[OmniOS-discuss] Status of TRIM support?

[Dan Swartzendruber]

> It looks to me like Sa¨o's design is active/standby failover.  Zpool
> import on the standby should obtain a clean transaction group as long
> as the originally active system is still not using the pool.  The
> result would be similar to the power fail situation.

As long as the right fencing is done in the case where the active node
goes south, agreed.  In my case, I have 3 server, two running vsphere and
one running illumos.  All active guests run on server V1, with V2 as a HA
backup for V1.  Since V2 is doing little else, it also hosts a virtualized
illumos appliance, which currently has two 1TB disks for a hourly zfs send
replication job.  I intend to put an HBA in V2 and pass it through to the
storage appliance and go from there.  The only fly in the ointment is that
while V1 can be readily fenced using the on-board IPMI, I have no easy way
to fence the virtualized appliance.  I seem to recall seeing a vmware
fencing agent, but it may not be reliable enough for me (e.g. what if the
reason the virtualized appliance is not working properly is because the
host is wigging out?)  It struck me that since nothing else normally runs
on V2, I can fence the virtualized appliance by fencing the host it runs
on using V2's onboard IPMI.  If a hard failover needs to be done, the
standby appliance will need to import the pool with '-f', which is scary
if your fencing is not extremely reliable...