Moving disks to a different controller - does the DG survive?

[vmjoe]

Hello

I'm planning a disk extension of a vSAN 7.0.3 (soon 8.0) cluster, and ended up with different theoretical options. One would involve the move of an entire disk group (one cache and two capacity SSDs) to a newly to-be-added controller (to keep disk groups and controllers aligned as one failure domain).

Question is: will the DG still be detected as present and intact, when connected via a different controller? (By the disk identifiers, or some vSAN signature?) Or would this move from one ThinkSystem 430-8i controller to an added identical one, destroy the DG?

I know I can evacuate DGs etc., but it takes time and increases the risk, time and involvement of local technicians significantly, so I would avoid it, and rather choose an option where the current DGs would be kept as-is, and just get extended and complemented by additional DGs.

[TheBobkin]

@vmjoe, Provided they are in passthrough mode and not RAID0 Virtual Devices then this *should* be fine - I am fairly certain we don't have a support-statement to this effect as it not guaranteed to work and likely not officially supported, but over the years I have seen it work multiple times (e.g. taking disks out of one server and putting them in a donor server 😝). I would advise you test it on one DG before deciding to do it on many.

vSAN uses the disk UUID, metadata and how this ties to Cache-tier and metadata stored on that describing the DG to maintain the structure e.g. mountpoint (and maybe even identifier) are not relevant.

[vmjoe]

Thanks for you reply.

The hosts are of course designed as vSAN nodes, so the 430-8i is a dumb HBA, does not have cache and does not even offer RAID functionality:

"Non-RAID (JBOD mode) support for SAS and SATA HDDs and SSDs (RAID not supported)"
https://lenovopress.lenovo.com/lp0649-thinksystem-430-8i-16i-internal-sas-hba

Problem with actually testing it, we would have to purchase a second controller already, which is difficult or impossible in a large corporation. We can either buy them with all the SSDs for all servers, or not. 🙄

But your statement about donor servers makes me believe it would work. Actually, I remember something similar: in a two-become-one exercise, we consolidated SSDs from hosts with too small capacities into fewer hosts, and when a cache and capacity device 'accidentally' ended up together in the same target server, they were recognized as a disk group (with issues, like second capacity drive missing)... 

Anyway, I think we should go with the configuration that does not need to move the disks that way. The options are: Two HBAs with two disk groups per HBA at 14 TB (3 capacity devices each), vs. only one DG per HBA with 28 TB (5 capacity devices each). Both configurations would be with the existing 750 GB cache device only. Price difference about 37,000 Euro...