[PATCH SERIES v2 pve-storage/qemu-server] add external qcow2 snapshot support

[DERUMIER, Alexandre]

Hi,

Any comment about this patch series ?

I really think that external snapshot could be a great feature (as I
still see report on the forum about freeze on snasphot deletion),
and support for lvm and shared san is really a feature than enterprise
users are waiting.  (To be honest, I have a lot of customers stuck on
vmware because of this)


About my previous patch serie, with lvm dynamic extent, I think I'll
give up, it seem to be too complex, with too many corner case.
(So keeping shared lvm + external snapshot without thin provisioning)


In Parallel, I have done more tests with gfs2/ocfs2, and I finally
found a 
way to have good performance on block allocation for thin non
preallocated qcow2 file.

Currently, I'm around 200 iops on gfs2 with 4k randwrite (instead
20000iops...)

(fio --rw=randwrite ---direct=1 -bs=4k --ioengine=libaio --iodepth=64 -
-filename=/dev/sdX)

qemu have "preallocate" filter feature
https://patchwork.kernel.org/project/qemu-devel/cover/20200814130348.20625-1-vsementsov@virtuozzo.com/

-drive driver=qcow2,file.driver=preallocate,file.prealloc-
size=1073741824,file.file.driver=file,file.file.filename=/mnt/pve/gfs2s
an/images/100/vm-100-disk-0.qcow2,id=drive-scsi2,if=none


which allow to prellocate on the fly more blocks than requested.
(for example, you need to write a 4k block on a unallocated block, I'll
reserve 128MB for example).
This reduce a lot the number of locks and round-trip network for fs
like ocfs2/gfs2  when you have a lot of write.

With qcow2 format allocating random blocks consecutively, that's
working very well.

I have done a small test, the fio result is around to 20~30k write.
I'll send a patch soon after doing more test.


Regards,

Alexandre

-------- Message initial --------
De: Alexandre Derumier <alexandre.derumier at groupe-cyllene.com>
À: pve-devel at lists.proxmox.com
Objet: [PATCH SERIES v2 pve-storage/qemu-server] add external qcow2
snapshot support
Date: 30/09/2024 13:31:50

This patch series implement qcow2 external snapshot support for files
&& lvm volumes

The current internal qcow2 snapshots have a lot of performance
problems.

I have tested through nfs and also local filesystem


I see that Fiona don't have same result than me, but I got something
like 200~300iops
vs 20000 iops with 4k randwrite when a snapshot exist.

The result is even worst on a shared filesystem like ocfs2 or gfs2.
(around 80 iops)
I think (I'm not 100% sure) this is mostly because metadatas are not
preallocated
anymore with qcow2 internal snap.

With external snapshot, I almost don't have any performance impact when
a snapshot exist.

Also other bugs are freeze/lock reported by users since years on
snapshots delete on nfs
https://antiphishing.vadesecure.com/v4?f=S1Zkd042VWdrZG5qQUxxWk5ps4t67k
NuHsBZzdzhpquLKuXqTZLIq2K1DfKr9N61yBafm7AuAITd6bHtRU4zEQ&i=MlZSTzBhZFZ6
Nzl4c3EyN5T6buHjA4kKs6Oz9IPjCIg&k=F1is&r=cm1qVmRYUWk2WXhYZVFHWA0PXtTaYx
z7-FIOTkZBm34_dHdSch-
gXn7ST9eGhQLN&s=64b60d6fd396d266b432ee693cc8f61d2632a8524491fef07cef3c3
f51c98871&u=https%3A%2F%2Fforum.proxmox.com%2Fthreads%2Fsnapshot-
removal-jams-the-vm.111648%2F
(The disk access seem to be frozen during all the delete duration)

External qcow2 snapshots also allow snapshot of raw devices ,so 0
performance impact without snapshots.

This also open doors for remote snapshot export-import for storage
replication.

This V2 introduce support for qcow2 external snapshot for lvm, extra
lvm
volume is created for each snapsphot and formated with qcow2.
This is a lot more performant than lvm (non-thin/nomedata) snapshot,
and allow to use
it for shared lvm.  (I have another patch series for thick lvm dynamic
extend, but if we could have at minimum
snapshot working, it could great :)

I have tested: snasphot, snap rollback, snap delete, clone, move disk,
rename disk, create_base. (online && offline)

lxc is not yet supported, but I think we could look to implement the
recent dm-qcow2 kernel block driver
https://antiphishing.vadesecure.com/v4?f=S1Zkd042VWdrZG5qQUxxWk5ps4t67k
NuHsBZzdzhpquLKuXqTZLIq2K1DfKr9N61yBafm7AuAITd6bHtRU4zEQ&i=MlZSTzBhZFZ6
Nzl4c3EyN5T6buHjA4kKs6Oz9IPjCIg&k=F1is&r=cm1qVmRYUWk2WXhYZVFHWA0PXtTaYx
z7-FIOTkZBm34_dHdSch-
gXn7ST9eGhQLN&s=1865f514f95ac1d8e0088b598376751d4d98fa25de6a8b2868a74f9
2ac661cfa&u=https%3A%2F%2Flore.kernel.org%2Flkml%2F164846619932.251310.
3668540533992131988.stgit%40pro%2FT%2F


storage.cfg example:

dir: local2
        path /var/liv/vz
        content snippets,vztmpl,backup,images,iso,rootdir
        snapext 1

lvmqcow2:test
        vgname test
        snapext 1
        content images


changelog v2:
 implement lvm with external qcow2 snapshots

pve-storage:

Alexandre Derumier (2):
  add external snasphot support
  add lvmqcow2 plugin: (lvm with external qcow2 snapshot)

 src/PVE/Storage.pm                |   2 +
 src/PVE/Storage/DirPlugin.pm      |   1 +
 src/PVE/Storage/LvmQcow2Plugin.pm | 460 ++++++++++++++++++++++++++++++
 src/PVE/Storage/Makefile          |   3 +-
 src/PVE/Storage/Plugin.pm         | 225 +++++++++++++--
 5 files changed, 665 insertions(+), 26 deletions(-)
 create mode 100644 src/PVE/Storage/LvmQcow2Plugin.pm


qemu-server:


Alexandre Derumier (1):
  implement external snapshot

 PVE/QemuServer.pm | 108 ++++++++++++++++++++++++++++++++++++++++------
 1 file changed, 95 insertions(+), 13 deletions(-)