Date: Tue, 10 Sep 2024 14:02:10 -0400 From: Charles Sprickman <spork@bway.net> To: andy thomas <andy@time-domain.co.uk> Cc: Allan Jude <allanjude@freebsd.org>, freebsd-fs <freebsd-fs@freebsd.org> Subject: Re: Does a failed separate ZIL disk mean the entire zpool is lost? Message-ID: <94C1B563-5C1A-4D48-AE88-9ABEE8841D99@bway.net> In-Reply-To: <alpine.BSF.2.22.395.2409101105040.74876@mail0.time-domain.net> References: <alpine.BSF.2.22.395.2409091634020.50467@mail0.time-domain.net> <535969cf-0b0b-48ca-a163-fc238f316bb7@gmx.at> <dabea42c-65d7-40ea-bd37-840148e855c5@freebsd.org> <alpine.BSF.2.22.395.2409101105040.74876@mail0.time-domain.net>
next in thread | previous in thread | raw e-mail | index | archive | help
I don't think your data is gone!
Note that the "-m" option to ignore the log device and possibly lose a =
few transactions does NOT require a mirrored ZIL:
-m Allows a pool to import when there is a missing log
device. Recent transactions can be lost because the =
log
device will be discarded.
It seems like it's absolutely worth trying.
Charles
> On Sep 10, 2024, at 6:35=E2=80=AFAM, andy thomas =
<andy@time-domain.co.uk> wrote:
>=20
> Thank you but I'm afraid I didn't use two mirrored ZIL devices since I =
didn't know this was possible at the time I set this server up (late =
2017 and before I was even aware of the 'FreeBSD Mastery: ZFS' book!) =
And there were no spare disk bays in the server's chassis to add another =
device and at the time PCIe > nvme adapters were not available. For data =
resilience I relied on an identical mirror server in the same rack =
linked via a 2 x 10GBit/sec bonded point-to-point network link but this =
server also failed in the data centre melt-down...
>=20
> It looks like the data is now lost so I won't waste any more time =
trying to recover it - this incident will hopefully persuade my employer =
to heed advice given years ago regarding locating mirror servers in a =
different data centre linked by a fast multi-gigabit connection.
>=20
> Andy
>=20
> PS: the ZFS and Advanced ZFS books are truly excellent, by the way!
>=20
> On Mon, 9 Sep 2024, Allan Jude wrote:
>=20
>> As the last person mentioned, you should be able to import with the =
-m flag, and only lose about 5 seconds worth of writes.
>>=20
>> The pool is already partially imported at boot by the other =
mechanisms, you might need to disable that to prevent the partial import =
at boot, so you can do the manual import.
>>=20
>> On 2024-09-09 12:20 p.m., infoomatic wrote:
>>> did you use two mirrored ZIL devices?
>>> You can "zpool import -m", but you will probably be confronted with =
some
>>> errors - you will probably lose the data the ZIL has not committed, =
but
>>> most of your data in your pool should be there
>>> On 09.09.24 17:51, andy thomas wrote:
>>>> A server I look after had a 65TB ZFS RAIDz1 pool with 8 x 8TB hard =
disks
>>>> plus one hot spare and separate ZFS intent log (ZIL) and L2ARC =
cache
>>>> disks that used a pair of 256GB SSDs. This ran really well for 6 =
years
>>>> until 2 weeks ago, when the main cooling system in the data centre =
where
>>>> it was installed failed and the backup cooling system failed to =
start up.
>>>> The upshot was the ZIL SSD went short-circuit across its power
>>>> connector, shorting out the server's PSUs and shutting down the =
server.
>>>> After replacing the failed SSD and verifying all the spinning hard =
disks
>>>> and the cache SSD are undamaged, attempts to import the pool fail =
with
>>>> the following message:
>>>> NAME SIZE ALLOC FREE CKPOINT EXPANDSZ FRAG CAP =
DEDUP
>>>> HEALTH ALTROOT
>>>> clustor2 - - - - - - - -
>>>> UNAVAIL -
>>>> Does this mean the pool's contents are now lost and unrecoverable?
>>>> Andy
>>=20
>>=20
>=20
>=20
> ----------------------------
> Andy Thomas,
> Time Domain Systems
>=20
> Tel: +44 (0)7866 556626
> http://www.time-domain.co.uk
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?94C1B563-5C1A-4D48-AE88-9ABEE8841D99>