streams-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Danny Sullivan <dsulliv...@hotmail.com>
Subject RE: [DISCUSS] Switching Streams from Camel deployment to .war deployment
Date Thu, 31 Oct 2013 12:40:44 GMT
Excellent, I'll write up something as a proof of concept and we can discuss further to make
sure everything is vanilla.

> Date: Wed, 30 Oct 2013 17:01:28 -0400
> Subject: Re: [DISCUSS] Switching Streams from Camel deployment to .war deployment
> From: jletourneau80@gmail.com
> To: dev@streams.incubator.apache.org
> 
> That sounds pretty promising to me.
> 
> On Wed, Oct 30, 2013 at 2:38 PM, Danny Sullivan <dsullivan7@hotmail.com> wrote:
> > Thanks for the feedback. You have an interesting point about the url linking to
a separate processing space. Let me tie my answer into your last question about "advocating
for the simplicity at registration to give up flexibility at registration, but retaining the
inner "guts" of EIP/messaging". Consider a new architecture:
> >
> > streams-web.war: single entry point to application, but functions ONLY as an entry
point. From here Camel routes the incoming requests to 4 separate jarssubscriber-registration.jar:
subscriber registration publisher-registration.jar: publisher registrationactivity.jar: returns
activity (also contains subscriber warehouse and storm activity aggregator)publish.jar: publishes
activitystreams-cassandra.jar: the above 4 jars would all have a hook into this jar which
would function as a hook onto the database. Each jar would have camel route output to this
jar.
> >
> > In this implementation, Camel would no longer be the entry and exit point of a client
to the application, but would handle the communication between components. The flow of activity
through the application would be method based in each jar. This would allow deployment on
up to 6 different process spaces. However, this does not address that there is a single server
entry point, but I'm not sure if it was a concern in the first place.
> >
> > My argument, at its basis, is that we should move away from using Camel as the entry
point to the application. I would be happy to maintain messaging between components.
> >
> >> Date: Wed, 30 Oct 2013 12:32:55 -0400
> >> Subject: Re: [DISCUSS] Switching Streams from Camel deployment to .war deployment
> >> From: jletourneau80@gmail.com
> >> To: dev@streams.incubator.apache.org
> >>
> >> An interesting use case that I am holding onto is the ability for
> >> publishers to register via a single URL (registration endpoint), but
> >> be sent a URL back to post to a different process space for actual
> >> publishing.  The same is true on the subscriber front.  Currently, the
> >> Camel/EIP infrastructure abstracts this because different components
> >> deployed in different process spaces handling the route creation can
> >> just be bolted onto a running Streams instance without new subs/pubs
> >> behaving any differently than existing.  This seems to be a
> >> potentially critical scaling point.  Is there a way to do this with
> >> the Spring solution?
> >>
> >> The persistence point is a good one, though I would classify that as
> >> "not implemented" vs "not possible" (not that you were).
> >>
> >> I'm not married to Camel, I just like the EIP approach to building
> >> something that is ultimately a messaging system.  There are known
> >> patterns that solve at least a subset of the problems Streams is
> >> trying to solve and implementations that can handle the load and I'll
> >> reiterate flexibility == complexity almost always.
> >>
> >> It comes right back to the central question: Do you want flexibility
> >> or simplicity?  It doesn't have to be black and white either I don't
> >> think...
> >>
> >> More pointedly: Where should we give up flexibility for simplicity?  I
> >> read that Danny is advocating for the simplicity at registration to
> >> give up flexibility at registration, but retaining the inner "guts" of
> >> EIP/messaging?  Thoughts?
> >>
> >> On Wed, Oct 30, 2013 at 11:37 AM, Danny Sullivan <dsullivan7@hotmail.com>
wrote:
> >> > My argument is not for the IoC pattern as that can be (and has been) implemented
alongside Camel. My main argument is that the syntax at the entry point is not only familiar
but much simpler. This wouldn't be a very strong argument if the Camel implementation wasn't
much more complicated but I feel that it is the case. Also, looking toward the future, if
the server is restarted, in-routes are lost in Camel. The way to curb this is to persist the
dynamic routes that Camel creates, and then on start up pull every one of these routes and
recreate a dynamic route for each one. Not only is this much easier to implement using the
Spring web implementation, but it already has been implemented and you can try it by checking
out the webservice branch, registering a subscriber, restarting tomcat, and using the same
url you had before. This will allow subscribers to hang on to their urls once they register.
(the same is true for publishers: you can post via the same url after restarting tomcat)
> >> >
> >> >> Date: Wed, 30 Oct 2013 11:00:21 -0400
> >> >> Subject: Re: [DISCUSS] Switching Streams from Camel deployment to .war
deployment
> >> >> From: jletourneau80@gmail.com
> >> >> To: dev@streams.incubator.apache.org
> >> >>
> >> >> To be fair, while the current implementation is heavily camel-based,
> >> >> all of the interfaces related to Streams functionality are not.  The
> >> >> current model maps to what Matt has outlined in my opinion, though
> >> >> packing names etc. probably don't follow that exact pattern.
> >> >>
> >> >> With regards to the complexity and different components in the
> >> >> registration process, this was a cut at the abstraction based on the
> >> >> assumption that different implementations may be plugged in and in
> >> >> fact may live on different processor space (ie. a polling publisher
vs
> >> >> a push publisher may be instantiated on different servers but the
> >> >> registration URL is staticly defined).
> >> >>
> >> >> Is the main argument I am seeing for Spring  the familiarity of its
> >> >> IoC pattern implementation and syntax at the entry point?
> >> >>
> >> >> On Wed, Oct 30, 2013 at 10:53 AM, Danny Sullivan <dsullivan7@hotmail.com>
wrote:
> >> >> > Could you clarify whether the same entry points would exist for
the camel implementation of the core (implementing the "process" method/ using a DynammicRouteBuilder)
or would the webservice be the sole entry point to Streams and after it enters would it hand
it off to Camel? And what would be the entry point for the Storm implementation?
> >> >> > -Danny
> >> >> >
> >> >> >> Date: Wed, 30 Oct 2013 10:13:04 -0400
> >> >> >> Subject: Re: [DISCUSS] Switching Streams from Camel deployment
to .war deployment
> >> >> >> From: m.ben.franklin@gmail.com
> >> >> >> To: dev@streams.incubator.apache.org
> >> >> >>
> >> >> >> On Tue, Oct 29, 2013 at 2:55 PM, Danny Sullivan <dsullivan7@hotmail.com>wrote:
> >> >> >>
> >> >> >> > My case for switching from OSGi is for simplicity in
design. To follow the
> >> >> >> > path of an activity through Streams in the webservice,
there is one main
> >> >> >> > things the developer needs to understand:
> >> >> >> > The @RequestMapping annotation specifies the HTTP entry
point
> >> >> >> > There are 4 @RequestMapping annotations that correspond
to each of the 4
> >> >> >> > ways a user enters the application: registering a publisher,
registering a
> >> >> >> > subscriber, publishing activity, and getting an activity
stream. Where
> >> >> >> > these are located in the source code can be found by
searching for the
> >> >> >> > paths specified in the documentation (search for "/publisherRegister",
> >> >> >> > "/subscriberRegister", "/publishActivity", "/getActivity"
which will all
> >> >> >> > lead you to StreamsWebController.java). From the methods
that process
> >> >> >> > requests, the flow through the application is through
methods which can be
> >> >> >> > understood by most Java programmers.
> >> >> >> > The flow of activities through the current trunk branch
is understood as
> >> >> >> > follows:
> >> >> >> > The string "/publisher/register" (the entry point to
register a publisher
> >> >> >> > specified in the documentation) is the value of the
> >> >> >> > consumer.registrationEndpoint property defined in streams.propertiesThe
> >> >> >> > camelContext.xml specifies an endpoint with the id
> >> >> >> > consumerRegistrationEndpoint to have a uri equal to the
propertyThe
> >> >> >> > consumerRegistrationEndpoint routes from uri direct:publisher
register with
> >> >> >> > the bean activityRegistrationProcessor nested in between
the routeThe
> >> >> >> > streams-eip-applicationContext contains a bean with the
id
> >> >> >> > activityRegistrationProcessor created for the class
> >> >> >> > ActivityPublisherRegistrationProcessorThe exchange will
enter the "process"
> >> >> >> > method of the class ActivityPublisherRegisitrationProcessor
and that this
> >> >> >> > is because this class implements the "Processor" interface
provided by
> >> >> >> > CamelThe direct:add-publisher-route takes the exchange
output from the
> >> >> >> > "process" method and routes it to the activityRegistrationProcessor
> >> >> >> > "register" method. The bean activityRegistrationProcessor
is defined in the
> >> >> >> > streams-eip-osgi-component-import.xmlThe output from
this method is then
> >> >> >> > sent to the "createNewRouteForConsumer" method of activityConsumerRouter.
> >> >> >> > This method creates a new route for the newly registered
publisher using
> >> >> >> > the private static final class DynamicConsumerRouteBuilder
which is
> >> >> >> > required to extend RouteBuilder which is provided by
Camel. This
> >> >> >> > DynamicConsumerRouteBuilder contains several methods:
> >> >> >> > getConsumerReceiveMethod() (which corresponds to @Value
> >> >> >> > ${consumer.receiveMethod} which corresponds to "receive"),
> >> >> >> > getConsumerSplitMethod() (@Value ${consumer.splitMethod},"split"),
and
> >> >> >> > getConsumerActivityQUri() (@Value ${consumerActivityQUri},
> >> >> >> > "direct:activityQ"). This is different than the camelContext.xml
in that
> >> >> >> > the route is being created programmatically. What this
is doing is routing
> >> >> >> > input from the inroute url (which Camel does automatically
through the
> >> >> >> > configure method which is required to be overridden),
to the "receive"
> >> >> >> > method of ActivityConsumer, then to the "split" method
of ActivityConsumer,
> >> >> >> > and then to the "direct:activityQ" which if you look
back in the
> >> >> >> > camelContext.xml routes to the "activemq:queue:activities"
which then
> >> >> >> > routes to "receiveExchange"
> >> >> >> > This is the process to register a publisher. The process
for registering a
> >> >> >> > subscriber is relatively the same though it involves
separate classes with
> >> >> >> > their own private static final RouteBuilder class. From
my perspective, the
> >> >> >> > two most difficult things with setting this project up
were understanding
> >> >> >> > that the "process" method of the class that implements
"Processor" is the
> >> >> >> > entry point and the DynamicConsumerRouteBuilder creates
the second entry
> >> >> >> > point (The 5th and last points). This made the project
very, VERY hard to
> >> >> >> > understand.
> >> >> >> > In addition to simplicity of design, the mvn clean install
of the web
> >> >> >> > service project is much faster and small scale activity
publishing is also
> >> >> >> > faster (see my email about load testing). These are minor
points though as
> >> >> >> > compilation has no effect on deployment. OSGi does add
the benefit of
> >> >> >> > modularized programming which is valuable, though I think
the added
> >> >> >> > complexity of Camel merits moving the project away from
this paradigm.
> >> >> >> >
> >> >> >>
> >> >> >> I agree that the project is pretty difficult to understand
ATM.  I think
> >> >> >> what we need to do is think about what the responsibilities
of the code are
> >> >> >> and allow for different implementations that are not so tightly
coupled as
> >> >> >> they are now.  For instance, having worked with Storm to ingest
millions of
> >> >> >> activities a day, I personally would like to see streams be
responsible for
> >> >> >> defining an over-arching orchestration model that can be implemented
within
> >> >> >> a single war or on top of a distributed system.  This would
look something
> >> >> >> like the follows:
> >> >> >>
> >> >> >> |__ Streams-Core  (Base classes, interfaces, utilities, extensions,
etc)
> >> >> >> |
> >> >> >> |__ Streams-Storm (Storm implementation of the core)
> >> >> >> |
> >> >> >> |__ Streams-Camel (Camel implementation of the core)
> >> >> >> |
> >> >> >> |__ Streams-WS (Web service implementation)
> >> >> >>
> >> >> >>
> >> >> >> > Danny
> >> >> >> >
> >> >> >> >
> >> >> >> > > From: dsullivan7@hotmail.com
> >> >> >> > > To: dev@streams.incubator.apache.org
> >> >> >> > > Subject: [DISCUSS] Switching Streams from Camel
deployment to .war
> >> >> >> > deployment
> >> >> >> > > Date: Tue, 29 Oct 2013 11:07:39 -0400
> >> >> >> > >
> >> >> >> > > The discussion thread for switching Streams from
Camel/osgi/Servicemix
> >> >> >> > to a single .war deployment
> >> >> >> >
> >> >> >> >
> >> >> >
> >> >
> >
 		 	   		  
Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message