Open baryluk opened 13 years ago
Thanks for contacting KQ Stor support.
W'll implement your suggested zfs iSCSI target support in our port.
Would be cool if IQN for tgtadm will be automatically derived from zpool and volume name/path, so no configuration will be needed. Probably removing and changing volume name will be harder, as just exporting iSCSI target using tgtadm is very simple. If UUID will be used then changing name will be not a problem, but then cloning will be. Probably best to have target acceible both using path as well UUID. I'm not sure how it is done on solaris, but AFAIK solaris exports volumes using UUID, and targets have additional Alias so it can be easly listed or accessed using alias from clients, improving discoverability and maintaince.
BTW. One need also to be aware that there is probably some limits in current kernel (probably only about 500 volumes can be exported, but I'm not sure).
Not only clones need to be considered, but also snapshots. I'm not sure if snapshots are accessible using iSCSI under Solaris, but they would be a read-only targets with obviously separate UUID, and path including @snapshotname. Also reanaming of snapshots needs to be taken into account (question remains what to do when clients is using a volume, clone or snapshot, and we rename or remove it - should it still use it until it closes connection, or we should close connections and unexport volume immiedietly ? i opt for the second one - administrator should know what he is doing).
I do not know i anybody was thinking about it, but ZFS on solaris have iSCSI target support, which can be easily enabled just by "zfs set shareiscsi=on some/path/to/volume"
On linux it could be just translated to some tgtadm commands with automatically adjusted target names and block device path (assuming some/path/to/volume is already available as a block device somewhere in the /dev/ ).