Re: Sudden zpool checksums errors

On 9/9/25 05:30, Andrea Venturoli wrote:
> On 9/9/25 01:54, David Christensen wrote:
> 
> 
>> What settings did you use?
> 
> The default ones: I just installed scterc-0.0.1 and added the following 
> to /etc/rc.conf:
> scterc_enable="YES"
> scterc_disks="da0 da1 da2 da4 da5 da6 da8 da9 da10 da11 da12 da13"
> 
> Again: it's wholly possible this is not the thing that fixed it; there 
> was a construction site nearby, so it's also possibly that some 
> vibrations were to blame.
> 
>   bye & Thanks
>      av.


Okay -- 7.0 read timeout, 7.0 second write timeout:

https://www.freshports.org/sysutils/scterc/

pkg-message:

   # specify read and write timeouts in tenths of second
   # here each is set to 7.0 seconds, which is also the default
   # for which you may omit these lines
   scterc_read_timeout="70"
   scterc_write_timeout="70"


The fact that your ZFS problems on the disks without ERC went away when 
you added scterc could be that ZFS implements a longer timeout on read 
and write calls, ZFS was faulting disks that hit the ZFS timeout, and 
that ZFS is more forgiving if the kernel times out sooner (?).


For my SOHO application, I use tower cases with vibration-isolated 
internal 3.5" drive bays.  I choose this primarily to reduce noise -- 
direct-to-metal disk bays transmit disk vibration to the case, which 
acts as a sounding board.  But, the vibration isolation should also help 
with drive-to-drive vibrations (e.g. RAID) and with external-to-drive 
vibrations (e.g. construction site):

https://www.fractal-design.com/products/cases/define/define-r5/black/


David