On Tue, Jun 12, 2007 at 02:33:28PM -0400, Bill Sommerfeld wrote:
> 
> am I correct in assuming that this would permit full operation of ZFS
> (in some "custom scenarios" -- clearly startup scripts would need a
> workover) without ever needing /etc/zfs/zpool.cache to exist or be
> writeable?
> 

Almost, but not quite.  With this change, it is possible for arbitrary
consumers to maintain their own copies of the pool configuration (as
some for of nvlist).  This will work for simple scenarios, but for
complex configuration where asynchronous changes are made to the config
(such as replacing disks, hot spares, etc), it would be difficult for
such a consumer to remain in sync.

One of the features we have talked about is the ability to expressly set
a particular cache file location on a per-pool basis.  This would both
allow consumers to control when and how such a pool is loaded, as well
as keep the kernel state in sync with the on-disk state.  Such a feature
was not included with this proposal because there are some non-trivial
implementation issues to sort out, there is little current demand for
it, and it is a logical follow-on once this case has been approved and
integrated.

- Eric

--
Eric Schrock, Solaris Kernel Development       http://blogs.sun.com/eschrock

Reply via email to