In message <
, Andrey Fesenko writes:
> On Sat, May 6, 2017 at 2:52 AM, Ngie Cooper <> wrote:
> > On Fri, May 5, 2017 at 4:43 PM, Cy Schubert <> wrot
> e:
> >
> > ...
> >
> >> You have a bad DIMM. I had this same problem on my laptop but not on my
> >> servers downstairs. That suggested that since all four machines were
> >> running the same software the difference between them was hardware.
> >> Replacing the memory in my laptop made this problem go away.
> >>
> >> I have a question for you. Do you use ZFS? ZFS exercises memory quite
> >> aggressively. I also had this problem when I replaced my UFS filesystems
> >> with ZFS on my testbed many moons ago. It even suffered random kernel
> >> panics. Here again, replacing the memory resolved the issue.
> >
> > We need more information first before saying "bad hardware" -- in
> > particular, was the machine overtaxed, were the input files proper,
> > etc?
> >
> > I'm asking because clang has a number of bugs in bugzilla where the
> > host ran out of memory trying to compile things and clang didn't fail
> > gracefully when allocating memory, handling inputs, etc.
> >
> > Thanks,
> > -Ngie
> Sorry, really my root ZFS pool is corrupt :(

That was another symptom of the problem I had on my testbed when it 
suffered random kernel panics due to marginal RAM. It was OK with UFS but 
inadequate for ZFS. My zpools became so horribly corrupt that nothing could 
correct them and at one point I couldn't even retrieve anything from one of 
the zpools. (Good thing it was testbed and that the data was replicated on 
my prod machine.) It was at that point I ran some RAM tests using 
memtest86+ from ports. It found no bad RAM. Not trusting it, I replaced all 
the RAM in that machine anyway. That was a few years ago and I haven't had 
a problem since.

There is a reason it is recommended that one use ECC corrected memory with 
ZFS. ZFS error correction is only as good as the installed RAM.

Cy Schubert <>
FreeBSD UNIX:  <>   Web:

        The need of the many outweighs the greed of the few.

_______________________________________________ mailing list
To unsubscribe, send any mail to ""

Reply via email to