If you crash, reset or lose power, ZFS can replay the ZIL from the slog device,
because it has battery/supercap etc. power-loss protection and is therefore
persistent.
If the slog device dies, all the data in the ZIL is still cached in RAM and can
be written out from there. ZFS automatically switches the ZIL from the slog to
the main pool.
If the slog device dies at the same instant you lose power, you just suffered a
drone strike and all bets are off.
Check this thread, which has some excellent comments from Richard Elling and
others:
<http://openindiana.org/pipermail/openindiana-discuss/2012-June/008249.html>
Richard also points out that we are saying ZIL when we mean slog. I stand
corrected.
Best,
Chris
Am 09.05.2014 um 13:34 schrieb Narayan Desai via smartos-discuss
<[email protected]>:
> On Fri, May 9, 2014 at 3:15 AM, Chris Ferebee via smartos-discuss
> <[email protected]> wrote:
>
> Therefore, write-endurance and power-loss protection are essential for a ZIL
> device. OTOH, if the ZIL fails, the system should keep running, just with
> reduced sync write performance. So you only need to mirror the ZIL if you
> can’t tolerate a drop in performance for the time it takes to provision a new
> ZIL device.
>
> Won't you lose data if you lose a ZIL? Presumably everything stored on it at
> failure time would be lost, right?
>
> I also vaguely recall that we've had problems with pools where the ZIL has
> failed, but I'm sure that someone on this list could definitively say one way
> or the other.
> -nld
-------------------------------------------
smartos-discuss
Archives: https://www.listbox.com/member/archive/184463/=now
RSS Feed: https://www.listbox.com/member/archive/rss/184463/25769125-55cfbc00
Modify Your Subscription:
https://www.listbox.com/member/?member_id=25769125&id_secret=25769125-7688e9fb
Powered by Listbox: http://www.listbox.com
