[HDFS-16244] Add the necessary write lock in Checkpointer#doCheckpoint()

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 3.4.0
• Component/s: namenode
• Labels: pull-request-available
• Assignee: JiangHua Zhu
• Reporter: JiangHua Zhu
• Created: 29/Sep/21 07:50
• Updated: 13/Oct/21 18:43
• Resolved: 13/Oct/21 09:21

Description

When BackupNode is enabled, Checkpointer#doCheckpoint() will start to work.
When the image file needs to be reloaded, there is a call link, for example:
FSImage#reloadFromImageFile()>FSNamesystem#clear()>FSDirectory#reset().
In FSDirectory#reset(), the write lock needs to be acquired in advance, for example:
void reset() {
writeLock();
try

{ ...... }

finally

{ writeUnlock(); }

}
However, no write lock has been acquired before this.
You will get an exception message at this time, for example:
java.lang.AssertionError: Should hold namesystem write lock

Issue Links

• links to: GitHub Pull Request #3497