is rectified.

Related Information

None.