[HADOOP-19171] S3A: handle alternative forms of connection failure

• Type: Sub-task
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 3.4.0, 3.3.6
• Component/s: fs/s3
• Assignee: Unassigned Assign to me
• Reporter: Steve Loughran
• Created: 13/May/24 13:15
• Updated: 14/May/24 18:23

Description

We've had reports of network connection failures surfacing deeper in the stack where we don't convert to AWSApiCallTimeoutException so they aren't retried properly (retire connection and repeat)

Unable to execute HTTP request: Broken pipe (Write failed)

 Your socket connection to the server was not read from or written to within the timeout period. Idle connections will be closed. (Service: Amazon S3; Status Code: 400; Error Code: RequestTimeout

note, this is v1 sdk but the 400 error is treated as fail-fast in all our versions and I don't think we do the same for the broken pipe. that one is going to be trickier to handle as unless that is coming from the http/tls libraries "broken pipe" may not be in the newer builds. We'd have to look for the string in the SDKs to see what causes it and go from there

Issue Links

• links to: https://github.com/aws/aws-sdk-java-v2/issues/1124
• links to: https://github.com/aws/aws-sdk-java/issues/2328