CMS: Heap Size and Occupancy before GC

[AlinaHovakimyan]

Heap size and occupancy before GC are returning heap size and occupancy after GC.

[dsgrieve]

@AlinaHovakimyan - Which collector are we talking about here? And can you submit a log fragment or whole log.

[AlinaHovakimyan]

@dsgrieve the log file used:
cms_gen.log

For example in 22
2024-08-16T15:20:51.002+0400: 34.193: [GC (CMS Initial Mark) [1 CMS-initial-mark: 0K(349568K)] 54146K(506816K), 0.0326151 secs] [Times: user=0.03 sys=0.00, real=0.03 secs]

For this line I was expecting to receive:
heap used before: 0 k
heap size before: 349568 K
heap used after: 54146 K
heap size after: 506816 K

Please let me know, if my expectations were incorrect.

Please see the received event in the screenshot:

[brunoborges]

/relevance-check

[github-actions]

Relevance Assessment: Still Relevant

• Summary: The reported behavior is confirmed in the current codebase — CMS Initial Mark events report identical values for heap occupancy "before" and "after" collection, which is confusing but rooted in how the parser models snapshot-only GC phases. No fix or related PR has been merged.

• Evidence:

  • The CMSTenuredPoolParser.initialMark() method (line 75–80) calls trace.getOccupancyWithMemoryPoolSizeSummary() which creates a MemoryPoolSummary with identical before/after values: new MemoryPoolSummary(occupancy, size, occupancy, size) (GCLogTrace.java:207).
  • For the log line [1 CMS-initial-mark: 0K(349568K)] 54146K(506816K), the parser correctly identifies 0K(349568K) as tenured and 54146K(506816K) as heap — but both getOccupancyBeforeCollection() and getOccupancyAfterCollection() on the heap MemoryPoolSummary return 54146K, since Initial Mark is a snapshot, not a before→after transition.
  • Existing tests in PreUnifiedCMSTenuredParserTest.java and UnifiedGenerationalEventsTest.java parse Initial Mark events but do not assert on heap before/after occupancy values for the pre-unified CMS path (they only check ClassCastException and event types), so the semantic issue is not caught by tests.
  • No commits or PRs have addressed this issue — searching commit history for CMS heap/occupancy/InitialMark fixes yields no results.
  • The GenerationalHeapParser.initialMark() (line 825–829) has the same pattern.
  • This is the only open issue about CMS heap occupancy values (CMS: Heap Size and Occupancy before GC #385). No duplicates found.

• Recommendation: ✅ Keep open — The issue is still valid. The underlying behavior (before == after for snapshot events) is by design in getOccupancyWithMemoryPoolSizeSummary(), but the reporter raises a fair point: it's semantically misleading. The fix could either (a) document that Initial Mark events are snapshots where before/after are intentionally equal, or (b) model these events differently so consumers can distinguish snapshot phases from actual collections. This is an actionable improvement that hasn't been addressed.

Generated by Relevance Check Agent for issue #385