Hi, kim Sorry for bothering you! recently I'm reading the code of your current pr about pool unit qos, but I encountered with some difficulty of understanding about how to statistic io of a pool to do QoS control, Could you please help me? is there exits some misunderstanding about it? thx! first: I think every real client corresponding to a objecter, every objecter record the sum of op refer to one pool since last request, but the objecter only statistic its own ops refer to one client. if we have two client request to one same pool, which corresponding to two objecter, every objecter statistic its own ops(every object record it own delta, rho corresponding to the pool), the two group of qos_params is independent, and the two client request may enter the mClockPoolQueue with two independent qos_params for the pool, the qos_params not really include hole the pool delta, rho . how could it throttle pool QoS. second is: why the whole io through one pool divide into several clients by osd_op_type_t, only client_op and osd_subop use the pool QoS, others op type use its own value by config. as the whole op type in pool include not only include client_op and osd_subop, how could it throttle the pool io under pool QoS? why it need to distinguish specific op types for one pool but not just use one pool as a client ? at last the statistic iops of pool only include of client_op or osd_subop? I'm waiting for your reply, thx! Any advise will be appreciate! -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
