approximating freebsd gmirror in dragonflybsd

[Roman Neuhauser]

# justin at shiningsilence.com / 2014-07-27 17:49:38 -0400:
> I've done similar - streaming to a local disk.  There's no way with
> just Hammer to have it automatically fail over; you'll have to
> manually mark the working disk as the master and throw out the failed
> former master.  For something truly automatic, you'd have to go for
> hardware RAID.  There's some things possible with software RAID -
> vinum or ccd or LVM, but you're wandering into uncharted territory
> there.

based on info from alexh re device mapper[0,1], ccd(4) and vinum(8)
i'd say i'm out of luck.

[0]
  18:52 <@alexh> mirror? it's a return-error shim.
  18:52 <@alexh> dmirror has a basic skeleton and some notes on how to
                 implement, but that's it
  18:53 <@alexh> personally I don't think that sort of raid is the way to
                 go; filesystems are a better place for that sort of thing

> On the other hand, you can stream to multiple targets.  A big enough
> power surge will kill both disks.  Streaming to a mirrored disk
> doesn't give you immediate failover, but streaming to a different
> location in addition to streaming to a local disk will make it more
> likely your data survives.

my internet connection goes down every now and then; the only
surge-related hw failure i've had was one of two onboard NICs.
that is to say: the remote streaming is very cool but solves somebody
else's problem, not mine -- or at least a different problem from the one
i'm seeking to kill here.

> A @reboot command for hammer mirror-stream in cron will suffice to
> keep it working.

what if the process crashes for whatever reason?  gmirror(4) is tied
to the kernel being up, whereas here i'll need daemontools or better.
the number of parts in the solution is much higher, so i'd expect
a corresponding hit to reliability (note that i'm talking purely about
the mirroring, not disks, networks or anything).
 

-- 
roman