[pve-devel] [PATCH v3 qemu-server 0/2] remote-migration: migration with different cpu

[Alexandre Derumier]

This patch series allow remote migration between cluster with different cpu model.

2 new params are introduced: "target-cpu" && "target-reboot"

If target-cpu is defined, this will replace the cpu model of the target vm.

If vm is online/running, an extra "target-reboot" safeguard option is needed.
Indeed, as the target cpu is different, the live migration with memory transfert
is skipped (as anyway, the target will die with a different cpu).

Then, after the storage copy, we switch source vm disk to the targetvm nbd export,
then shutdown the source vm and restart the target vm.
(Like a virtual reboot between source/target)



Changelog v2:

The first version was simply shuting down the target vm,
wihout doing the block-job-complete.

After doing production migration with around 400vms, I had
some fs corruption, like some datas was still in buffer.

This v2 has been tested with another 400vms batch, without
any corruption.


Changelog v3:

v2 was not perfect, still have some 1 or 2 fs corruption with vms doing
a lot of write.

This v3 retake idea of the v1 but in a cleaner way

- we migrate disk to target vm
- source vm is switching disk to the nbd of the target vm. 
  (with a block-job-complete, and not a block-job-cancel with standard disk migration).
  We are 100% sure it that no pending write is still pending in the migration job.
- source vm is shutdown
- target with is restart


We have redone a lot of migration this summer( maybe another 4000vm),
0 corruption, windows or linux guest vms.



Alexandre Derumier (2):
  migration: move livemigration code in a dedicated sub
  remote-migration: add target-cpu && target-reboot params

 PVE/API2/Qemu.pm   |  28 ++-
 PVE/CLI/qm.pm      |  11 ++
 PVE/QemuMigrate.pm | 451 ++++++++++++++++++++++++---------------------
 3 files changed, 281 insertions(+), 209 deletions(-)

-- 
2.39.2