Mailing-List: contact dev-help@cloudstack.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@cloudstack.apache.org
Received-SPF: neutral (nike.apache.org: local policy)
Message-ID: <53D20332.7020101@widodh.nl>
Date: Fri, 25 Jul 2014 09:11:46 +0200
From: Wido den Hollander <wido@widodh.nl>
User-Agent: Mozilla/5.0 (X11; Linux x86_64;
 rv:31.0) Gecko/20100101 Thunderbird/31.0
MIME-Version: 1.0
To: dev@cloudstack.apache.org
Subject: Re: [CLOUDSTACK-6181] RBD Support not implemented
References: 
 <CACKLxkCR2noAh18x6e0xi+PK1ZGSHEj2DtGKLGHOH+3XUsNtDg@mail.gmail.com>
	<53CEC512.9080908@widodh.nl>
	<CACKLxkCQWo_c=pZnhccFC_VtaNJatq068Y5E0wzS3Fy9Nn+roQ@mail.gmail.com>
	<53CEC9EF.1000304@widodh.nl>
	<CACKLxkD1mQBTJx7naR5CmCUUnqZvPJLSvETfabALD3WqxDOcUQ@mail.gmail.com>
	<CACKLxkD5-1ZJcJBt+imWjouBjMfRcUmpk9bzQ0sEqWskvG85LQ@mail.gmail.com>
	<CACKLxkAr+iMGTPY6Ojc4R1TzA8ww-7cYW+JpXMmhM2jieFPAHw@mail.gmail.com>
 <CACKLxkACHL8UigtVdPWFUSDG1+9XsBNJAwww=h7baAHq+-H-jg@mail.gmail.com>
In-Reply-To: 
 <CACKLxkACHL8UigtVdPWFUSDG1+9XsBNJAwww=h7baAHq+-H-jg@mail.gmail.com>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: 7bit


On 07/24/2014 06:38 PM, Logan Barfield wrote:
> I think I've found the cause of the resize failure.  When using RBD for
> primary storage the disk is not allocated before the VM starts.  The resize
> operation runs before the disk exists, and instead of failing completely it
> just updates the database with the new disk size and returns success.
>   Deploying the VM in a stopped state doesn't work either, because the disk
> isn't created until the VM starts.
>

Ah, ok. So that's something that has to be fixed there.

> I've been able to get a resize operation to work by manually running
> deployVM -> wait for ~5 seconds -> resizeVolume.  This obviously doesn't
> help when passing the 'rootdisksize' parameter to deployVM.
>

True, true. Weird however that the resize runs prior to the creation of 
the drive.

> To fix this I think the deployment logic would need to be changed to
> allocate the VM's disk before starting it, or to run the resize operation
> once the disk has been created.  I'm assuming the only reason this issue is
> appearing with RBD is due to how templates are deployed from an RBD
> snapshot instead of being copied from secondary storage every time.
>
> Thoughts?
>

Fully agree with you. I think the logic sounds sane.

Wido

>
> Thank You,
>
> Logan Barfield
> Tranquil Hosting
>
>
> On Wed, Jul 23, 2014 at 7:13 PM, Logan Barfield <lbarfield@tqhosting.com>
> wrote:
>
>> The change I mentioned appears to fix the resize error.  Resizing the root
>> disk with deployVM still doesn't work however.
>>
>> To test I created a VM from a 5GB template with "rootdisksize" = "20".
>>   The VM is successfully created, and CloudStack lists the ROOT volume as
>> 20GB, but the RBD volume is still 5GB in size.
>>
>> Resizing the volume manually via virsh seems to work fine.  I'll look
>> through it again tomorrow.
>>
>>
>> Thank You,
>>
>> Logan Barfield
>> Tranquil Hosting
>>
>>
>> On Wed, Jul 23, 2014 at 5:49 PM, Logan Barfield <lbarfield@tqhosting.com>
>> wrote:
>>
>>> Just an FYI, RBD resizing still fails because libvirt is throwing an
>>> invalid flags error.  I'm testing the following patch:
>>>
>>>
>>> plugins/hypervisors/kvm/src/com/cloud/hypervisor/kvm/resource/LibvirtComputingResource.java:
>>>
>>> -                    if (conn.getLibVirVersion() > 1001000 &&
>>> vol.getFormat() == PhysicalDiskFormat.RAW) {
>>> +                    if (conn.getLibVirVersion() > 1001000 &&
>>> vol.getFormat() == PhysicalDiskFormat.RAW && pool.getType() !=
>>> StoragePoolType.RBD) {
>>>                            flags = 1;
>>>                        }
>>>
>>> If it resolves the issue I'll submit it.
>>>
>>>
>>>
>>>
>>> Thank You,
>>>
>>> Logan Barfield
>>> Tranquil Hosting
>>>
>>>
>>> On Tue, Jul 22, 2014 at 4:38 PM, Logan Barfield <lbarfield@tqhosting.com>
>>> wrote:
>>>
>>>> Wido,
>>>>
>>>> Excellent.  As long as it passes testing we'll be golden.
>>>>
>>>>
>>>> Thank You,
>>>>
>>>> Logan Barfield
>>>> Tranquil Hosting
>>>>
>>>>
>>>> On Tue, Jul 22, 2014 at 4:30 PM, Wido den Hollander <wido@widodh.nl>
>>>> wrote:
>>>>
>>>>>
>>>>>
>>>>> On 07/22/2014 10:13 PM, Logan Barfield wrote:
>>>>>
>>>>>> Wido,
>>>>>>
>>>>>> I appreciate the confirmation.  Would you mind posting an update here
>>>>>> when
>>>>>> you've submitted the patch?  As I mentioned in another post we will
>>>>>> probably end up just having to go into production with a non-official
>>>>>> build.  We won't be able to wait for 4.5 or 4.4.x unfortunately.
>>>>>>
>>>>>>
>>>>> I just pushed a patch: https://git-wip-us.apache.org/
>>>>> repos/asf?p=cloudstack.git;a=commitdiff;h=
>>>>> 173909e99d85cfcc85b017bc426950f9f16fddf0
>>>>>
>>>>> That should also apply on 4.4 very easily, a simple cherry-pick should
>>>>> be sufficient.
>>>>>
>>>>>
>>>>> Wido
>>>>>
>>>>>
>>>>>> Thank You,
>>>>>>
>>>>>> Logan Barfield
>>>>>> Tranquil Hosting
>>>>>>
>>>>>>
>>>>>> On Tue, Jul 22, 2014 at 4:09 PM, Wido den Hollander <wido@widodh.nl>
>>>>>> wrote:
>>>>>>
>>>>>>
>>>>>>>
>>>>>>> On 07/22/2014 09:53 PM, Logan Barfield wrote:
>>>>>>>
>>>>>>>   I was testing on a recent 4.4 build today and noticed that volume
>>>>>>>> resizing
>>>>>>>> for RBD volumes is not working as intended.
>>>>>>>>
>>>>>>>> In LibvirtComputingResource.java the 'getResizeScriptType()' function
>>>>>>>> doesn't have logic for type = RBD and format = RAW or QCOW2, so it
>>>>>>>> returns
>>>>>>>> 'null' and the resize operation fails.
>>>>>>>>
>>>>>>>> In CLOUDSTACK-6181 there was some discussion regarding RBD support,
>>>>>>>> and it
>>>>>>>> was signed off on by Wido (who appears to be the only contributor
>>>>>>>> actively
>>>>>>>> supporting Ceph in CloudStack).  Was the mentioned functionality just
>>>>>>>> never
>>>>>>>> implemented, or am I overlooking something?
>>>>>>>>
>>>>>>>>
>>>>>>>>   So it seems you are right. I can't remember anymore why I signed it
>>>>>>> off
>>>>>>> (probably because it worked locally), but what you are saying is true.
>>>>>>>
>>>>>>> RBD are block devices which do not exist in kernel space, so a script
>>>>>>> can
>>>>>>> never check if the volume exists.
>>>>>>>
>>>>>>> The fix is rather simple, since libvirt can resize the volume it's
>>>>>>> just a
>>>>>>> matter of a few if-statements in LibvirtComputing resouce, I'll get
>>>>>>> right
>>>>>>> to that, but it will never make it into 4.4. Hopefully 4.4.1
>>>>>>>
>>>>>>> Wido
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>> Thank You,
>>>>>>>>
>>>>>>>> Logan Barfield
>>>>>>>> Tranquil Hosting
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>
>>>>
>>>
>>
>