Hardware
The IT-mode tradeoff
This one was not really a choice. The controller could only present hardware RAID arrays, and Unraid cannot work that way, so the card had to be reflashed before the server could do the job I bought it for. The interesting part is what that cost me, and why I would have done it anyway.
The problem
- The controller as shipped could only present hardware RAID arrays. It wants to own the disks and hand the operating system a finished volume.
- Unraid needs the opposite: direct visibility of individual drives, because it manages parity and per-disk allocation itself.
- Given an abstracted volume, the thing that is supposed to be managing the array cannot see the array. Unraid does not work at all in that configuration.
What I weighed
- Leave the card in RAID mode and give up on Unraid entirely, which would also mean giving up mixing drive sizes and expanding a disk at a time.
- Buy a card that ships in pass-through mode instead of reflashing the one already in the chassis.
- Crossflash the existing controller, accepting the risk of writing new firmware to hardware that was working.
The choice, and the real reason
- Crossflashed the controller to pass-through so drives are presented directly to the operating system. Necessity forced it, but the second-order benefit is what made me comfortable with it.
- The array stays reconstructible on any Linux machine. With hardware RAID, recovering after a card failure can mean sourcing a compatible controller with compatible firmware, which is a dependency on a specific piece of aging silicon still being findable years from now.
- Pass-through removes that dependency. The disks are just disks, and the array metadata belongs to the operating system.
What it cost
- Out-of-band management no longer sees the drives at all. In IT mode the controller stops reporting them, so the server's management interface, which is exactly the tool meant to work when the machine is otherwise unreachable, is blind to disk health.
- Inside Unraid, visibility is complete and works fine. The loss is specifically the out-of-band path, not the operating system view.
- That concentrates drive health into a single reporting path. If the storage OS is down, nothing is watching the disks.
- Crossflashing itself is a one-way-feeling operation on hardware that was working beforehand.
Where that leaves monitoring
- Honestly: thinner than it should be, and I know it.
- Today that is Unraid reporting drive health, plus rolling checksum verification against recorded hashes so silent corruption surfaces as an alert rather than as a file that will not open later.
- Beszel is planned and not yet deployed. Until it is, a single monitoring path is the accepted cost of a choice I did not really have an alternative to.
- The lesson generalizes: when you consolidate a capability into one component, you have also consolidated its failure into one component. Worth knowing which ones you have done that to.