[RFC] Adding timestamp to guest's serial console log

[Peter Krempa]

On Tue, Aug 29, 2023 at 11:11:09 +0200, Michal Prívozník wrote:
> On 8/29/23 09:04, Shaleen Bathla wrote:
> > ping
> > 
> > -----Original Message-----
> > From: Shaleen Bathla <shaleen.bathla at oracle.com> 
> > Sent: Tuesday, June 20, 2023 2:44 PM
> > To: libvir-list at redhat.com
> > Subject: [RFC] Adding timestamp to guest's serial console log
> > 
> > Hi,
> > 
> > Need some comments regarding the following feature :
> > Addition of timestamp support for serial console logs of a guest.
> > 
> > We can implement it as a configurable attribute in xml.
> > For example :
> > <serial type='pty'>
> >   <log file='$PATH/$GUESTNAME-serial0.log' append='on' timestamp="on/off"/>
> >   <target type='isa-serial' port='0'>
> >     <model name='isa-serial'/>
> >   </target>
> > </serial>
> > 
> > We can add a timestamp after every '\n' character received from qemu.
> > Can I have some comments regarding this change like what I should keep
> > in mind while implementing, whether it is a welcome addition or not,
> > issues I might face, any qemu changes required.
> > 
> 
> Yeah, this might be useful. We definitely need additional attribute to
> control this behavior, otherwise we might break some scenarios (e.g.
> where users transfer binary data via this channel.
> 
> What I worry though, is that with stdio_handler="file", i.e. when the
> file is passed to QEMU directly, then this knob won't work. So we would
> need some changes to QEMU. But then, taking this idea further, if QEMU
> is changed, then does it make sense to have the same code in libvirt
> (for stdio_handler='logd' case)?

Nowadays I don't think that anybody disables logd.

You can always report an error if the configuration is not supported.

Note that '<log>' is a separate output from qemu, so adding timestamps
to that only (via logd) will not break anything trying to communicate
via the real PTY side. Just the log file will be binary data intermixed
with timestamps at certain points.