PVE 4.0 Beta 2: Difficulties with zfs {Large kmem_alloc}

[windinternet]

Hello all,

We have some intermittent crashes on a PVE 4.0 Beta 2 server. Server loses connectivity and kernel log shows a message that is correlated in time. We run ZFS with a raid10 setup done with the installer.

This message shows up in kern.log:

Large kmem_alloc(35496, 0x1000), please file an issue at:
CPU: 3 PID: 251 Comm: zvol Tainted: P O 4.2.0-1-pve #1
Hardware name: Supermicro X10SL7-F/X10SL7-F, BIOS 2.00 04/24/2014
0000000000000000 ffff8807ef8abbd8 ffffffff817c76d3 ffff88081fcd1178
000000000000c210 ffff8807ef8abc28 ffffffffc00d6c39 0000000000000017
00000000f7f59a98 ffff8807ef8abc18 ffff8807faaa6b88 00000005f4adf000
Call Trace:
[<ffffffff817c76d3>] dump_stack+0x45/0x57
[<ffffffffc00d6c39>] spl_kmem_zalloc+0x159/0x1c0 [spl]
[<ffffffffc01b2df1>] dmu_buf_hold_array_by_dnode+0xa1/0x4b0 [zfs]
[<ffffffffc01b32dd>] dmu_buf_hold_array+0x5d/0x80 [zfs]
[<ffffffffc01b4585>] dmu_write_req+0x65/0x1c0 [zfs]
[<ffffffffc0254ea7>] zvol_write+0x117/0x430 [zfs]
[<ffffffffc00da2b1>] taskq_thread+0x221/0x450 [spl]
[<ffffffff8109f530>] ? wake_up_q+0x70/0x70
[<ffffffffc00da090>] ? taskq_cancel_id+0x110/0x110 [spl]
[<ffffffff810947ab>] kthread+0xdb/0x100
[<ffffffff810946d0>] ? kthread_create_on_node+0x1c0/0x1c0
[<ffffffff817ce55f>] ret_from_fork+0x3f/0x70
[<ffffffff810946d0>] ? kthread_create_on_node+0x1c0/0x1c0

It seems as though this particular error may have been fixed upstream in ZFS on Linux at 28^th august with a fix for issue 3684.

Did anyone else see this error and know how to handle it?

 

[tom]

update to stable 4.0 (including zfs updates) and try again.

 

[windinternet]

Hello Tom,

Thanks for the reply. Could you confirm that this commit made it in into PVE 4 release?

Dietmar Maurer [Thu, 24 Sep 2015 10:47:22 +0000]
update pkg-zfs to master/debian/jessie/0.6.5.1-4

Best regards,
Gerrit Venema
WIND Internet

 

[dietmar]

we already updated to master/debian/jessie/0.6.5.2-2

 

[windinternet]

Hello Dietmar,

Ok, great. You guys are really up to date

We will update also...

Best regards,
Gerrit