Replacing a failed drive

[maniek]

The fine manual says this about replacing an encrypted drive

Wait until the resilvering is complete. Next, restore the encryption keys to the pool.​

This is seriously scary. Resilvering is a process which lasts multiple hours on a nontrivial install. It is possibly the most work the NAS has ever done, so possibility of crash is real (if only because of thermal issues).

Can you restore the encryption keys just after resilvering has started? Is it dangerous? What will actually happen if the NAS crashes during resilvering?

 

[cyberjock]

Well, yes it's scary to the casual observer, but it's not really scary at all.

When you replace a failed disk, the disk being resilvered is not useful until the resilvering is complete (pretty much the same for hardware RAID). So issues from crashes and such are no more risky than unencrypted pools, with the exception that you have to remember to rekey.

All that being said, if your server wasn't already tested and verified to be able to handle a high workload, you haven't adequately prepared your server to handle the potential workload anyway. You should NOT be having problems with thermals or any other problem. You should also be doing regular scrubs, which are almost the same kind of workload as a resilver. So you are almost implying that you are regularly having thermal problems if that is a problem.

So I don't think there's anything to worry about.

 

[maniek]

I am doing regular scrubs, have an UPS, the NAS has never crashed, and I just installed an extra fan. I should be OK :) Resilvering is currently in progress...

I find it strange that replacing a disk invalidates the encryption keys. Why is that so?

 

[cyberjock]

I am doing regular scrubs, have an UPS, the NAS has never crashed, and I just installed an extra fan. I should be OK :) Resilvering is currently in progress...

I find it strange that replacing a disk invalidates the encryption keys. Why is that so?

They keys aren't put on the hard drive until you rekey (this may or may not be correct in the latest version, there's some confusion among some). I know back in 8.3.1 you had to rekey (which would invalidate the current keys, so you *had* to rekey the hard drive, then redo the recovery key and regular key for everything to be fully functional again. This may not be true in 9.3. I haven't tried to test this yet, so I can't tell you for certainty if the steps are redundant or not.