Hi, we recently added a new SAS shelf to our FAS3140 and have realised it is cabelled incorrectly with the SAS connections from IOM A(circle and square) both going to controller A and the SAS connections from IOM B both going to controller B as below.
Can we simply swap the cables connected to the circle ports on the new shelf to resolve the issue?
Note the disks are not yet in use but have been assigned with controller B as their owner and are marked as spares on controllerB, we do however have an urgent requirement to provision the new disks in a new aggregate asap.
Is it a completely new stack with no other shelves on it?
Yes, you can swap around cables going into circle ports, but I am not 100% sure if it is supported to do it on-line - whilst hot-adding shelves is supported, hot-removal isn't (you will be removing a single path only though).
Thanks for the reply, it is a completely new stack with only this shelf attached and I believe it is possible to remove the single connections online as it will just transition the shelf to single path from multipath, however not sure if the system will be happy with the cables plugged back into different ports.
Hot removing a path is ok for example when adding to a stack. But plugging in a port to a new location can cause a crash. Removing with the node down in maintenance mode to clear things prior to recabling mpio is supported but definitely open a support case.
It's actually the other way around - hot-removing a cable isn't supported, whilst plugging it back in (hot-adding a second path) is supported.
I'd stick to the scenario described by Scott.
There is no hot removal of a shelf but you can do loop or stack maintenance when a node is down and in this case leave a single connection per node.
The First thing I would do is open a case to make sure you are supported. Then you likely will use cluster takeover and giveback. Probably Take one node down. Disconnect the single incorrect port commection leaving one correct connection. Boot and giveback then the same on the other node. End result is the stack will have one connection per node...with the wrong connection removed while the node is taken over. Then hot add the mpio connections with both nodes up. But run it by support first. Several steps but should be able to get there using the cluster to keep services running. Cifs will have to reconnect but other protocols stay running with each cluster takeover/giveback.
I have opened a case with NetApp concurrently with this discussion. I think your suggestion might be safest as I know I can remove the connections safely whilst the system is running (the controller would simply transition to single path) but adding them back into different ports would cause issues I believe. I will await NetApp supports response and report back to this discussion.
We did this,
Failover the node and halt the node. Redo the SAS cabling like moving from other to other. boot in maintenance mode and verified to make sure that node can access the disk shelves through new ports.
Halt the system, which will exit from maintenane mode. boot the system. Failback the node.
Repeated the same steps for other node in HA pair, It worked without any issues.