Noticed a commit added this to FreeBSD stable. Skimming through the
commit and screenshots from the differential revision (haven't tested it
personally yet) I see that it describes it as changing options to the
ZFS pool but is then followed by changing options for datasets within
the pool. There is a difference for options that apply to a pool versus
options that apply to a dataset. I thought reading and writing
properties is only sometimes interchangeable between pool and dataset
with ZFS tools doing what is intended instead of what was asked while
other times you just get an error. Can this be clarified for adjusting
dataset(s) properties instead of pool properties if that is what is
being done?
Example screenshots form discussion while it was worked on normally
showed compression=lz4 and atime=off which is the default for all but 1
dataset last I reviewed what the installer is doing. I figure altering
compression, encryption, (ref)reservation/quota would be more likely to
be adjusted and away from defaults when adjustments are made by users.
Setting options that apply to all datasets on the one pool is a handy
quick override. Further expanding how users are prompted to be able to
adjust each dataset individually would be more useful than having to
change all datasets to use same override. This then opens up options
like a read only dataset that won't be used for storage getting
refreservation to force a certain amount of space to remain free; this
example is a dataset property that is easy to add after the fact but
likely more steps for someone wanting to automate an install with it
being done during install but some proerties require being set during
creation or have limitations and consequences otherwise.
It would be useful if the installer gave us disk selection, likely
combined with pool(s) creation but both could be separate configuration
screens with one feeding to the next; that opens up options for multiple
pool creation during install and with different devices, layouts, and
options. Then having control of pool specific layout+options which I
presume is quite hard to implement in a way giving full control to the
user. After that, a list of all partitions/datasets where selecting them
allows changing individual partition/dataset properties would be handy;
inheriting the properties to other datasets should be part of the user
prompt when adjusting datasets that others depend on.
Such an installer layout should make adding geli, labels, etc. start
to become easier to provide as additional options or entirely new layers
between these for the installer's interface.
