Re: Enabling balloon(4) in Xen dom0/domU kernels?

[Hugo Silva <hugo%barafranca.com@localhost>]

On 04/27/11 00:00, Jean-Yves Migeon wrote:
> On 26.04.2011 23:12, Hugo Silva wrote:
>> Had it briefly enabled with a 5.99.49 box, but with ballooning enabled
>> it rebooted on its own 2 or 3 times.
>>
>> I think the culprit might be swap on cgd though (or an interaction with
>> the ballooning code? I have no idea).
> 
> Do you have ddb.onpanic set to 1?

No, and the machine is some 300km away. Altough I visit from time to
time, won't be near it again for about a month.

> 
>> On several 5.1 domUs with very low memory (16-32M) I notice a hang where
>> the domU will be consuming 100% CPU (xm top from dom0) and xm console
>> seems to work but hangs right away, and ssh stops working too.
> 
> Balloon has not been pulled up to 5.1. What code did you take? The "old"
> one with the BALLOON option in config file, or the recent one (committed
> about a week ago)?
> 

I didn't clarify: I tested ballooning on dom0 only.
If the code is one week old, I must have tested the old code: my
/usr/obj mtime is from Apr 13 (is there a more elegant way to check?).

I do remember enabling it by uncommenting XEN_BALLOON in the config file.

>> Notably, to be able to compile a kernel on these VMs, I had to disable
>> swap on cgd.
>>
>> I'm not 100% sure it's 2 different problems, since the domU hangs at
>> 100% and the dom0 just reboots. I have the dom0 running with swap on cgd
>> and 768M mem for a week now, no crashes (it did crash with ballooning).
>> There is that.
>>
>> It's all a bit vague I know, but maybe someone else steps up with the
>> same problem and we might be able to track it down..
>>
>> Hugo
>>
>> Wish I had more precise information, I know it's a bit vague.
> 
> Turn on ddb.onpanic, hopefully we'll get a bit more details on the context.
> 
> Note that balloon(4) cannot be triggered unless explicitly signaled,
> either by starting a new domU (in which case, the Xentools will try to
> get back memory to start the domain), or via the xm mem-set command.
> Rest of the time, the associated LWP will remain idle and should not
> modify memory reservation, both for dom0 or domUs.
> 
> I'll try reproducing this with swap on cgd. Is only swap backed by
> cgd(4), or other file systems too (domain images, specific partitions, ...?)
> 

Inside some domains, I have some filesystems on cgd too. I am not 100%
sure anymore, but *I think* back when the crashes occured swap on cgd
was the only cgd configured.

In the dom0 there were no cgds configured (now I run swap on cgd @ the
dom0 too)

The logs have been rotated in the mean time, I remember seeing messages
from UVM and perhaps something about targets and numbers changing;
eventually the dom0 would reboot.

But this was probably with the old code.

I'll assume we're dealing with two different issues here then: dom0
rebooting with ballooning enabled (gotta make sure it was with the old
code, becomes irrelevant then) and 5.1 domUs hanging with 100% CPU,
apparently something to do with low memory and swap on cgd.