Hi all, I have a question regarding OSD memory behavior during read workloads. During sequential read tests, I observe that a small number of OSD daemons sometimes grow their memory usage far beyond configured limits and end up consuming most of the host memory. This memory is not reclaimed even after the workload stops, unless the OSD is restarted. I have already set explicit limits such as osd_memory_target and bluestore cache size (with autotune off), but some OSDs still exceed them significantly. My main questions are: 1. Is it considered normal for OSDs to exceed configured memory targets by a large margin during read workloads? 2. Is there any supported way to strictly enforce OSD memory limits (beyond current tuning options), without relying on container OOM or suffering major performance degradation? If this behavior is expected or already known, I’d appreciate any pointers. I can provide more details if needed. Thanks in advance. Best regards, Chanu Yu