Description
HADOOP-14747 added support for CanUnbuffer to S3A. One issue for applications that rely on the CanUnbuffer functionality, is that the current S3A code only merges input stream stats when the stream is closed. Applications using CanUnbuffer might not close streams for a long time, so any stats reported by input streams don't get reported in the filesystem-wide metrics.
This JIRA proposes merging input stream statistics when S3AInputStream#unbuffer is called.
Attachments
Issue Links
- is required by
-
IMPALA-8544 Expose additional S3A / S3Guard metrics
- Open
- relates to
-
SPARK-28091 Extend Spark metrics system with user-defined metrics using executor plugins
- Resolved
- links to