Hi, everyone: I'm currently encountering a performance issue in Nautilus Bluestore. With the perf dump stats by mgr prometheus module, I found Bluestore will produce tens of milliseconds in prepare stage. The latency is calculated with l_bluestore_state_prepare_lat logger. The max latency I have seen is about 70ms, and those abnormal latency only occur in high IO load sometimes. When the abnormal latency occurs, the backend nvme disk is also suffering a high await.
As far as I know, prepare stage in Bluestore does not contain disk operations, how that huge latency be possible? The `bluestore alloctor fragmentation block` shows about 0.5, and with re-constructing osd, it indeed reduced latency in the whole IO link to an acceptable level, including prepare stage. Could anyone give some tips on this? Thanks! _______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-le...@ceph.io
