[HADOOP-17181] Handle transient stream read failures in FileSystem contract tests

• Type: Sub-task
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 3.3.0
• Fix Version/s: 3.3.1, 3.4.0
• Component/s: fs/s3
• Labels: pull-request-available
• Assignee: Steve Loughran
• Reporter: Steve Loughran
• Created: 03/Aug/20 18:45
• Updated: 08/Sep/21 12:15
• Resolved: 09/Sep/20 11:02

Description

Seen 2x recently, failure in ITestS3AContractUnbuffer as not enough data came back in the read.

The contract test assumes that stream.read() will return everything, but it could be some buffering problem. Proposed: switch to ReadFully to see if it is a quirk of the read/get or is something actually wrong with the production code.

Issue Links

• links to: GitHub Pull Request #2286