On Mon, May 2, 2011 at 6:23 AM, Steven Sim <unixan...@gmail.com> wrote:

> Firstly, can I create another filesystem (dataset) within the webpool e.g.
> say webpool/webdata and assigned it as a dataset to the webzone? Would that
> be considered a bad practice as it is a form cascading mount?

You *can* but you *shouldn't*.

In general, and I don't recall if I have seen a "Best Practice" on
this or not, if you are using a zpool for multiple datasets for an NG
Zone you should not use the dataset that is at the root of the zpool,
but build child datasets. The issue is not necessarily ZFS figuring
out how / when to mount the dependent datasets, that works fine, it is
the interaction with both the Zones infrastructure and Live Upgrade
(if you are going to ever do any of that). I have had all sorts of odd
issues with hierarchical zfs datasets and NG Zones and LU.

The layout I have been using, and have not had any issues with Live
Upgrade or patching, is the following:

zpool mountpoint=none
zpool/zfs(zone root) mountpoint=/zones/fred/root (where fred is the zonename)
zpool/zfs(application) mountpoint=/zones/fs/fred/<foo> and a lofs
mount into the NG Zone
zpool/zfs(data) mountpoint=/zones/fs/fred/<bar> and a lofs mount into
the NG Zone

The above came about after many, many different configurations were
tried (and most failed). We adopted NG Zones before they were
supported by LU, and rolled out NG Zones on ZFS before that was
supported by LU. A long, hard path to where we are now.

Paul Kraus
-> Senior Systems Architect, Garnet River ( http://www.garnetriver.com/ )
-> Sound Coordinator, Schenectady Light Opera Company (
http://www.sloctheater.org/ )
-> Technical Advisor, RPI Players
zones-discuss mailing list

Reply via email to