[libvirt] Syslog "nested job is dangerous" message

[Sam Bobroff]

On 17/07/14 12:50, Eric Blake wrote:
> On 07/16/2014 07:52 PM, Sam Bobroff wrote:
>> Hello everyone,
> 
> [Can you configure your mailer to wrap long lines?]

[No problem, done.]

>> After performing a migration, the syslog often contains several
>> messages like this:
>> 
>> "This thread seems to be the async job owner; entering monitor
>> without asking for a nested job is dangerous"
> 
> The sign of a bug that we need to fix.  What version of libvirt are
> you using?  We may have already fixed it.

I've been testing on 1.2.6, but I will re-test on the latest code from git.

>> The message makes it sound as if the user has done something
>> dangerous but, after looking at the code that produces the message,
>> it appears to be more to do with how the libvirt job code is used.
> 
> Rather, libvirt itself has done something dangerous.
> 
>> 
>> The commit that added the warning
>> (6eede368bc8e3df2c94c2ec1a913885ce08e99db) doesn't explain why it
>> might be dangerous...
> 
> Jiri can probably explain better, but it means there is a race
> condition where libvirt can lose track of a long-running job and
> cause memory corruption in its management of the guest.
> 
>> Would it be appropriate to re-word the message, or perhaps change
>> it to a debug message so that it's not normally seen by users?
> 
> No.
> 
>> 
>> Is it important to track down the cases that are generating the
>> warning and fix them? (Could it cause some kind of significant
>> problem?)
> 
> Yes.

Acknowledged.

If they still occur on the latest code I'll start tracking them down :-)

Cheers,
Sam.