BUG: ZFS replication hangs the VM after qemu fs-freeze (only when guest is CloudLinux/WHM)

[mailinglists]

Hi,

i was doing routine offline migration of VMs. For each and every VM it worked except for CloudLinux VM.
After I set up replication this is what happens.

Log on host:

2019-02-28 13:59:01 108-0: start replication job
2019-02-28 13:59:01 108-0: guest => VM 108, running => 22427
2019-02-28 13:59:01 108-0: volumes => local-zfs:vm-108-disk-1

log of replication stops here. Inside the VM, the filesystem gets frozen, like one can not to even touch somefile.
VM can not be cleany shut down, can do nothing. qm agent 108 ping, etc... returns nothing.
This is how it stays untill I kill / stop the VM using PM and then replication continues.

2019-02-28 13:59:02 108-0: freeze guest filesystem
2019-02-28 14:01:17 108-0: create snapshot '__replicate_108-0_1551358741__' on local-zfs:vm-108-disk-1
2019-02-28 14:01:17 108-0: thaw guest filesystem
2019-02-28 14:01:17 108-0: incremental sync 'local-zfs:vm-108-disk-1' (__replicate_108-0_1551358138__ => __replicate_108-0_1551358741__)

The PM hosts tested were the latest from no subs repo: pve-manager/5.3-11/d4907f84 (running kernel: 4.15.18-11-pve) as well as some older versions: pve-manager/5.2-1/0fcd7879 (running kernel: 4.15.17-1-pve).

Guest is CloudLinux release 7.6 (Vladimir Lyakhov) 3.10.0-962.3.2.lve1.5.24.8.el7.x86_64, qemu-guest-agent.x86_64 10:2.12.0-2.el7.

I can reproduce it every time.
Where should I open bug report?

 

[wolfgang]

Hi,

I would say this is a kernel bug of Cloudlinux and has nothing to do with the replication.
Or maybe qemu-guest-agent.

 

[mailinglists]

So you think I should report it to CloudLinux?

 

[wolfgang]

To prove my thoughts you can deinstall qemu-guest-agent and try again.
In this case, I guess you will have no problems.
And it would prove the problem is at the freeze-fs what is done by the qemu-guest-agent what comes from CloudLinux.

 

[mailinglists]

If anyone hits this bug, there is a workaround to disable secure tmp in WHM and it will not freeze anymore.