On Mon, 17 Aug 2026 20:46:15 GMT, David Holmes <[email protected]> wrote:
>> pulled the gc logs from the two 1g failures. the majors did run but >> reclaimed almost nothing right before the OOMs >> >> GC(4) Major Collection (Allocation Stall) 1024M >> 1024M 2.424s Out Of >> Memory (Allocator0) >> GC(4) Major Collection (Allocation Stall) 1024M >> 1018M 2.419s Out Of >> Memory (Allocator3) >> >> three majors completed before the first internal OOM in both runs, seven or >> eight before the java OOME. later majors in the same runs reclaim fine, >> 1024M >> 682M and 1024M >> 272M, but the allocators already got OOM by then > > @shivshah-oracle I'm waiting to hear @stefank 's view on what the logs are > showing. Something doesn't seem right to me. @dholmes-ora I need full -Xlog:gc* logs to tell more. My guess is that there's a starvation problem. That one thread tries to allocate, wait for the major collections to complete, but while doing so, other threads are using up all the memory before the major collection can finish and hand out memory to the waiting thread. ------------- PR Comment: https://git.openjdk.org/jdk/pull/32229#issuecomment-5325428451
