git.net

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: KVM NFS template image


Thanks Andrija, will try to work with the solution proposed.

Ran


On 11/22/2018 02:25 PM, Andrija Panic wrote:
I confirm bug...tmpl not removed...

And I confirm a possible solution:

"volumes" table, field "template_id" should be set to NULL for this
particular volume,  after volume restored from snapshot - on next storage
scavenger run it will be marked properly for GC and removed...
(volumes of VM deployed from ISO file, also have NULL for "template_id"
filed"







On Thu, 22 Nov 2018 at 22:29, ran huang <ran.huang134@xxxxxxxxx> wrote:

Hi Andrija,

That is precisely the step I went through.

However the template was not cleaned up after expected interval when no
other volume have it as a backing image.

regards,
Ran
On 11/22/2018 12:53 PM, Andrija Panic wrote:
Hi Run,

not sure what you mean  (I did not quite understand your explanation) -
but
here is an exercise from my side (just done it):

https://pasteboard.co/HOowNao.png

Check the image - explanation below:


Centos55 minimal (builtin) template, spin new VM:
--- new volume created with UUID/PATH (name on NFS files
system): 021e8602-235b-4e0d-b9e4-04f0ff46399f
--it's backing file: backing file:

/mnt/63a3ae7b-9ea9-3884-a772-1ea939ef6ec3/93682641-e7f6-11e8-8f64-089e01d943be
Create snapshots via GUI - there is qcow2 snapshots created, whole
snapshot
copied over (converted via qemu-img - "ps aux | grep qemu-img") tool to
Secondary NFS Storage - and snapshot REMOVED from original volume on
Primary NFS Storage (qemu-img snapshot -l
021e8602-235b-4e0d-b9e4-04f0ff46399f shows zero snaps after ACS has
finished creating snapshots)
Volume still points to it's backing file - no changes so far (as
expected)
Then I restore volume from snapshots.
CloudStack will (my conclusions from the exercise), remove original
volume
on NFS Primary Storage (021e8602-235b-4e0d-b9e4-04f0ff46399f), then it
will
copy back (convert via qemu-img) a qcow2 file from Secondary Storage back
to Primary Storage - but it will use SAME NAME, so you again see
021e8602-235b-4e0d-b9e4-04f0ff46399f on your NFS mount point.

This time when you check the image with qemu-img info - it will show it
has
NO backing file at all - since it's brand new volume/qcow2 image created
(as a copy fom Secondary Storage)

that is how it works

I assume the template will be again cleaned-up/removed from Primary
Storage
if no other VMs/volume use it as it's backing (parent) image.

Makes sense ?

Cheers

On Thu, 22 Nov 2018 at 21:18, ran huang <ran.huang134@xxxxxxxxx> wrote:

Thanks Andrija, just tested myself with expunge and works as expected.

However, for KVM, when I revert a qcow disk from snapshot, which removes
the backing chain to template, the template will not be removed.

So it seems like despite the qcow disk is no longer backed by the
template, the template will still consider the disk as its children in
this case(revert from snapshot).

regards,
Ran

On 11/19/2018 10:43 AM, Andrija Panic wrote:
new template, deployed new VM, destroyed VM (with Exunge option)...

up to 120sec later... (storage.cleanup.interval=120 sec,  global config
option)

2018-11-19 19:35:59,525 DEBUGStorage pool garbage collector found 1
templates to clean up in storage pool: Primary-storage - NFS
2018-11-19 19:35:59,525 DEBUG [c.c.s.StorageManagerImpl]
(StorageManager-Scavenger-1:ctx-2c88c8e0) (logid:040f4ad1) Storage pool
garbage collector has marked template with ID: 219 on pool 4 for
garbage
collection

Another  120sec later... (storage.cleanup.delay=120sec)

2018-11-19 19:37:59,598 DEBUG [c.c.s.StorageManagerImpl]
(StorageManager-Scavenger-2:ctx-f9dd338d) (logid:9ae40975) Storage pool
garbage collector found 1 templates to clean up in storage pool:
Primary-storage - NFS
...
2018-11-19 19:37:59,638 DEBUG [c.c.t.TemplateManagerImpl]
(StorageManager-Scavenger-2:ctx-f9dd338d) (logid:9ae40975) Evicting
TmplPool[37-219-4-563ea0f5-5164-4ac4-a183-728f418269b7]
2018-11-19 19:37:59,643 DEBUG [c.c.h.o.r.Ovm3HypervisorGuru]
(StorageManager-Scavenger-2:ctx-f9dd338d) (logid:9ae40975)
getCommandHostDelegation: class
com.cloud.agent.api.storage.DestroyCommand
...
2018-11-19 19:37:59,665 DEBUG [c.c.t.TemplateManagerImpl]
(StorageManager-Scavenger-2:ctx-f9dd338d) (logid:9ae40975) Successfully
evicted template andrija-test-tmpl from storage pool null

template "andrija-test-tmpl" deleted...

Hope that helps Run.

Cheers
Andrija

On Mon, 19 Nov 2018 at 19:11, Andrija Panic <andrija.panic@xxxxxxxxx>
wrote:
True (at least I'm sure for SolidFire) - but I believe in general also
(will test this now...)

On Mon, 19 Nov 2018 at 18:51, Dag Sonstebo <
Dag.Sonstebo@xxxxxxxxxxxxx>
wrote:

Developers please correct me... but as far as I remember there is a
garbage collector which does remove the templates from primary
storage
once
they are not needed (i.e. have no more "child VMs"). This is
controlled by
the global setting "storage.template.cleanup.enabled".

Regards,
Dag Sonstebo
Cloud Architect
ShapeBlue


On 16/11/2018, 22:51, "ran huang" <ran.huang134@xxxxxxxxx> wrote:

       Hi Andrija,

       Thanks for the clarification and quick response

       regards,
       Ran

       On 11/16/2018 02:15 PM, Andrija Panic wrote:
       > Hi Ran,
       >
       > templates stays on Primary Storage "forever", at least for
NFS
(they are
       > moved from Secondary to Primary when you deploy a very first
VM
from
       > specific template). All VMs have this templates qcow2 as
baking
(parent)
       > image.
       >
       > This template is a qcow2 copy of a file from Secondary
Storage -
and is
       > considered a "parent" image, from which all child images  (VM
volumes) are
       > created - as you stated baking file (qcow linked clones, in
official
       > terminology)
       >
       > you can have i.e. 100 VMs all linking (having it's backing
file...)
to a
       > template qcow2 file.
       > So in other words, it's not supposed to be removed.
       >
       > Does this make sense?
       >
       > Cheers
       >
       >
       >

Dag.Sonstebo@xxxxxxxxxxxxx
www.shapeblue.com
Amadeus House, Floral Street, London  WC2E 9DPUK
@shapeblue



On Fri, 16 Nov 2018 at 22:38, ran huang <ran.huang134@xxxxxxxxx>
wrote:
       >
       >> Greetings All,
       >>
       >> For qcow2 format images, when creating a new VM in KVM, the
template
       >> image is copied from secondary storage to primary storage,
and
the
root
       >> volume image is created with the template image as a backing
file.
       >>
       >> But when I break this backing chain on primary(expunge VM or
revert to a
       >> snapshot previously created on the root volume image), the
template
       >> image is not deleted.
       >>
       >> Might I ask how is the template image going to be cleaned
from
the
       >> primary storage?
       >>
       >>
       >> addendum:
       >> CS ver 4.9.2 on CentOS 7.2
       >>
       >> regards,
       >> Ran
       >>
       >




--

Andrija Panić