[SPARK-20656] Incremental parsing of event logs in SHS

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Incomplete
• Affects Version/s: 2.3.0
• Component/s: Spark Core
• Labels: bulk-closed
• Assignee: Unassigned
• Reporter: Marcelo Masiero Vanzin
• Created: 08/May/17 18:16
• Updated: 25/May/21 01:54
• Resolved: 25/May/21 01:43

Description

This feature is mentioned in the spec attached to SPARK-18085 but there's not a lot of discussion about it.

It would be good to implement incremental parsing of event logs in the SHS. With the new work, UI data is stored on disk, so it should be possible to save enough metadata about the event log and the state of the listeners to allow one to resume parsing the log of a live application at the point where it stopped in the previous iteration.

This would considerably speed up parsing on updates, and could be done speculatively so that UIs for new applications are available in the SHS almost immediately.

I'm filing this as a separate enhancement because I don't want to block SPARK-18085 on this.

Issue Links

• is blocked by: SPARK-18085 SPIP: Better History Server scalability for many / large applications
• is duplicated by: SPARK-22264 History server will be unavailable if there is an event log file with large size
• is related to: SPARK-28867 InMemoryStore checkpoint to speed up replay log file in HistoryServer
• links to: GitHub Pull Request #26821