On Mon, Mar 25, 2019 at 1:43 PM Richard Laager <[email protected]> wrote:
> On 3/25/19 8:15 AM, Josh Paetzel wrote:
> > "-o features=portable" means that features that are currently supported
> by all tier-1 platforms. This definition will change over time.
> Specifically, as of the current date, the following all supported the
> features:
> > - a numbered release of FreeBSD (currently FreeBSD 11.2)
> > - a numbered release of ZFSonLinux (currently ZoL 0.7.12)
> > - the code in illumos-gate as of 1 year prior (currently December 7,
> 2017, commit b8e5ecd6 for bug 8895).
>
> Is this the current version of each tier-1 platform, or the oldest
> still-supported version of each tier-1 platform?
>
The current (most recent) version.
>
> Based on your example of FreeBSD 11.2, it sounds like the oldest
> still-supported version.
>
The examples above were prepared some time ago, sorry for the confusion.
>
> Using the oldest still-supported release, rather than the latest
> release, will cause more lag in features=portable. Is this desirable or
> necessary, given the portable-YYYY options would exist?
>
> > From these definitions, users can make the following conclusions:
> > - I upgraded this machine sometime in 2017, therefore if I want to
> bring a pool to it, I can use "-o features=portable-
> > 2017" or earlier.
>
> For Linux users, this doesn't necessarily hold true. The ZoL source
> release version isn't necessarily as relevant as their distro's packaged
> version, which additionally can be complicated by backports or similar.
>
That's true - the "tier 1 release" is of ZoL, not a specific distro. The
implication is that you upgraded the machine "to a tier 1 release" sometime
in 2017. As you point out, that might not be what folks really care about,
if they get their ZoL bits from a distro. The distros can lag arbitrarily
far behind, and we have no control over them. That said, I agree that this
is a good point of discussion.
--matt
>
> Looking at now, so features=portable not features=portable-2017...
>
> For Debian, that is currently 0.6.5.9 in Stretch or 0.7.12 via
> stretch-backports; the upcoming Buster release will have 0.7.12. For
> Ubuntu, that is currently 0.6.5.6 in Xenial, 0.7.5 in Bionic with 0.7.9
> via the HWE ("backport") kernel, and 0.7.9 in Cosmic (non-LTS); the
> upcoming Disco (non-LTS) release will have 0.7.12. Gentoo and Arch seem
> to both have 0.7.13.
>
> If you expand "tier-1 platforms" to include ZoL source (plus the Fedora
> and RHEL/CentOS packages), Debian, Ubuntu, Gentoo, and Arch,
> features=portable would be as old as ZoL 0.6.5.6 (from Ubuntu Xenial) if
> you consider all supported versions, 0.6.5.9 for the latest LTS distro
> releases (from Debian Stretch), or 0.6.5.9 for the latest released
> distros ignoring LTS (from Debian Stretch).
>
> In a few months, once Ubuntu Disco and Debian Buster are released, that
> would change to ZoL 0.6.5.6 (from Ubuntu Xenial) if you consider all
> supported versions, 0.7.5 for the latest LTS distro releases (from
> Ubuntu Bionic), or 0.7.12 for the latest released distros ignoring LTS
> (from Ubuntu Disco and Debian Buster).
>
> Personally, I'd like to see the above distros included in the definition
> of tier-1 platforms, and I would use "latest released LTS version" for
> distros where LTS is relevant (e.g. Ubuntu), but not including Ubuntu's
> HWE kernel packages or Debian's backports. So that would currently be
> ZoL 0.7.5, changing to 0.7.12 later this year.
>
> Ubuntu releases an LTS version every two years, on schedule. Debian
> releases "when ready", but releases are about every two years. This is
> in line with FreeBSD releasing about every two years, so expanding the
> definition of "tier-1 platform" to include these specific Linux distros
> is not likely to significantly slow the progress of features=portable.
>
> I'm not a FreeBSD user, but a similar definition would use the latest
> FreeBSD stable release, so that'd be FreeBSD 12.0 at the moment.
>
> Considering only Linux and FreeBSD, Linux would currently be blocking
> device_removal, obsolete_counts, spacemap_v2, and zpool_checkpoint and
> FreeBSD would be blocking edonr, large_dnode, and userobj_accounting.
>
> --
> Richard
>
> ------------------------------------------
> openzfs: openzfs-developer
> Permalink:
> https://openzfs.topicbox.com/groups/developer/Td66d70a086ae5a9e-M7e1f9ec33b74af96a04e9159
> Delivery options:
> https://openzfs.topicbox.com/groups/developer/subscription
>
------------------------------------------
openzfs: openzfs-developer
Permalink:
https://openzfs.topicbox.com/groups/developer/Td66d70a086ae5a9e-M95f4f9ba0c2d96c700a92615
Delivery options: https://openzfs.topicbox.com/groups/developer/subscription
