[OpenIndiana-discuss] Simple zfs vs zpool space question.

[Jim Klimov]

On 2014-04-13 01:33, John Ryan wrote:
> I had a problem once where I was unable to import a pool with a missing
> log disk. I was unable to recover the pool. Now I mirror the logs

How long ago was that (or, rather, with how old a ZFS version)?
There was this "feature" of requiring logs for import initially,
but since then there appeared a command-line flag to allow pool
import without processing the logs (forfeiting "sync" transactions
that were only logged there). Probably this would indeed require
a bit of work in single-user mode (if your pools are imported by
default with /etc/zfs/zpool.cache) or soon after boot (if another
pool importing mechanism is used, such as an SMF service), since
on serious systems the serious decisions are by design delegated
to humans - even if you can in fact automate the imports with the
forcing switches in a script on your home rigs, at a risk of some
conflicts (i.e. relevant on corporate shared storage) or recently
written data loss (TXG rollback, log ignore, etc.) that you then
won't be well informed about.

Also note that if the (only) separate log device dies or is removed
while the pool is up, this should be detected by the OS pretty soon
and the logs would be written to main pool devices like they are
on a pool without a log device at all. So the "unprotected" period
in this case is relatively short, one or a few TXGs.

Of course, by Murphy's law, that is bound to happen rarely - rather,
your server goes down with a power surge from a lightning, and when
it comes up - you discover that the log device is toast. Maybe both
of them mirrored logs ;)

HTH,
//Jim Klimov