[HADOOP-19200] Reduce the number of headObject when opening a file with the s3 file system

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: 3.4.0, 3.3.6
• Component/s: fs/s3
• Assignee: Unassigned
• Reporter: Oliver Caballero Alvarez
• Created: 08/Jun/24 12:20
• Updated: 10/Jun/24 14:07
• Resolved: 10/Jun/24 13:32

Description

In the implementation of the S3 filesystem, of the hadoop aws package, if you use it with spark, every time you open a file for anything you will have to send two Head Objects, since to open the file, you will first look to see if this file exists, executing a HeadObject, and then when opening it, the implementation, both of sdk1 and sdk2, forces you to make a head object again. This is not the fault of the implementation of this class (S3AFileSystem), but of the abstract FileSystem class of the Hadoop core, since it does not allow the FileStatus to be passed but only allows the use of Path.

If the FileSystem implementation is changed, it could be used to not have to request that HeadObject again.

Issue Links

• depends upon: SPARK-48571 Reduce the number of accesses to S3 object storage
• duplicates: HADOOP-15229 Add FileSystem builder-based openFile() API to match createFile(); S3A to implement S3 Select through this API.
• is depended upon by: HADOOP-19199 Include FileStatus when opening a file from FileSystem
• is related to: HADOOP-19199 Include FileStatus when opening a file from FileSystem