|
Hi Pascal,
I noticed that the engine.target that you attached has this IU
<unit id="org.eclipse.equinox.p2.transport.ecf"
version="0.0.0"/>
listed *twice*. At first look this doesn't seem right...i.e. that
it's listed twice...although maybe it doesn't matter...if it
doesn't matter then please ignore.
Thanks,
Scott
On 5/12/2014 1:58 AM, Pascal Rapicault wrote:
Hi
Kai,
p2 does not offer any support for Concierge. That said I'm
attaching a target definition called engine.target that contains
the bundles required to just get the engine going on Equinox (note
that you will need to wait for tonight's I build to be available
because when trying to set this up I realized that there was a
missing optional flag on an import package
(http://git.eclipse.org/c/equinox/rt.equinox.p2.git/commit/?id=445e1dbcd4a4fa72662d0455c2b4e325b8037a67)).
Trying to use this target definition against Kepler would work but
it would work but would bring sat4j and the director (but they
would not be used at runtime).
An even lighter way would be to just use simpleconfigurator (and
the manipulator) to read and write the bundles.info file (see
configurator.target). However in this case, it would be your code
responsibility to download the jars, store them somewhere, and
finally update the bundles.info. So now you may ask why bother
with bundles.info? Because you need a way to do the install and
update without disturbing the running system, and these 2 bundles,
by letting the framework run undisturbed provide just that. And
btw, iirc, these two bundles would run on concierge unmodified.
Now how to use the engine on concierge?
There are two ways:
- Provide an implementation of the frameworkadmin service to deal
with concierge. This service is abstracting behind an interface
the format on disk of the various configuration files that are
specific to a fwk. For example frameworkadmin.equinox takes care
of the details of reading and writing the various files dealt with
by equinox (config.ini, eclipse.ini, bundles.info (through the
simpleconfigurator bundles))
- Provide a touchpoint implementation specific for concierge. In
this case you create a custom touchpoint that only cover your
needs and can do everything you need :) If you wanted to be able
to reuse the metadata published into the p2 repositories (to be
specific, the touchpoing actions like setStartLevel,
installBundle, etc.) you would need to provide actions that cater
for this but this is not complicated.
In any even, some code would have to be written
HTH
Pascal
On 10/05/2014 3:23 PM, Kai Kreuzer wrote:
Hi Pascal,
For small devices, you most likely
don't want p2 dependency resolution (or any other) to run on
the device because it will likely be too much computation.
However if you still want a transactional installation you
could go by using the p2 engine to perform the download and
the installation in the runtime. Then how the engine gets the
list of IU to install / uninstall / update is an
implementation detail :) For example you could use the OMA DM
from OSGi (not sure how it would work with the LWM2M server),
or implement something custom where the device contacts a
server, the server does the resolution and returns the list of
things to install, or use remote service, ... I'm sure there
are many other ways to do this and these are just things that
I know could be made to work.
As Jens, I am also looking for a very lightweight update
mechanism for Eclipse SmartHome. What you describe sounds like a
good option - moving the dependency resolution to the server and
just sending the runtime the list of things to install.
What is the minimal set of bundles required for the plain p2
engine to run on e.g. Concierge?
Best regards,
Kai
Am 10 May 2014 um 19:21 schrieb Pascal Rapicault
<pascal@xxxxxxxxxxxxx>:
HI Jens,
Sorry for the late reply. Find below more details
On 29/04/2014 3:58 AM, Jens Reimann wrote:
Hello Pascal,
Thanks for your answers.
We already start Eclipse without the native launcher in most
cases.
Wrapped by a Python script to find the correct jar. However
you
sometimes get some native bundles in there. I am not sure if
this is
Equinox or the Eclipse Runtime though.
The eclipse.exe companion jar comes to mind (e.g.
equinox.launcher_win32.win32) and it may be brought in by some
features you depend on but equinox itself has no native
dependencies.
For P2 I find it rather hard to find
god starting points for alternative
solutions. I think it does what it is intended to do, and it
does it
good. But you pretty soon get lost when it comes to
alternative
launching and provisioning ways (beside the Eclipse IDE). I
don't blame
anyone, I know how hard it is to write documentation ;-)
Some of these other starting points are not documented
because they never got beyond the prototype stage a few years
ago because nobody ended up need them.
Also is the whole Eclipse way a bit
more powerful than plain OSGi. As
far as I know OSGi proposes OBR for provisioning. However I
did find no
support for that in Equinox or P2. I think it was Felix that
allowed
installing bundles directly from Maven repositories, which
would also be
an alternative way.
If you are talking about OBR as found in Felix, know
that it is not more standard than p2 is.... What has been
standardized a couple years ago is a repository format, a
service to access repositories and an API to perform
dependency resolution. This is far from solving the problem of
downloading bundles and then installing them in a
transactional way in your runtime.
I also have to admit that I am not
sure what the best way actually is
for small devices. On the one hand you want to be flexible
and only
install the bundles that you need. It may be nice to do this
manually
("clicking" together your system with dependencies). On the
other hand
if you have 100s of devices and every device is special you
don't want
to maintain this manually.
For small devices, you most likely don't want p2
dependency resolution (or any other) to run on the device
because it will likely be too much computation. However if you
still want a transactional installation you could go by using
the p2 engine to perform the download and the installation in
the runtime. Then how the engine gets the list of IU to
install / uninstall / update is an implementation detail :)
For example you could use the OMA DM from OSGi (not sure how
it would work with the LWM2M server), or implement something
custom where the device contacts a server, the server does the
resolution and returns the list of things to install, or use
remote service, ... I'm sure there are many other ways to do
this and these are just things that I know could be made to
work.
HTH
Pascal
But I guess all this should be subject
to a different post :)
And yes "more independence from Equinox in your application"
correct.
Jens
On 04/27/2014 04:17 AM, Pascal Rapicault wrote:
Hi Jens
Answer embedded.
On 24/04/2014 4:52 AM, Jens Reimann wrote:
Hello Pascal,
I don't want to sound negative, but from the viewpoint
of Eclipse SCADA
we are pretty happy with Equinox ;-)
But we also learned, the hard way, that Equinox seems to
make a lot of
"special things". That might be easier at first, but
creates problems
when other OSGi containers become a topic. At the moment
we are not able
to deploy to other OSGi containers since we need P2 and
EMF
For EMF we really do need EMF. But for P2, we only
require a some sort
of build -> provision toolchain. Which is Maven Tycho
(p2) -> P2
Director at the moment.
Our long term goal is to also have other, more pure,
OSGi containers.
Also not needing any native code in order to launch the
framework. And
for this we would like to use technology coming from the
Eclipse
eco-system.
The framework can be started w/o the native launcher.
For example
you can start the Eclipse IDE by running java -jar
plugins/org.eclipse.equinox.launcher_....jar and you can
also directly
start equinox itself in a similar way (java -jar
org.eclipse.osgi_...jar). You can also see code embedding
Equinox in
Tycho.
For this we would need Declarative
Services (DS) and some sort of
provision mechanism. P2 is a bit bulky, but works fine
for our case. We
do use the P2 director to assemble our applications
(depending from the
System Configuration the user created). This done on
either the
configuration machine or the target machine. There is no
"online
update". So if we have a system configuration with 10
servers and each
server has 2 different applications, all applications
are provisioned
using the P2 director from the initial set of P2
repositories.
It is true that p2 is a bit bulky and is currently
known to only
run on Equinox. However p2 has been designed to run and
provision
other frameworks. In fact back when we started we were
able to
provision Felix. At this point I don't know exactly what
it would take
to make that work again, but I think this is in the real
of
possibilities.
The other thing you may be interested in is in
knowing that there
has been designs and prototypes done in p2 to make it work
"after the
fact". Meaning that you would not need to have the p2
folder and all
that in your application. Instead p2 would be able to
reason about the
currently installed bundle and take it from there to
perform the
installation (still bringing transactionality and
dependency resolution).
However we also see a shift at the
moment from this deployment scheme,
to a more distributed scheme. In the the past we had two
big servers,
and that's it. At the moment systems go up to 20
different nodes which
are (compared to then) less powerful. And we now start
to have ARM based
devices which will have even less performance and
greater numbers.
So in the end the three most important reasons for us to
move to a
different OSGi container are:
-) Better provisioning
Maybe a topic of the p2-dev mailing list, but I would
be
interested in knowing what are the criteria for a better
provisioning.
-) Less (or better no) native
dependencies (launcher)
See note above.
-) Better compatibility with OSGi
I m assume you mean more independence from Equinox in
your
application, or do you have something else in mind
Thanks
Pascal
Jens
On 04/23/2014 06:19 PM, Pascal Rapicault wrote:
Hi,
Since Concierge has been released, I noticed a lot of
interest from
the variou IoT projects to move to concierge from
Equinox and I would
like to understand the motivation of such a move? I
can obviously get
the size argument, but is there more to it?
Thanks
Pascal
_______________________________________________
iot-wg mailing list
iot-wg@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/iot-wg
_______________________________________________
iot-wg mailing list
iot-wg@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/iot-wg
_______________________________________________
iot-wg mailing list
iot-wg@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/iot-wg
_______________________________________________
iot-wg mailing list
iot-wg@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/iot-wg
_______________________________________________
iot-wg mailing list
iot-wg@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/iot-wg
_______________________________________________
iot-wg mailing list
iot-wg@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/iot-wg
_______________________________________________
iot-wg mailing list
iot-wg@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/iot-wg
|
