[HADOOP-16711] S3A bucket existence checks to support v2 API and "no checks at all"

• Type: Sub-task
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 3.2.1
• Fix Version/s: 3.3.0
• Component/s: fs/s3
• Labels: performance
• Target Version/s: 3.3.0

Release Note

The probe for an S3 bucket existing now uses the V2 API, which fails fast if the caller lacks access to the bucket.
The property fs.s3a.bucket.probe can be used to change this. If using a third party library which doesn't support this API call, change it to "1".

to skip the probe entirely, use "0". This will make filesystem instantiation slightly faster, at a cost of postponing all issues related to bucket existence and client authentication until the filesystem API calls are first used.

• Assignee: Mukund Thakur
• Reporter: Rajesh Balamohan
• Created: 14/Nov/19 19:24
• Updated: 13/Jul/22 14:21
• Resolved: 21/Feb/20 13:48

Description

1. in S3AFileSystem.initialize(), we check for the bucket existing with verifyBucketExists(), which calls s3.doesBucketExist(). But that doesn't check for auth issues.

s3. doesBucketExistV2() does at validate credentials, and should be switched to. This will help things fail faster

2. For performance it would be good to skip verifyBuckets call during S3A filesystem init(). This would save a call to S3 during init method.

Issue Links

• breaks: HADOOP-16995 ITestS3AConfiguration proxy tests fail when bucket probes == 0
• causes: HADOOP-17801 No error message reported when bucket doesn't exist in S3AFS
• contains: HADOOP-15990 S3AFileSystem.verifyBucketExists to move to s3.doesBucketExistV2
• is depended upon by: SPARK-24000 S3A: Create Table should fail on invalid AK/SK
• is related to: HADOOP-15990 S3AFileSystem.verifyBucketExists to move to s3.doesBucketExistV2
• is superceded by: HADOOP-17454 [s3a] Disable bucket existence check - set fs.s3a.bucket.probe to 0
• links to: GitHub Pull Request #1838
• links to: GitHub Pull Request #1845