On 9/12/22 4:45 PM, Dmitry Bogdanov wrote: >> Remember how ESX used to send a RTPG to one port and expect that it got >> every group and that the state info was all in sync (basically opposite >> if what's in the spec now)? >> >> The spec and ESX were updated, but I don't know if other OSs did this and >> if/when everyone was updated. Do you know this info? Are the old ESX versions >> that worked like that end of life? > ESXi is kinda a pain. But fortunately it has nothing to do with that > patch 😄 I didn't get why that is. How do you set up a distributed/cluster/HA target? I'm probably missing that part. Software drivers like iscsi I get, but for HW drivers I didn't see how you do it. For example, if you have 2 systems/nodes running LIO which each export the same device via 1 port each where one is active/optimized and the other is standby and you are using qla2xxx, then on the local node would you create 2 groups: [root@ol8n4 alua]# pwd /sys/kernel/config/target/core/iblock_0/device0/alua [root@ol8n4 alua]# ls default_tg_pt_gp local remote Then under the mapped lun: [root@ol8n4 lun_0]# pwd /sys/kernel/config/target/..../tpgt_1/lun/lun_0 in the alua_tg_pt_gp file you set that to local. That would then have tg_pt_gp_members set, but remote would not. Before your patch, windows and ESX could do a RTPG to just one port/path and we would return the default, local and remote groups. We don't want the default group, but we wanted the local and remote one returned. With your patch we only return the the local one now. I wasn't sure how that works for drivers like qla2xxx. For iscsi, you could just mirror the remote node locally, so you would have a second tpg: [root@ol8n4 lun_0]# pwd /sys/kernel/config/target/..../tpgt_2/lun/lun_0 and in there set alua_tg_pt_gp to remote. Your patch works fine for that because both groups then have tg_pt_gp_members set so if the OS just does a RTPG to one path/port you get all the groups.
