[
https://issues.apache.org/jira/browse/FLINK-41587?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
ASF GitHub Bot updated FLINK-41587:
-----------------------------------
Labels: pull-request-available (was: )
> Bump the SeaweedFS test image to 4.54
> -------------------------------------
>
> Key: FLINK-41587
> URL: https://issues.apache.org/jira/browse/FLINK-41587
> Project: Flink
> Issue Type: Technical Debt
> Components: Tests
> Reporter: Martijn Visser
> Priority: Major
> Labels: pull-request-available
>
> The SeaweedFS image that our S3 tests use (chrislusf/seaweedfs:4.47, in
> DockerImageVersions and common_s3_seaweedfs.sh) mishandles the trailing
> checksum of aws-chunked uploads. When the trailer lines arrive in separate
> TCP segments, it rejects the request with "400 The Content-Md5 you specified
> is not valid", see https://github.com/seaweedfs/seaweedfs/issues/11582. AWS
> SDK v2 sends this trailer by default on every PutObject and UploadPart.
> With s3.crt.enabled=true, every write from flink-s3-fs-native fails this way,
> because the CRT HTTP client sends the trailer in small pieces. The default
> Apache client can hit it too, depending on how the body is split in transit.
> Real S3 accepts the same requests.
> SeaweedFS fixed this in 4.49.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)