you said "hybrid HDD"
is this possibly about write-back vs write-through cache integrity and
some confusion in a driver over what is committed back in disk, and
what is not?
this feels like a very nasty corner case. Could you be explicit about
versions and vendors?
I am asking for selfish reasons: I have a lot of dependencies in a
large SSD backed ZFS postgres server on Dell, and I am about to commit
to a lenovo X1 Carbon 7/8th gen which would be SSD and almost
certainly was intended to be ZFS-SSD in FreeBSD.
-George
On Wed, Feb 26, 2020 at 9:22 AM Mario Olofo <mario.olofo at gmail.com>
wrote:>
> Hello,
>
> I reinstalled FreeBSD 12.1 on my SSD (in the swap partition of my Linux to
> test) and on my Hybrid HDD.
>
> Just configured rc.conf to start my wifi dongle, downoaded git, node and
> npm via pkg and... as you can see in my screenshot,
> the ZFS already shows corrupted data...
>
> Can't been able to load the FreeBSD from the HDD though, don't know
why, if
> someone knows how to load the
> kernel from the HDD via loader on SSD or grub2, I can try =)
>
> Mario
>
> Em ter., 25 de fev. de 2020 ?s 20:18, Mario Olofo <mario.olofo at
gmail.com>
> escreveu:
>
> > Hello,
> >
> > I reinstalled FreeBSD 12.1 on my SSD (in the swap partition of my
Linux to
> > test) and on my Hybrid HDD.
> >
> > Just configured and rc.conf to start my wifi dongle, downoaded git,
node
> > and npm via pkg and... as you can see in my screenshot,
> > the ZFS already shows corrupted data...
> >
> > Can't been able to load the FreeBSD from the HDD though, don't
know why,
> > if someone direct me how to load the
> > kernel from the HDD via loader or grub2, I'll try =)
> >
> > Em ter., 25 de fev. de 2020 ?s 18:56, Mario Olofo <mario.olofo at
gmail.com>
> > escreveu:
> >
> >> Hello,
> >>
> >> I reinstalled FreeBSD 12.1 on my SSD (in the swap partition of my
Linux
> >> to test) and on my Hybrid HDD.
> >>
> >> Just configured and rc.conf to start my wifi dongle, downoaded
git, node
> >> and npm via pkg and... as you can see in my screenshot,
> >> the ZFS already shows corrupted data...
> >>
> >> Can't been able to load the FreeBSD from the HDD though,
don't know why,
> >> if someone direct me how to load the
> >> kernel from the HDD via loader or grub2, I'll try =)
> >>
> >> Mario
> >>
> >> Em ter., 25 de fev. de 2020 ?s 12:11, Karl Denninger <karl at
denninger.net>
> >> escreveu:
> >>
> >>>
> >>> On 2/25/2020 9:53 AM, John Kennedy wrote:
> >>> > On Tue, Feb 25, 2020 at 11:07:48AM +0000, Pete French
wrote:
> >>> >> I have often wondered if ZFS is more aggressive with
discs, because
> >>> until
> >>> >> very recently any solid state drive I have used ZFS
on broke very
> >>> quicky. ...
> >>> > I've always wondered if ZFS (and other
snapshotting file systems)
> >>> would help
> >>> > kill SSD disks by locking up blocks longer than other
filesystems
> >>> might. For
> >>> > example, I've got snapshot-backups going back, say, a
year then those
> >>> blocks
> >>> > that haven't changed aren't going back into the
pool to be rewritten
> >>> (and
> >>> > perhaps favored because of low write-cycle count). As
the disk fills
> >>> up, the
> >>> > blocks that aren't locked up get reused more and
more, leading to
> >>> extra wear
> >>> > on them. Eventually one of those will get to the point
of erroring
> >>> out.
> >>> >
> >>> > Personally, I just size generously but that isn't
always an option
> >>> for
> >>> > everybody.
> >>>
> >>> I have a ZFS RaidZ2 on SSDs that has been running for several
/years
> >>> /without any problems. The drives are Intel 730s, which Intel
CLAIMS
> >>> don't have power-loss protection but in fact appear to;
not only do they
> >>> have caps in them but in addition they pass a "pull the
cord out of the
> >>> wall and then check to see if the data is corrupted on
restart" test on
> >>> a repeated basis, which I did several times before trusting
them.
> >>>
> >>> BTW essentially all non-data-center SSDs fail that test and
some fail it
> >>> spectacularly (destroying the OS due to some of the in-flight
data being
> >>> comingled on an allocated block with something important; if
the
> >>> read/erase/write cycle interrupts you're cooked as the
"other" data that
> >>> was not being modified gets destroyed too!) -- the Intels are
one of the
> >>> very, very few that have passed it.
> >>>
> >>> --
> >>> -- Karl Denninger
> >>> /The Market-Ticker/
> >>> S/MIME Email accepted and preferred
> >>>
> >>
> _______________________________________________
> freebsd-stable at freebsd.org mailing list
> https://lists.freebsd.org/mailman/listinfo/freebsd-stable
> To unsubscribe, send any mail to "freebsd-stable-unsubscribe at
freebsd.org"