Re: [Qemu-devel] chardev's and fd's in monitors

["Dr. David Alan Gilbert" <dgilbert@redhat.com>]

* Markus Armbruster (armbru@redhat.com) wrote:
> "Dr. David Alan Gilbert" <dgilbert@redhat.com> writes:
> 
> > * Markus Armbruster (armbru@redhat.com) wrote:
> >> "Daniel P. Berrange" <berrange@redhat.com> writes:
> [...]
> >> > Realistically all the major backend subsystems (chardev, network, block,
> >> > ui and migration) need to be converted to Error ** propagation, since
> >> > they all ultimately call into some common code that reports Error **.
> >> 
> >> Infrastucture generally doesn't know how it's used, which means
> >> error_report() is generally wrong there.  Sufficiently simple functions
> >> can keep returning -errno, null, whatever, but the interesting stuff
> >> needs to use Error.
> >> > Very few places will end up being able to stick with -errno, or plain
> >> > error_report in the long term.
> >> 
> >> Not sure about "very few".  Less than now.  We'll see.
> >
> > I'd also prefer we got the very-few level; Migration used to be
> > characterised by getting a 'load of migration failed -22' and having
> > no clue in the logs to why; I've slowly fought back to be able
> > to get an error from the lowest level that caused the failure.
> > I want more of that, so that when someone gets a rare failure in the field
> > I can see why.
> 
> When it's about details that are only useful for debugging, logging
> might be a practical alternative.  No excuse for shoddy error reporting,
> of course.

It's detail I want in the initial error report rather than future debug;
to give a real example:

 qemu-system-ppc64: 9223477658187168481 != 9223477658187151905
 qemu-system-ppc64: Failed to load cpu:env.insns_flags
 qemu-system-ppc64: error while loading state for instance 0x0 of device 'cpu'
 qemu-system-ppc64: load of migration failed: Invalid argument

I wouldn't want to log the value loaded from each field unless I'm in the
pain of really bad debugging; but when one goes wrong like this getting the
value mismatch, the field name, and the device that failed is what I want
in the logs.

One point here is that those lines each come from a different function 
as we come back up out of the failure;  the one at the bottom doesn't
have the information on the device it's loading, just that it's loading
an integer and it's not the expected value.

Dave

--
Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK