reactor examples

[Rafael Schloming <rh...@alum.mit.edu>]

I pushed some basic reactor examples into the proton tree under
examples/reactor/py. These are geared mostly towards
understanding/explaining the basic processing model of the event loop.

Please have a look. They are still a bit of a work in progress, as is the
API itself, but all feedback is welcome.

--Rafael

Re: reactor examples

[Rafael Schloming <rh...@alum.mit.edu>]

On Thu, Feb 12, 2015 at 8:29 AM, Gordon Sim <gs...@redhat.com> wrote:

> On 02/05/2015 05:10 PM, Rafael Schloming wrote:
>
>> On Wed, Feb 4, 2015 at 8:39 PM, Ted Ross <tr...@redhat.com> wrote:
>>
>>> I'm a bit confused by this push.  How does this set of examples relate to
>>> the examples Gordon has been developing in examples/engine/py?
>>>
>> [...]
>
>> The short summary is that I've worked to implement in C a lot of the
>> concepts Gordon already developed so that we can gain the benefits of his
>> work across all languages, not just Python.
>>
>
> I didn't really develop any concepts. I focused on getting a set of
> working examples that demonstrated a level of usability and functional
> completeness I believe is necessary, along with utilities (essentially some
> handlers and a 'container' class) that enable this. The examples and the
> utilities are currently in python.
>

Concepts might not be quite the right word, but you did develop most of the
dispatch model necessary to support the simplified scenarios (scoped
handlers, derived events, etc). My hope is by providing this model in C we
can a) keep the model consistent across all the bindings and b) give them
all a head start when it comes to implementing something similar, i.e. they
can focus just on whatever convenience layer makes the most sense for that
given language and not have to redevelop the same plumbing.


> Ultimately, we'll want to
>> reconcile the Python container with the C Reactor. As you point out they
>> are actually quite close already.
>>
>
> The purpose of the container (as well as being the simple entry point for
> applications), was as a place to locate the convenience functions for
> connecting, establishing links etc, that allow applications to remain
> succinct and (I hope) intuitive. I.e. the important difference is exactly
> what the container is seeking to add.
>
> What the reactor replaces is the plumbing that was necessary to drive
> everything. The container can be viewed as a layer on top of that or an
> extension to it.
>
>  Gordon and I have discussed how some of
>> the remaining differences can be reconciled. I'll be posting some notes on
>> this in the next few days.
>>
>
> I have all the 'container'  examples working over the reactor now. There
> is a small change to the reactor itself to allow for more flexibility in
> configuration of ssl and sasl:
>
> https://reviews.apache.org/r/30926/
>
> There is a trivial temporary workaround for the lack of sufficient
> transport events being emitted when using ssl:
>
> https://reviews.apache.org/r/30927/
>
> Then there are a couple of small additions to the python wrapped version
> of the reactor:
>
> https://reviews.apache.org/r/30928/
>
> https://reviews.apache.org/r/30929/
>
> With these in place the two attached patches will apply and complete the
> work (I'm not sure how to put patches up on reviewboard that depend on
> other patches). The only real change to the api is that the schedule method
> takes a relative duration, where previously it took an absolute time. It
> also allows a handler to be specified for the specific scheduled event. I
> think these are both improvements.
>
> (I split the attachments into two to make it easier to review. The first
> is an entirely non-functional change that simply moves code, unaltered.
> This makes it clearer that there are no further changes to the reactor. As
> a single diff it was a lot harder to read).
>

I'm +1 on all the patches (both reviewboard and attached).

--Rafael

Re: reactor examples

[Gordon Sim <gs...@redhat.com>]

On 02/05/2015 05:10 PM, Rafael Schloming wrote:
> On Wed, Feb 4, 2015 at 8:39 PM, Ted Ross <tr...@redhat.com> wrote:
>> I'm a bit confused by this push.  How does this set of examples relate to
>> the examples Gordon has been developing in examples/engine/py?
[...]
> The short summary is that I've worked to implement in C a lot of the
> concepts Gordon already developed so that we can gain the benefits of his
> work across all languages, not just Python.

I didn't really develop any concepts. I focused on getting a set of 
working examples that demonstrated a level of usability and functional 
completeness I believe is necessary, along with utilities (essentially 
some handlers and a 'container' class) that enable this. The examples 
and the utilities are currently in python.

> Ultimately, we'll want to
> reconcile the Python container with the C Reactor. As you point out they
> are actually quite close already.

The purpose of the container (as well as being the simple entry point 
for applications), was as a place to locate the convenience functions 
for connecting, establishing links etc, that allow applications to 
remain succinct and (I hope) intuitive. I.e. the important difference is 
exactly what the container is seeking to add.

What the reactor replaces is the plumbing that was necessary to drive 
everything. The container can be viewed as a layer on top of that or an 
extension to it.

> Gordon and I have discussed how some of
> the remaining differences can be reconciled. I'll be posting some notes on
> this in the next few days.

I have all the 'container'  examples working over the reactor now. There 
is a small change to the reactor itself to allow for more flexibility in 
configuration of ssl and sasl:

https://reviews.apache.org/r/30926/

There is a trivial temporary workaround for the lack of sufficient 
transport events being emitted when using ssl:

https://reviews.apache.org/r/30927/

Then there are a couple of small additions to the python wrapped version 
of the reactor:

https://reviews.apache.org/r/30928/

https://reviews.apache.org/r/30929/

With these in place the two attached patches will apply and complete the 
work (I'm not sure how to put patches up on reviewboard that depend on 
other patches). The only real change to the api is that the schedule 
method takes a relative duration, where previously it took an absolute 
time. It also allows a handler to be specified for the specific 
scheduled event. I think these are both improvements.

(I split the attachments into two to make it easier to review. The first 
is an entirely non-functional change that simply moves code, unaltered. 
This makes it clearer that there are no further changes to the reactor. 
As a single diff it was a lot harder to read).

Re: reactor examples

[Rafael Schloming <rh...@alum.mit.edu>]

On Wed, Feb 4, 2015 at 8:39 PM, Ted Ross <tr...@redhat.com> wrote:

> Rafael,
>
> I'm a bit confused by this push.  How does this set of examples relate to
> the examples Gordon has been developing in examples/engine/py?
>

I tried to explain this a while back when I originally posted about this
work:


http://qpid.2158936.n2.nabble.com/c-reactor-gordon-s-examples-td7618339.html

The short summary is that I've worked to implement in C a lot of the
concepts Gordon already developed so that we can gain the benefits of his
work across all languages, not just Python. Ultimately, we'll want to
reconcile the Python container with the C Reactor. As you point out they
are actually quite close already. Gordon and I have discussed how some of
the remaining differences can be reconciled. I'll be posting some notes on
this in the next few days.

--Rafael

Re: reactor examples

[Ted Ross <tr...@redhat.com>]

Rafael,

I'm a bit confused by this push.  How does this set of examples relate 
to the examples Gordon has been developing in examples/engine/py?

They are both "reactor" APIs with very similar structure yet they appear 
to be completely different and unrelated.  Yours uses Reactor, 
on_delivery, and on_reactor_start.  Gordon's uses Container, on_message, 
and on_start.

Gordon has been developing this API in the open for the last several 
months.  Why have you suddenly contributed a seemingly overlapping API?

-Ted


On 01/29/2015 06:23 PM, Rafael Schloming wrote:
> I pushed some basic reactor examples into the proton tree under
> examples/reactor/py. These are geared mostly towards
> understanding/explaining the basic processing model of the event loop.
>
> Please have a look. They are still a bit of a work in progress, as is the
> API itself, but all feedback is welcome.
>
> --Rafael
>