rmustacc: zfs doesn't have lfs's "segment cleaner" which would empty out a metaslab for fresh sequential writing.

alanc: reads mainly like "whoever ported zfs to linux didn't understand linux's user namespaces"

jbk: I think you have the metaslab spacemap thing (which indeed uses less space to keep track of free/allocated when the slab is nearly full) confused with intent logs (which use blocks that aren't marked allocated in the spacemaps..)

yeah it could be the spacemaps i'm think of... it's been a _long_ time since I've had to look that deeply :)

intent logs seem like they'd work well since you're really just appending

in other news I gave Hans's NVME host memory buffer patch (code.illumos.org/c/illumos-gate/+/4929) a spin on a test machine where it would used and .. I guess it works? nothing blew up, things seem to be performant.

→ CODE REVIEW 4929: 18307 nvme: want support for Host Memory Buffer feature (NEW) | illumos.org/issues/18307

I don't know nvme-level things well enough to seriously review it.