> It seems fine, perhaps, to allow __additional__ feature sets this way, but I 
> think the "portable" family really needs to be something compiled in.  The 
> core value proposition is that "zpool create -o features=portable-2016" is 
> the same __everywhere__, and allowing it to be easily overridden seems likely 
> to erode that value.
The core value proposition is that a pool created with zpool create -o 
features=portable-2016 *can be imported* on all systems this tag applies to, so 
one dosn't have to remember what exact features can be enabled without breaking 
compatibility.

It dosn't imply that all features that *would be* compatible *have to* be 
activated, should case a user not want a feature (for whatever reason) he 
should be able to remove it from the predefined list so *his* pools that *he* 
creates don't use it.

The side value proposition is shutting up zpool status about upgrading a pool 
by being able to tell it that it should only complain when features included in 
a tag are not enabled.
I for one at least want to be able to add a tag named 'no-nag' (with an empty 
feature list) just to get rid of zfs insisting on my pools urgently needing 
upgrades.

> By allowing a custom define you've now also added responsibility for having 
> that define available on every system that imports the pool. 
What is your reasoning for that conclusion? Importing a pool is (and should 
stay) a question of the activated features being compatible with the running 
ZFS implementation. 

Having the prerequisite of a tag needing to be defined on a system to import a 
pool with it would boil down to introducing backward incompatibility (inability 
to add new not-bleeding edge tags as they would break import on older systems). 
I really hope it's not designed that way, in case it is: back to the drawing 
board, please.

------------------------------------------
openzfs: openzfs-developer
Permalink: 
https://openzfs.topicbox.com/groups/developer/Td66d70a086ae5a9e-M68300d0964b67d9d08f67725
Delivery options: https://openzfs.topicbox.com/groups/developer/subscription

Reply via email to