Question about ZFS with log and cache on SSD with GPT

Freddie Cash fjwcash at gmail.com
Sat Jan 21 23:02:24 UTC 2012


On Jan 21, 2012 2:06 PM, "Alexander Leidinger" <Alexander at leidinger.net>
wrote:
>
> On Sat, 21 Jan 2012 19:18:31 +0100 Willem Jan Withagen
> <wjw at digiware.nl> wrote:

> > Our are you more concerned out losing the the LOG en thus possible
> > losing data.
>
> Yes. If one piece of the involved hardware dies, you lose data.

To clarify this a bit.  You will only lose data is:
  - data is written to ZIL device
  - entire system crashes before the data in the ZIL is written to disk
  - ZIL device is not available at pool import time

If you write data to the ZIL, then the system crashes *but all data in ZIL
is already written to the pool*, and the ZIL device is not available at
pool import time, then no data is lost, and the pool import will continue
without the separate ZIL.

ZFS pools prior to ZFSv19 could not cope with a missing ZIL device at pool
import time, so those pools were effectively lost. The only way to recover
data was through some finicky zdb stuff. Thus, you had to use a mirrored
log device to mitigate this risk.

Since ZFSv19,  though, a pool can be imported with a faulted ZIL device,
and carry on. Only data that's only in the ZIL, not written to the pool, is
lost.


More information about the freebsd-fs mailing list