BRRABill's Field Report With XenServer

olivier

"Tasks" are a XAPI object, telling you something happening on your host.

The progress value is broken in all current versions of XenServer (fixed in Dundee).

Making a VM copy is a complex task:

we start to download chunks on the HTTP export handler of the origin host
at the same time, we start to upload those chunks (on the fly) on the destination host

Each one of them had their respective tasks (export and import) but they don't "know" we are in fact "streaming" stuff between them (for them, it's more like a classical import or export).

Plus, XAPI is a pain to know when the transfer is finished (we had to use some hacks, because they don't use HTTP standard). Furthermore, we don't have the size of what we are exporting, so we can't "guess" the progress by ourselves.

It could fail on any side (fail export, fail import, for many reasons). So we catch anything we could.

But for now, there is no "XO" tasks, just multiple "XenServer tasks".

BRRABill

@olivier

Perhaps I am not asking my question properly.

In XC (which is far less featured than XO in my opinion!), you get the following type of status:

"Succeeded","Copying VM 'XenOrchestra' to 'Copy of XenOrchestra' on SR 'SATA Array'
Copied
Time: 00:00:45","xenserver-1","Mar 29, 2016 3:00 PM"

There is no way to see that in XO? I can use XC for this, but XO is so much nicer to use, being web-based.

Also, XC seems to have a progress bar that worked when I did this copy. Are they working around the bug somehow?

olivier

@BRRABill Can you provide a real world example with a XC screenshot?

BRRABill

olivier

This is the XenServer task list. Is the task stays at "Success" when it's done? (eg with the migration?)

BRRABill

@olivier said:

This is the XenServer task list. Is the task stays at "Success" when it's done? (eg with the migration?)

Yes, it stays at success when it is done, with the amount of time it took.

As it moves along, the progress bar there also moves.

olivier

@BRRABill Okay so the only difference with us is they keep ended tasks for a time (how much time?)

About the progress: XenCenter uses the pure XML-RPC way of calling XAPI, which is less prone to bugs, but very CPU intensive (I mean, VERY cpu intensive: decoding tons of XML is horrible)

Using JSON encapsulated in XML-RPC made a small difference for larger infrastructure, around 250 times faster than the shitty XML (for larger infrastructure). See https://xen-orchestra.com/blog/improving-xen-orchestra-performances/ That's the only way for us to have XO working on infrastructure with 1k VMs or more (the problem starts with 100 VMs)

But sadly, they made mistakes in JSON and it's only fixed in Dundee.

BRRABill

@olivier said:

@BRRABill Okay so the only difference with us is they keep ended tasks for a time (how much time?)

It disappears when you exit XC.

It would just be nice to be able to check in and see if the task actually finished OK, or finished but with errors, or didn't finish at all.

As it stands now, you really don't know, right?

Is that something that COULD be added? Or is it useless and I"m missing the reason why it's useless? (Which is entirely, entirely possible!)

BRRABill

@olivier

Do you have to enable auto-power-on on the pool, before you can do it to individual VMs in XO?

My servers are set to auto-power-on in XO, but do not come back up when XS reboots.

olivier

@BRRABill Yes It could be added, but XAPI tasks are such a mess. We can't be sure about them, but yes it's possible to fetch them and store them somewhere. But it will be only informative and not the "truth".

About auto power on: if you did it with XO, it's normally done in both pool and VM at the same time (see https://xen-orchestra.com/blog/auto-start-vm-on-xenserver-boot/ ).

To check your current pool status with autopower on: xe pool-param-get uuid=<POOL_UUID> param-name=other-config

BRRABill

@olivier said:

To check your current pool status with autopower on: xe pool-param-get uuid=<POOL_UUID> param-name=other-config

This is what that command returned:
cpuid_feature_mask: ffffff7f-ffffffff-ffffffff-ffffffff; memory-ratio-hvm: 0.25; memory-ratio-pv: 0.25

olivier

Hmm no autopower_on activated. I imagine you already tick the checkbox on the XO GUI? Normally, this will add the right parameter in the pool. If not, I have to check this on my side too.

BRRABill

@olivier said:

Hmm no autopower_on activated. I imagine you already tick the checkbox on the XO GUI? Normally, this will add the right parameter in the pool. If not, I have to check this on my side too.

You mean the "auto power" dropdown on the VM? Yes.

Just to test, I enabled it though the CLI, and then re-ran the commands, and this is what it said. Note I did NOT adjust the parameter for the VM itself through the CLI, so perhaps XO just needed the pool setting changed. I rebooted, and it did indeed auto-start.

[root@xenserver-TEST ~]# xe pool-param-get uuid=3526ffee-1afe-6162-0afc-6362580d10cc param-name=other-config

auto_poweron: true; cpuid_feature_mask: ffffff7f-ffffffff-ffffffff-ffffffff; memory-ratio-hvm: 0.25; memory-ratio-pv: 0.25

[root@xenserver-TEST ~]# xe vm-param-get uuid=d09990ec-73ce-9946-9698-8b8ba70621fe param-name=other-config

auto_poweron: true; vgpu_pci: ; mac_seed: b3cddb7b-b5ec-f3c8-368a-2ba33059f96a; install-methods: cdrom; base_template_name: Windows Server 2012 R2 (64-bit)

olivier

AFAIK, XO would do both (activate the "autopower" on your VM but also on its pool).

If it's not the case anymore, maybe something is broken on our side. Let me check that.

BRRABill

@olivier said:

AFAIK, XO would do both (activate the "autopower" on your VM but also on its pool).

If it's not the case anymore, maybe something is broken on our side. Let me check that.

Let me know if there is anything I can do to help.

BRRABill

@olivier said:

AFAIK, XO would do both (activate the "autopower" on your VM but also on its pool).

If it's not the case anymore, maybe something is broken on our side. Let me check that.

They did in recent versions apparently remove the "auto power on" from the GUI due to issues with HA. I wonder if that broke something on your end?

olivier

Don't know.

But we are doing this in the code:

    if auto_poweron
      yield xapi.call 'VM.add_to_other_config', ref, 'auto_poweron', 'true'
      yield xapi.setPoolProperties({autoPowerOn: true})
    else
      yield xapi.call 'VM.remove_from_other_config', ref, 'auto_poweron'

So in theory, this should work.

BRRABill

On my Server 2003 VM, whenever I go to console, it loads the following dialog box. I hit CANCEL and the console shows up.

Does not do this on my Server 2012 eval box.

Any thoughts on why it does this? Is this standard?

Dashrender

@BRRABill said:

On my Server 2003 VM, whenever I go to console, it loads the following dialog box. I hit CANCEL and the console shows up.

Does not do this on my Server 2012 eval box.

Any thoughts on why it does this? Is this standard?

Yep...Because the older version of RDP doesn't have authentication like the new one. So assuming you're using Windows 7 or newer on the client side, the RDP client in Win7 or newer is looking for that authentication - which it gets from Server 2012, but not from 2003 - and it's warning you of such.

BRRABill

@Dashrender

Is the "console" tab using RDP?