FreeBSD 8.2 Release, ZFS + Samba, running out of memory

Tue Feb 22 22:21:21 UTC 2011

On Tue, Feb 22, 2011 at 10:55:37PM +0100, Henner Heck wrote:
> 
> Hello,
> 
> i experience freezing of my FreeBSD machine when performing certain
> operations
> on a Samba share.
> 
> Technical info:
> - FreeBSD 8.2 Release 64 Bit (it also happened with 8.2 RC3)
> - Samba 3.5.6.1
> - Athlon II Quadcore, 4 GB Ram
> - 1 SSD with a ZFS pool (No.0) containing the FreeBSD system
> - 12x2TB RaidZ2 pool (No.1) for data, created on 12 GEOM eli encrypted
> partitions on 12 disks,
>     shared to a Windows 7 PC with Samba,
>     8 of the disks are attached to 2 Marvell SATA controllers, 4 to the
> onboard controller
> - ZPool v15, ZFS v4
> 
> Scenarios (checked using top):
> 
> A:
> When copying files from one directory in pool 1 to another, the free
> memory drops from
> about 3700M to abaout 200M in the process, but seems to stabilize then.
> 
> B:
> When copying the files onto a Windows machine using the Samba share,
> the free memory seems to stabilize at about 100M.
> 
> C:
> When computing a hashvalue of files from the share on Windows or doing a
> binary compare to copies of the files stored on the Windows PC (using
> Total Commander),
> the free memory on the FreeBSD machine drops even lower and shortly
> after the BSD system freezes.
> Here is the last top output i got via ssh:
> 
> /last pid:  1328;  load averages:  4.53,  2.23,  0.99    up 0+00:04:39 
> 22:07:50
> 263 processes: 43 running, 201 sleeping, 19 waiting
> CPU:  0.9% user,  0.0% nice, 23.1% system,  4.2% interrupt, 71.9% idle
> Mem: 720K Active, 516M Wired, 144K Cache, 320K Buf, *39M Free*
> Swap: 4096M Total, 12M Used, 4084M Free, 3008K In, 5124K Out
> 
>   PID USERNAME  THR PRI NICE   SIZE    RES STATE   C   TIME   WCPU COMMAND
>    11 root        4 171 ki31     0K    64K RUN     0  15:54 303.61% idle
>  1321 root        1  52    0 27812K   704K swread  1   0:24 14.26% smbd
>    12 root       19 -60    -     0K   304K WAIT    0   0:21 12.45% intr
>    16 root        1  48    -     0K    16K psleep  2   0:01  3.76%
> pagedaemon
>     3 root        1  -8    -     0K    16K RUN     0   0:06  3.27% g_up
>     4 root        1  -8    -     0K    16K -       3   0:05  2.69% g_down
>     0 root      108  -8    0     0K  1712K -       0   1:02  1.86% kernel
>     8 root        6  -8    -     0K    88K tx->tx  1   0:00  1.27% zfskern
>  1268 root        1  44    -     0K    16K geli:w  1   0:03  0.98%
> g_eli[1] gpt
>  1225 root        1  45    -     0K    16K RUN     3   0:02  0.98%
> g_eli[3] gpt
>  1267 root        1  44    -     0K    16K geli:w  0   0:02  0.98%
> g_eli[0] gpt
>  1237 root        1  44    -     0K    16K RUN     0   0:02  0.88%
> g_eli[0] gpt
>  1214 root        1  44    -     0K    16K RUN     2   0:02  0.88%
> g_eli[2] gpt
>  1244 root        1  44    -     0K    16K RUN     2   0:02  0.78%
> g_eli[2] gpt
>  1243 root        1  44    -     0K    16K RUN     1   0:02  0.78%
> g_eli[1] gpt
>  1212 root        1  44    -     0K    16K RUN     0   0:02  0.78%
> g_eli[0] gpt
>  1215 root        1  44    -     0K    16K RUN     3   0:02  0.78%
> g_eli[3] gpt
>  1213 root        1  44    -     0K    16K RUN     1   0:02  0.78%
> g_eli[1] gpt
>  1240 root        1  44    -     0K    16K RUN     3   0:02  0.78%
> g_eli[3] gpt
>  1217 root        1  44    -     0K    16K RUN     0   0:02  0.78%
> g_eli[0] gpt
>  1242 root        1  44    -     0K    16K RUN     0   0:02  0.68%
> g_eli[0] gpt
>  1238 root        1  44    -     0K    16K RUN     1   0:02  0.68%
> g_eli[1] gpt
>  1248 root        1  44    -     0K    16K RUN     1   0:02  0.68%
> g_eli[1] gpt
>  1252 root        1  44    -     0K    16K RUN     0   0:02  0.68%
> g_eli[0] gpt
>  1249 root        1  44    -     0K    16K RUN     2   0:02  0.68%
> g_eli[2] gpt
>  1269 root        1  44    -     0K    16K geli:w  2   0:02  0.68%
> g_eli[2] gpt/
> 
> It looks like a caching problem to me, but i don't know how to fix it.
> I am also a bit confused, since i don't see an obvious difference
> between scenario B and C.
> I had a similar setup with 5 disks RaidZ1 and Samba running on 8.1 Release,
> and never experienced such a freeze.
> 
> Does anyone have advice on how to get rid of this problem?
Try the patch from rev. 218795.

If it indeed help, we would need an errara notice.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 196 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-stable/attachments/20110222/eaa27fc0/attachment.pgp