[OAK-7008] Estimation for FULL can be off sometimes

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 1.7.12, 1.8.0
• Component/s: segment-tar
• Labels: compaction gc
• Epic Link: TarMK GC scalability
• Assignee: Michael Dürig
• Reporter: Michael Dürig
• Created: 30/Nov/17 15:23
• Updated: 08/Dec/17 14:23
• Resolved: 30/Nov/17 15:45

Description

Since OAK-6883, FULL estimation compares segmentstore size with the previous FULL. There can be cases where the current segmentstore is smaller than the previous FULL (i.e. due to TAIL cleaning up more). This leads to FULL being skipped for much more than anticipated.

A case to illustrate this scenario:

Start Oak with a 10 GB repo
GC #1: run FULL results in segmenstore of 20GB
GC #2: run TAIL results in segmentstore of 11GB
GC #3: run FULL (saturday) - skipped because the reference is 20GB from the previous FULL

FULL be executed again only when the segmentstore grows back above 20GB, which might be too late.

Estimation should take this situation into account this and take a better decision.