[PVE-User] KVM tun device dying when reciving/writing large streams of data

Lindsay Mathieson lindsay.mathieson at gmail.com
Thu Oct 22 14:42:32 CEST 2015


And more :) console reboot doesn't work, I have to use:



echo b > /proc/sysrq-trigger

to do a hard reset


On 22 October 2015 at 22:37, Lindsay Mathieson <lindsay.mathieson at gmail.com>
wrote:

> Also, and more seriously, its seems to bolix the cluster - all nodes start
> showing with red dots and the VM lists doesn't show any names. So far the
> only fix I have is to reboot every node.
>
> Oddly it only seems that pve services are effected. ceph etc keeps
> working, I can still ssh onto each node.
>
> On 22 October 2015 at 21:56, Lindsay Mathieson <
> lindsay.mathieson at gmail.com> wrote:
>
>> Proxmox 3.4 cluster with 3.10
>>
>> I'm currently trying to test gluster 3.7.5 using 3 debian 8.2 VM's. The
>> VM's are setup with 3.10.0-13-pve kernel
>> - kernel 4.2
>> - Gluster 3.7.5
>> - A replica 3 sharded datastore
>> - 1 virtual nic per virtual gluster node
>>
>> The gluster datastore is exposed via gluster NFS to prxomox storage so I
>> can test running VM's off it.
>>
>> But whenever I try to move a VM disk to the gluster storage (via the NFS
>> share) the nic on the VM just stops responding and its IP drops off the
>> network. Sometime its only happens after 23GB, once got as far as 30GB.
>>
>>  However the VM is still there and I can use the novnc console to access
>> it and logon. Its still thinks its nic is working but it can't ping
>> anything.
>>
>>
>> I've tried with VIRTIO and e1000.
>>
>> Any suggets as to what to look at next?
>>
>>
>> --
>> Lindsay
>>
>
>
>
> --
> Lindsay
>



-- 
Lindsay
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://pve.proxmox.com/pipermail/pve-user/attachments/20151022/fa8f7a90/attachment-0015.html>


More information about the pve-user mailing list