It's one of my least favourite things about vault as a product - it would have been technically feasible to not require unsealing for read-only operations, and thus not made a 2am restart a critical failure. I made a PoC once (I don't think it's still published anywhere) that does exactly this. Unfortunately they chose not to do this.

Admittedly simplicity is a feature in and of itself, and the read-only unsealing requires more complex asymmetric cryptography, but the excellent nacl[0] makes this a lot easier than it used to be.

0: https://nacl.cr.yp.to/box.html

I kid you not, the main reason we haven't implemented vault yet, is not because we're worried about security. It's because we're deathly afraid of locking ourselves out by making a mistake, taking the whole system down.

Meaning, this makes Vault more of a "let's really dedicate time to think of every possible scenario" type implementation rather than "let's just keep adding a couple of secrets a week".

What has other people's experiences been?

They're moving Cloud Auto Unseal to the open source Vault soon, thankfully.

That said, for most organizations, I've never been all that convinced that the multi-key-holder model provides much benefit.

do not use single node instance, in a HA fashion it is not a problem. Even with the open source version, which we are using.