Thanks Rohith -- I wonder (well Robert wondered to me ... I'm just the wonder messenger here) if this maybe has been the cause of some of our JDK 25 build timeouts? E.g. https://lists.apache.org/thread/pgqkr2g5fjs8k2gg510l14hsyk97bhxq maybe?
I think the "GC stuck at 100% heap" issue ( https://bugs.openjdk.org/browse/JDK-8375467) happens when Lilliput (compact object headers - COH) is enabled, and you're using ParallelGC, with JDK 25? But it's not clear if COH is enabled on some/none/all of our Jenkins JDK 25 build configs? Robert mentioned the Jenkins Lucene builds use Temurin JDK builds but Temurin seems not to enable COH by default (according to Gemini), so ... Mike McCandless http://blog.mikemccandless.com On Fri, Jan 16, 2026 at 11:29 AM Rohith Reddy Nedhunuri < [email protected]> wrote: > Hi Lucene community, > > > We've identified a significant performance regression in Amazon when > upgrading to JDK 25 that affects applications using Parallel GC > > > Issue: JDK-8375467 - Unproductive Full GC cycles with Compact Object > Headers enabled > > • Link: https://bugs.openjdk.org/browse/JDK-8375467 > > • Affects: JDK 25 with Parallel GC + Compact Object Headers > > > Our Experience: We observed heap usage spiking to 100%, dramatically > slower indexing, and much larger segment merges in production. > > > Root Cause: The new Parallel GC implementation for Lilliput (Compact > Object Headers) has a bug where non-maximum-compaction full GCs become > unproductive, leading to repeated unproductive GC cycles. > > > The OpenJDK team is working on a fix. > > > Best regards, > > Rohith Reddy Nedhunuri >
