[HDFS-16316] Improve DirectoryScanner: add regular file check related block

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.9.2
• Fix Version/s: 3.4.0
• Component/s: datanode
• Labels: pull-request-available
• Assignee: JiangHua Zhu
• Reporter: JiangHua Zhu
• Created: 11/Nov/21 07:19
• Updated: 31/Oct/23 04:03
• Resolved: 22/Feb/22 02:16

Description

Something unusual happened in the online environment.
The DataNode is configured with 11 disks (${dfs.datanode.data.dir}). It is normal for 10 disks to calculate the used capacity, and the calculated value for the other 1 disk is much larger, which is very strange.
This is about the live view on the NameNode:

This is about the live view on the DataNode:

We can look at the view on linux:

There is a big gap here, regarding'/mnt/dfs/11/data'. This situation should be prohibited from happening.

I found that there are some abnormal block files.
There are wrong blk_xxxx.meta in some subdir directories, causing abnormal computing space.
Here are some abnormal block files:

Such files should not be used as normal blocks. They should be actively identified and filtered, which is good for cluster stability.

Issue Links

• links to: GitHub Pull Request #3861