[HADOOP-19131] WrappedIO to export modern filesystem/statistics APIs in a reflection friendly form

• Type: New Feature
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 3.4.0, 3.3.6
• Fix Version/s: 3.5.0, 3.4.1
• Component/s: fs, fs/azure, fs/s3
• Labels: pull-request-available
• Target Version/s: 3.4.1
• Assignee: Steve Loughran
• Reporter: Steve Loughran
• Created: 28/Mar/24 16:50
• Updated: 23/Sep/24 10:53
• Resolved: 21/Aug/24 13:22

Description

parquet, avro etc are still stuck building with older hadoop releases.

This makes using new APIs hard (PARQUET-2171) and means that APIs which are 5 years old such as HADOOP-15229 just aren't picked up.

This lack of openFIle() adoption hurts working with files in cloud storage as

• extra HEAD requests are made
• read policies can't be explicitly set
• split start/end can't be passed down

HADOOP-18679 added a new WrappedIO class.

This jira proposes extending this with

• more of the filesystem/input stream methods
• iOStatistics
• Pull in parquet DynMethods to dynamially wrap and invoke through tests. This class, DynamicWrappedIO is intended to be copied into libraries (parquet, iceberg) for their own use.
• existing tests to use the dynamic binding for end-to-end testing.

+then get into the downstream libraries and use where appropriate

Issue Links

• causes: HADOOP-19285 [ABFS] Restore ETAGS_AVAILABLE to abfs path capabilities
• depends upon: HADOOP-18679 Add BulkDelete API for paged delete of files and objects
• is depended upon by: PARQUET-2493 HadoopInputFile to pass down FileStatus when opening file.
• is depended upon by: PARQUET-2486 Improve Parquet IO Performance within cloud datalakes
• is related to: HADOOP-19199 Include FileStatus when opening a file from FileSystem
• links to: GitHub Pull Request #6686
• links to: GitHub Pull Request #6995