Description of problem: Allow overriding of the following LIO configuration settings via the "[target]" section in "/etc/ceph/iscsi-gateway.cfg" and/or via a new "/iscsi-target/<IQN> reconfigure <key> <value>" action within the "gwcli". cmdsn_depth immediate_data initial_r2t max_outstanding_r2t first_burst_length max_burst_length max_recv_data_segment_length max_xmit_data_segment_length Tweaking these settings are useful for high IOPS/throughput environments.
Do we have performance numbers? Is it for any release of lio (for example, applicable for next RHEL)? Will it be helpful for any lio users with high iops, not just ceph?
I don't have any numbers recorded (since this wasn't a perf and scale exercise and I have limited access to resources). However, for an example, the default command depth is 64 IOs, so for fast storage w/ high queue-depth workloads, it would be nice to tweak 'cmdsn_depth' up. This is just about exposing the knobs and levers that LIO already exposes via the Ceph 'gwcli' interface (multi-node-aware replacement for targetcli).
Is there a specific use case for supporting setting params in iscsi-gateway.cfg? If not, let's only allow setting params like this via tools like gwcli for now unless it is specifically documented in the RHCS guide. gwcli allows us to only have to reset the iscsi sessions so it is less disruptive. It should only cause a a couple second delay in running IO. The rbd-target-gw service reload support should do something similar, but I think we still have some bugs when invalid values are used.
Since the problem described in this bug report should be resolved in a recent advisory, it has been closed with a resolution of ERRATA. For information on the advisory, and where to find the updated files, follow the link below. If the solution does not work for you, open a new bug report. https://access.redhat.com/errata/RHBA-2019:0020