[OpenIndiana-discuss] HDD Upgrade problem

[Jim Klimov]

2012-06-10 18:45, michelle wrote:
> The system seems to have hung again, any query to the ZFS system hangs
> the session.
>
> Nothing in /var/adm/messages.
>
> I'm wondering whether the combination of having a mirrored zfs set with
> one drive on e-sata and one drive on USB could be upsetting it.



The situation sounds reminiscent of my large data deletions on a
deduped pool, when the system was running out of kernel memory and
collapsed into a "scanrate-hell" while processing deferred-frees
of zfs blocks. You can monitor this with "vmstat 1" in an SSH
terminal, if this condition is the case, the "free" memory will
go under 32Mb (or 128Mb in some cases), and the "sr" column may
reach millions. I am not sure if there's much one can DO about
this, except reset the box and wait - my deferred-frees cleared
up after a couple of weeks...

I've made a "freeram-watchdog" program which tries to catch the
condition and reset the system before it is too late, but it
often misses the moment (kstat checks can't be more frequent
than 1 second, and this is sometimes enough to bring the box down).
I later learned of the "deadman timer" but haven't got to test
if it applies in this case. See here:

http://wiki.illumos.org/pages/viewpage.action?pageId=1146983
http://81.5.113.5/~jim/freeram-watchdog-20110531-smf.tgz

HTH,
//Jim Klimov