Re: 7-BETA3 everyday reboot

Alexandre Biancalana wrote:
> On Nov 30, 2007 9:48 AM, Gary Jennejohn <gary.jennejohn_at_freenet.de> wrote:
>   
>> On Fri, 30 Nov 2007 07:47:51 -0300
>> "Alexandre Biancalana" <biancalana_at_gmail.com> wrote:
>>
>>     
>>> After the reboot, the kernel running have the patch suggested by
>>> ivoras_at_ (http://people.freebsd.org/~pjd/patches/vm_kern.c.2.patch),
>>> but how the patch is related to "kmem_map too small", I don't think
>>> that will be usefull with this panic.
>>>
>>> I *really* need some solution :(
>>>
>>>  Any ideas ?
>>>
>>>       
>> The patch is potentially useful because the new code tries harder to
>> reclaim pages (8 times instead of once with a sleep in between the
>> attempts).
>>
>> The idea is that pages may become available at some time during one
>> of the sleeps.
>>
>> Just try it and see whether it helps. Can't do any harm.
>>     
>
> After apply the patch the machine survives to 3 days of work, but
> paniced again after the system start using swap because some
> applications were using more memory, here is the panic message:
>
> Dec  4 03:12:33 Manny syslogd: kernel boot file is /boot/kernel/kernel
> Dec  4 03:12:33 Manny kernel: panic: vm_fault: fault on nofault entry,
> addr: fffffffff7a3e000
> Dec  4 03:12:33 Manny kernel: cpuid = 0
> Dec  4 03:12:33 Manny kernel: Uptime: 3d5h5m25s
> Dec  4 03:12:33 Manny kernel: Physical memory: 3061 MB
> Dec  4 03:12:33 Manny kernel: Dumping 1788 MB: 1773 1757 1741 1725
> 1709 1693 1677 1661 1645 1629 1613 1597 1581 1565 1549 1533 1517 1501
> 1485 1469 1453 1437 1421 1405 138
> 9 1373 1357 1341 1325 1309 1293 1277 1261 1245 1229 1213 1197 1181
> 1165 1149 1133 1117 1101 1085 1069 1053 1037 1021 1005 989 973 957 941
> 925 909 893 877 861 845 829 813
> 797 781 765 749 733 717 701 685 669 653 637 621 605 589 573 557 541
> 525 509 493 477 461 445 429 413 397 381 365 349 333 317 301 285 269
> 253 237 221 205 189 173 157 141 12
> 5 109 93 77 61 45 29 13
> Dec  4 03:12:33 Manny kernel: Dump complete
> Dec  4 03:12:33 Manny kernel: Automatic reboot in 15 seconds - press a
> key on the console to abort
> Dec  4 03:12:33 Manny kernel: Rebooting...
> Dec  4 03:12:33 Manny kernel: cpu_reset: Stopping other CPUs
>
> Any other ideas ?!
>   
I have found that turning off zil and prefetch seem to keep things 
happier on one of the heavily loaded servers that I look after.
It also appears to prevent a deadlock under very heavy load - something 
I've not yet had time to debug.
Try adding:

vfs.zfs.zil_disable=1
vfs.zfs.prefetch_disable="1"

to /boot/loader.conf

and let us know if it makes a difference.

Cheers,
    Benjamin