Re: [Qemu-devel] QMP: RFC: I/O error info & query-stop-reason

[Anthony Liguori <anthony@codemonkey.ws>]

On 06/06/2011 04:25 AM, Kevin Wolf wrote:
> Am 02.06.2011 20:09, schrieb Luiz Capitulino:
>>>> I'm ok with either way. But in case you meant the second one, I guess
>>>> we should make "reason" a dictionary so that we can group related
>>>> information when we extend the field, for example:
>>>>
>>>>    "reason": { "no space": false, "no permission": true }
>
> Splitting up enums into a number of booleans looks like a bad idea to
> me. It makes things more verbose than they should be, and even worse, it
> implies that more than one field could be true.

I agree.  What I had suggested was to not have a reason at all.

>
> If this new schema thing doesn't support proper enums, that's something
> that should be changed.

It does BTW.

>>>
>>> Why would we ever have "no permission"?
>>
>> It's an I/O error. I have a report from a developer who was getting
>> the BLOCK_IO_ERROR event and had to debug qemu to know the error cause,
>> it turned out to be no permission.
>
> And I want to add that it's a PITA to handle bug report when the only
> message you get from qemu is "something went wrong". Sorry, that's not
> useful at all. I want to see the real error reason (and at least for
> debugging this means, I want to see the errno value/string).
>
> Finding out that it was -EACCES in fact cost me (and QA who ran into the
> problem) much more time than it should have. It's simply too much that
> you need to attach gdb to find out what really happened.

You want a log file and you want libvirt to actually let QEMU write to a 
log file.

Since we already have trace points, could we have a white list of trace 
points that are written to a log file by default?

QMP is a stable interface that we have to maintain forever.  We can 
write whatever we want to a log file and change what gets written at will.

Regards,

Anthony Liguori

> Kevin
>