[HDFS-15552] Let DeadNode Detector also work for EC cases

• Type: Sub-task
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 3.3.0
• Component/s: dfsclient, ec
• Labels: pull-request-available
• Target Version/s: 3.5.0
• Assignee: imbajin
• Reporter: dark_num
• Created: 01/Sep/20 12:03
• Updated: 17/Jan/24 11:05

Description

Currently, the EC stream (`DFSStripedInputStream`) is not handled properly while exception occurs.

For example, while reading EC-blocks, if the client timed out when connecting to the DataNode, it will throws `SocketTimeoutException` , then add current DN to localDeadNode.

However, the local dead nodes will not be removed until the stream is closed, which will cause missing block IOException to be thrown in the use scenario of Hbase.

So we need to use detector to deal with dead nodes under EC to avoid reading failures.

 

Issue Links

• links to: GitHub Pull Request #2287