[KAFKA-6530] Use actual first offset of messages when rolling log segment for magic v2

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.0.0
• Assignee: Dhruvil Shah
• Reporter: Jason Gustafson
• Created: 03/Feb/18 07:25
• Updated: 25/Apr/18 09:42
• Resolved: 17/Mar/18 17:30

Description

We've implemented a heuristic to avoid overflowing when rolling a log segment to determine the base offset of the next segment without decompressing the message set to find the actual first offset. With the v2 message format, we can find the first offset without needing decompression, so we can set the correct base offset exactly.

Issue Links

• links to: GitHub Pull Request #4660