[HADOOP-14084] Shell.joinThread swallows InterruptedException

• Type: Bug
• Status: Patch Available
• Priority: Minor
• Resolution: Unresolved
• Assignee: Xiaobing Zhou
• Reporter: Shixiong Zhu
• Created: 15/Feb/17 20:52
• Updated: 29/Mar/17 03:19

Description

In "Shell.joinThread", when the user tries to interrupt the thread that runs Shell.joinThread, it will catch InterruptedException and propagate it to thread t. However, it doesn't set the interrupt state of the current thread before returning, so the user codes won't know it's already interrupted.

See https://github.com/apache/hadoop/blob/9e19f758c1950cbcfcd1969461a8a910efca0767/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/util/Shell.java#L1035

Issue Links

• relates to: SPARK-19617 Fix the race condition when starting and stopping a query quickly
• relates to: HADOOP-10622 Shell.runCommand can deadlock
• relates to: SPARK-19599 Clean up HDFSMetadataLog