SAS Cabling - splitting a SAS Stack Without Shutdown

[martro]

I have the following situation

FAS6290, CDOT 8.2.3

It has 2 mixed (SSD & SAS) stacks

- Stack #1: 4xDS2246x600GB SAS + 1xDS2246x200GB SSD (2 x sas cables to controller 1, 2 x sas cables to controller 2)

- Stack #2: 4xDS2246x600GB SAS + 1xDS2246x200GB SSD (2 x sas cables to controller 1, 2 x sas cables to controller 2)

SSD shelves are the first (last) of the stack in both stacks above.

I want to separate SSD and SAS shelves, splitting each of the stacks in 2, WITHOUT SHUTDOWN. That will end with 4 stacks (2 x SAS, 2 x SSD)

If I remove ONLINE the SAS cables between the SSD shelf and the first of the SAS shelves in each stack, will those 2 stacks be automatically converted in 4 stacks?

Can I then close these 4 stacks with 8 SAS cables going back to the controllers to restore the redundant connection?

It is clear I can do this WITH a shutdown, but can it be done WITHOUT?

[aborzenkov]

Online removal is supported only for completely unused shelves. What you are doing is effectively removing shelves from stack. Go figure ...

[martro]

Thanks for your answer. I had already read this:

Hot-removing disk shelves in systems running Data ONTAP 8.2.1 or later

 https://library.netapp.com/ecmdocs/ECMP1367947/html/GUID-2B80FBD2-007D-4D19-8EB1-9CCEED211001.html

but I don't think it fits what I am doing, since I am not removing the shelf (all shelves remain connected to both controllers at any time)

May be the following quick hand-drawing makes it more clear.

In step #1, I remove 2 SAS cables, but the removal keeps all shelves connected (single path, instead of dual).

In step #2, I reconnect the 4 disconnected shelf SAS ports, but to new SAS ports in the controller.

It's clear I can power down the storage and do this changes offline, but can I do them without disrupting service?

[aborzenkov]

Well, hot cable move between adapters is supported as of 8.2.1; see kb2023388. The catch is "between adapters". Not between shelf and adapter 🙂 I think this is one of those twilight zones where you do not get direct answers.

[shatfield]

That looks dicey.  But I suspect odds of success would improve if you did a takeover/giveback between step 1 and step 2 so each node gets to boot in a dual stack single path ha config.  Interesting lab experiment, but wouldn't try it on a production filer.  If you do try it, tell us if it panics on you.