[OmniOS-discuss] [discuss] crash dump speed painfully slow

[Richard Elling]

On Nov 5, 2014, at 12:42 PM, Schweiss, Chip via illumos-discuss <discuss at lists.illumos.org> wrote:

> 
> On Wed, Nov 5, 2014 at 2:36 PM, Dan McDonald <danmcd at omniti.com> wrote:
> 
> > On Nov 5, 2014, at 3:31 PM, Schweiss, Chip via illumos-discuss <discuss at lists.illumos.org> wrote:
> >
> > I had a system crash this morning when I/O on a pool was hung with failmode=panic.
> 
> What version of OmniOS are you running?
> 
> r1510010
>  
> 
> > That was 7 1/2 hours ago and it's at 64%.    I'm going to let it complete so the failure can be analyzed more, but I really need to find a way to significantly speed up the crash dumps.
> >
> > This system has 256GB ram and a dedicated dump pool of two 320GB mirrored SATA disks.
> >
> > I'm aware of the multi-threaded dump issue so I have these to lines in /etc/system
> >
> > set dump_plat_mincpu=0
> 
> That should be the default now.  I can't remember which rev of OmniOS it showed up in, but certainly it's in 010 or later.
> 
> > It's not clear where the bottleneck is.  I don't think it's the disk.  The disk activity lights are a light flicker, by no means being hit hard.  Unless it's a serial latency issue.
> >
> > What would be my best approach to get good dumps from a system like this in minutes, not hours?   Assuming that is possible.
> 
> You weren't dumping to the same pool or the same controller where you had the I/O hang, were you?
> 
> No, I setup a dedicated dump pool to a pair of mirrored SATA disks connected to the motherboard AHCI controller. 

Try dumping to a raw slice instead of ZFS.
 -- richard

> 
> -Chip
> 
> Dan
> 
> 
> illumos-discuss | Archives  | Modify Your Subscription	 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://omniosce.org/ml-archive/attachments/20141105/4fa05b63/attachment.html>