Re: [PATCH v6 6/8] chardev/char-mux: implement backend chardev multiplexing

[Roman Penyaev]

On Fri, Jan 10, 2025 at 9:58 AM Daniel P. Berrangé <berrange@redhat.com> wrote:
>
> On Fri, Jan 10, 2025 at 09:43:52AM +0100, Roman Penyaev wrote:
> > On Thu, Jan 9, 2025 at 7:07 PM Daniel P. Berrangé <berrange@redhat.com> 
> > wrote:
> > >
> > > On Thu, Jan 09, 2025 at 01:56:40PM +0100, Roman Penyaev wrote:
> > > > Hi,
> > > >
> > > > On Tue, Jan 7, 2025 at 3:57 PM Marc-André Lureau
> > > > <marcandre.lureau@redhat.com> wrote:
> > > > > Whether we talk about multiplexing front-end or back-end, the issues
> > > > > are similar. In general, mixing input will create issues. Teeing
> > > > > output is less problematic, except to handle the buffering...
> > > >
> > > > I understand your concerns. What exact issues do you have in mind?
> > > > Are these issues related to the input buffer handling, so technical 
> > > > issues?
> > > > Or issues with usability?
> > >
> > > While the design / impl technically allows for concurrent input to be
> > > sent to the frontend, from multiple backends, in practice I don't think
> > > we need to be particularly concerned about it.
> > >
> > > I don't see this as being a way for multiple different users to interact
> > > concurrently. Rather I'd see 1 user of the VM just deciding to switch
> > > from one backend to the other on the fly. IOW, although technically
> > > possible, the user will only be leveraging one at a time to send input.
> > >
> > > We very definitely do need all backends to receive output from the guest
> > > concurrently too, as you'd want the historical output context to be
> > > visible on whatever backend you choose to use at any given point in time.
> > >
> > > If a user decides to be crazy and send input from multiple backends
> > > concurrently, then they get to keep the mess.
> > >
> > > > > > Do you think we need to artificially introduce multiplexing logic 
> > > > > > to be fully
> > > > > > compliant with multiplexer naming? It's not hard to do, repeating
> > > > > > `mux_proc_byte()` from `mux-fe`. In my use-case, I'll still need to 
> > > > > > disable
> > > > > > multiplexing in favor of 'mixing', for example with the 'mixer=on' 
> > > > > > option,
> > > > > > i.e. '-chardev mux-be,mixer=on,...`. Or do you think it should be 
> > > > > > some
> > > > > > completely different beast, something like mixer chardev?
> > > > >
> > > > > I think it would be saner to have the muxer be selectors: only work
> > > > > with one selected be or fe. Otherwise, we can run into various issues.
> > > >
> > > > In multiplexing (not mixing) for the use-case that I am describing, 
> > > > there is one
> > > > serious drawback: as soon as you switch the "focus" to another input 
> > > > device
> > > > (for example from vnc to socket chardev), you will not be able to 
> > > > s]witch back
> > > > from the same input console - the input now works on another device. 
> > > > This looks
> > > > strange and does not add convenience to the final user. Perhaps, for a 
> > > > case
> > > > other than console, this would be reasonable, but for console input -
> > > > I would like
> > > > to keep the mixer option: the front-end receives input from both 
> > > > back-ends.
> > >
> > > Agreed, I think this is desirable. If you did the exclusive access mode,
> > > it'd complicate things as you now need a way to switch between active
> > > backends, while also reducing the usefulness of it.
> > >
> > > The main thing I'm not a fan of here is the naming 'mux-fe', as I think we
> > > should have something distinct from current 'mux', to reduce confusion
> > > when we're talking about it.
> >
> > The idea to have mux-fe and mux-be (current implementation) was born to
> > distinguish what exactly we multiplex: front-ends or back-ends.
> >
> > As Mark-Andre rightly noted, input from back-end devices is not multiplexed,
> > but rather mixed.
> >
> > >
> > > How about 'overlay' or 'replicator' ?
> >
> > Overlay for me has a strong association with the filesystem concept. This
> > would work for me if combined back-end inputs function by layering one
> > on top of another, with potentially higher-priority inputs overriding
> > lower-priority ones. It implies a hierarchical or layered merging approach.
> > Not quite well describes a simple mixing strategy.
> >
> > Replicator - this can be a good name from front-end device point of view:
> > suggests a mechanism for distributing the same input (front-end) to 
> > different
> > destinations (back-ends).
> >
> > Two more: what about 'aggregator' or even 'hub' ?
>
> Yes, those are ok
>
> > Also 'mixer'? So we have '-chardev mux' and '-chardev mix' (try not to get
> > confused :)
>
> AFAIR, users would not use '-chardev mux', but instead set 'mux=on' on the
> real chardev backend.

Yeah, right, I forgot about this peculiarity.

If Mark-Andre or anyone else has no objections, I'll drop all changes
to the original front-end 'mux' (I tried to make all names and variables reflect
the 'front-end' nature to reduce confusion with the back-end mux) and will
resend modified series introducing a new 'hub' chardev. I will keep the
possibility to send input from several back-ends to a single front-end, as
in the current implementation.

--
Roman