[LEDE-DEV] [OpenWrt] [OpenWrt-Devel] Project proposal: The GNUnet of autonomous Things

[Daniel Golle]

Hi Hauke,

On Mon, Dec 05, 2016 at 01:54:20PM +0100, Hauke Mehrtens wrote:
> On 2016-12-05 11:57, Daniel Golle wrote:
> > Hi Felix,
> > 
> > On Thu, Dec 01, 2016 at 04:51:30PM +0100, Felix Fietkau wrote:
> > > On 2016-12-01 16:38, Daniel Golle wrote:
> > > > Hi Felix,
> > > >
> > > > On Thu, Dec 01, 2016 at 04:12:38PM +0100, Felix Fietkau wrote:
> > > >> On 2016-12-01 16:05, Daniel Golle wrote:
> > > >> > I was following your posts and do believe there is quite some overlap.
> > > >> > It would thus be feasible to generalize the common parts (ubus call
> > > >> > proxy, ubus service proxy, ubus remote monitor) by agreeing on a shared
> > > >> > interface the actual implementations shall use. In that way, people
> > > >> > can choose whether they want WebSockets, TR-069 or a suitable P2P
> > > >> > framework depending on their specific needs.
> > > >> > Has anything of your current approach at IOPSYS been made available
> > > >> > publicly (eg. on github?)
> > > >> >
> > > >> > From what I can tell there is also some overlap with Felix' proposed
> > > >> > System Configuration Abstraction Layer, just that my envisioned use
> > > >> > exceeds system configuration as it includes sensors, events and actors
> > > >> > rather than just access to a configuration model.
> > > >> If it makes sense, I'd be open to extending my abstraction layer to make
> > > >> it suitable for your use case as well.
> > > >> Feel free to propose changes to it if you like.
> > > >
> > > > Having a first deeper look at scal I believe that access to sensors
> > > > and actors could be implemented inside scal similar to the existing
> > > > shell and system backends. That would be nice, as then scal would
> > > > make things available on ubus and provide the ACL mechanics.
> > > Nice. Maybe we can reinterpret the acronym as "System Communication
> > > Abstraction Layer". I'd be fine with renaming it to something else as
> > > well, I just didn't find a better name for it yet.
> > > 
> > > I think a good approach would be to add a dlopen plugin API to the
> > > json
> > > plugin itself, so you can use json files to parameterize access to
> > > sensors and other devices.
> > 
> > To me the question remains whether access to devices should happen
> > directly inside those scal-json-plugins or if it'd be better to
> > expose a service on ubus ("urthings") handling them (which was my
> > original plan) and then have scal access them via ubus. The latter
> > would come with the advantage that other local services (think:
> > collectd) could also access them via that urthings service instead of
> > having to go through scal.
> 
> I would like to have an API which can be used locally easily not just from
> remote, I haven't found the time to look into scal .
> Like a Luci web UI to switch on and off your lamps and also some API so that
> others can easily integrate own application running on the device which are
> managing and controlling the things. Probably people will also run
> applications to connect the devices to existing clouds with existing
> interfaces.

I agree. Having a simple service on ubus allowing to expose sensors and
actors is still the best first-step which everything else can later on
build-upon. Using and, if necessary, extending SCAL to allow remote
access to specific objects exposed by that service shouldn't be too hard
either.

> 
> > I'd be glad to here more opinions, because this obviously has to be
> > decided early in the design of the IoT integration approach. Find
> > me on IRC (dangole at freenode) in the next couple of hours, maybe we
> > can collect some ideas about the edges to be cut before the meeting
> > at 3pm CET.
> > 
> > > 
> > > Event handling could also be scripted through .json files using
> > > json_script.
> > 
> > I thought of it like that, similar to how procd's hotplug json_scripts
> > look like. In addition I thought about adding ubus input and output
> > support to collectd (so events can be triggered depending on conditions
> > defined over polled sensors), but maybe something much simpler and more
> > thightly designed for that specific job -- polling sensors, (maybe)
> > caching & monitoring/triggering events -- could also be sufficient.
> > However, I reckon this can be decided at a later stage, and as I'm
> > already quite familiar with collectd, I'd just go ahead and create
> > a ubus plugin there and who ever is unhappy with that may suggest
> > what ever else could be better.
> 
> I think we should focus on Phase 1 first to get these stuff properly into a
> generic API in LEDE and then use some generic interfaces to expose it to
> remote hosts.

The above is more about whether polling/monitoring/caching/aggregating
should be taken care of by a to-be-created service or if implementing a
collectd input and output plugin for ubus would be agreable for
everyone...?


Cheers


Daniel

> 
> > 
> > 
> > Cheers
> > 
> > 
> > Daniel
> > 
> > 
> > 
> > > 
> > > > I'll have a deeper look and play with it to see whether that can
> > > > work.
> > > >
> > > > Ideally, data collection (think: interface counters and such things
> > > > which need to be polled) and triggering events (think: link status
> > > > updates) should also be made accessible.
> > > >
> > > > A local database which exceeds UCI state as suggested by Luka could
> > > > also be very useful, eg. for renewable energy or other applications
> > > > where loss of connectivity should never imply loss of collected data.
> > > Makes sense.
> > > 
> > > - Felix
> > 
> > _______________________________________________
> > Lede-dev mailing list
> > Lede-dev at lists.infradead.org
> > http://lists.infradead.org/mailman/listinfo/lede-dev
> 
> 
> _______________________________________________
> Lede-dev mailing list
> Lede-dev at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/lede-dev