Question

A Java service running G1GC has fine average GC behavior but develops periodic ~1.5s full-GC pauses (not normal young/mixed pauses) that cause request timeouts a few times an hour. Heap usage looks moderate (~55% of 16GB) and there's no obvious memory leak (heap returns to baseline after each pause). GC logs show frequent 'humongous allocation' entries and a rising humongous-region count, plus occasional 'to-space exhausted.' A week ago a feature began returning larger result payloads (some multi-MB byte arrays / lists). How do you triage and mitigate?