KVE live migration on VM with lots of diry RAM pages (takes forever)

[mailinglists]

Hi guys,

i have a situation where migration link is slower than VMs RAM, which is constantly being changed.
So when doing an online migration from 6.0 to 6.2, it never completes, just synces RAM forever and ever.

Does anyone have any ideas for online migration in this case?
VM disk in on a ZVOL.

 

[che]

Hi, what is the reason for the constantly changing memory pages? Is limiting the traffic/services or processes on the VM in order to reduce the amount of dirty pages an option?

 

[spirit]

if it can't converge, proxmox increase the final pause timeout at the end, and then retry again.
and qemu is also slowing down cpu inside the guest.

to speedup transfert, you could try to disable ssh tunnel for migration

in /etc/pve/datacenter.cfg
"migration: type=insecure"

if it's still can't migrate it, you need faster links.
(I have seen java applications doing crap with memory for example)

can you provide migration log ?

 

[mailinglists]

Hi guys, thanks for replys and suggestions.

Later I noticed, that dirty blocks are also on disk, not just in RAM and gave up, as the 1Gbps link would always be to slow.
FYI some of the VMs actually migrated after a day or two of migrating. But I was notified that live / online migration took down the VM for 30 minutes, so I stopped doing live migration.

I did an offline migration with restart and replication set beforehand and got it done very quickly. I just have had to shut down VMs, but managed to find a time frame where it was acceptable (read, during night hours .