[TAJO-1959] Improve AWS S3 file system support

• Type: Improvement
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: S3, Storage
• Assignee: Unassigned
• Reporter: Hyunsik Choi
• Created: 03/Nov/15 03:48
• Updated: 08/Jan/16 07:02

Description

We need to improve AWS S3 support. As you know, S3 has different characteristics as follows:

• No move operation. Move is emulated by copy and remove.
• too slow directory listing
• unnecessary locality (i.e., always remote access)
• eventual consistency

Emulating S3 via just HDFS implementation may cause lots of performance degradation. We need to mitigate the performance degradation points.

This is an umbrella issue to track sub tasks.

Issue Links

• links to: How does Qubole improve S3 performance?
• links to: Optimizing Hadoop for S3 – Part 1

Sub-Tasks

• Separate HDFS space handler and S3 space handler: Resolved
• S3TableSpace needs to extend FileTableSpace: Resolved
• Use list S3 files using AmazonS3Client instead of using S3A: Open
• Insert clause to partitioned table fails on S3: In Progress
• Implement S3TableSpace::commitTable.: Resolved
• Support DirectOutputCommitter for AWS S3 file system: Open
• Optimize Partition Table Split Computation for Amazon S3: Open