I have developed my application with Akka on a single JVM. Now I want to distribute the workload across many machines. I've started to read the documentation and got confused.
There are two ways of making Akka app to be distributed by clustering and remoting.
I don't get the difference between the two.
If I understand properly both are excluding themselves mutually since in configuration one need to use different provider for actor reference:
akka.remote.RemoteActorRefProvider
akka.cluster.ClusterActorRefProvider
So what are the use cases? When I would choose one instead of the other?
Maybe clustering is something like superset of remoting or maybe it is the other way around?
They are not mutually exclusive since clustering is implemented on top of remoting. The main feature of remoting is location transparency for ActorRefs. Clustering adds distributed membership on top of that. It is rarely useful to use remoting directly, clustering is preferred for most of the use cases.
Always best to look to the code
https://github.com/akka/akka/blob/c2983c7225eeaa035af99e0affeb5f5c841668c4/akka-cluster/src/main/scala/akka/cluster/ClusterActorRefProvider.scala
private[akka] class ClusterActorRefProvider ... extends RemoteActorRefProvider
First of all ClusterActorRefProvider is just extension of the RemoteActorRefProvider
Core features of the cluster-akka added to the remoting are:
cluster wide failure detection
routers with rootes group & pool capabilities
you'll find flavour of all this addons in the source code.
Related
I have an interesting conundrum. I have been challenged with identifying the most suitable process in which to create a "browser front-end" to an existing multi-user application built within the TigerLogic/Pick D3 environment. My research indicates there are many ways to do this; but I am struggling to decide which method is best or where to start. I have "played" with a few technologies but a commitment to one is needed to get started.
These methods include:
Creation of a complex webservice using MVS Toolkit; and engineering a client from WSDL either from scratch or using maven/wsimport. Tests indicate there is a lot more to this process than originally though for a simply WSDL.
Development of a Java based web app that harnesses the MVSPJavaAPI - I am not a JAVA developer so this means learning a new language. Development would most likely take place in Eclipse.
Using TigerLogics FlashCONNECT - resulting in additional expenditure to clients so not preferable - and more or less ruled
out.
There is also the .NET option - but I have ruled this out on the basis of needing portability.
My question is, has anyone else out there done anything like this and could you share your experience? My first task is to build a web-app that will reliably give me the D3 TCL prompt in a browser that I can customise.
I am not sure there is a definitive answer here but would like peoples thoughts and will label the most useful as the answer.
What path you choose depends in some part on your existing skill-set and whether that fits in with your portability needs. It is very difficult to give you a concrete answer to your question becaue of not knowing in which part of the chain you need the portability.
It is however possible to develop a web-browser front-end using .NET which will run on Linux or Windows, so I don't see an issue with portability here. Your web server will have to be windows based but it shouldn't matter whether D3 is running on Linux or Windows at the server-end, or whether the client desktops are running Linux or Windows.
You could try TigerLogics MVSP .NET API but I do not know if it has the power to deliver based on your needs. I believe you may find mv.NET from Bluefinity could fulfill your needs. This is in my opinion the leading product on the MultiValue market for achieving the goals you have in mind. This will mean spending money of course. For this you will get a very powerful set of tools. Also, the cost of investing in a good tool could end up smaller than the cost in terms of time, effort and potential complications of trying to do this piecemeal without spending any extra money. I am sure Flashconnect would also do the job. You would have to weigh up the cost of the different options to find out which one is right for you both technically and financially.
Not knowing whether or not you have .NET in your skill-set, I don't know whether the .NET option would be easier for you. It is however technically possible.
I would suggest using Rocket's D3 (formerly TigerLogic D3) .NET APIs and create a Web API RESTful service that you can consume with JavaScript in any other web technology and if you need to call from a D3 subroutine (in case you would) then use the MVS Toolkit.
Requirements though are D3 9.0 or later.
I've used all of the technologies described, and many more to interface with D3. I agree with #Glenn and will add... I understand you're edging away from .NET. That's fine, you don't need it. But consider that most LAMP implementations separate the DBMS servers from the web servers. That topology introduces short delays between the tiers but decouples them in case you want to use multiple web servers or multiple databases - a common topology even with D3 / MV.
I have a client where we have a Java/Grails front-end over Linux, with all data queries filtered through a single, elegant data provider class that's abstracted from application logic. That uses a web service call which I wrote in Java, calling to a .NET web service. The service is easily generated/modified, as is the client from the WSDL. From there IIS carries inbound queries to D3 via mv.NET, and at this point it doesn't matter if the D3 DBMS is in Linux or Windows. My web service could have as easily been in Linux with Java but it would then lack a pooling mechanism - see below.
If you want all Linux then you can go with the MVSP Java library. TigerLogic (now acquired by Rocket Software) committed to a PHP binding for MVSP some months ago. Rather than wait, one of my clients created a PHP wrapper around mv.NET, though MVSP is as easy. So the resulting application is essentially LAMP, but with the M = Multivalue. I have written code like this too - we can write a wrapper in any language which exposes a useful API and abstracts both connectivity method and OS dependencies. In other words it doesn't matter what languages we want to use or what OS's are involved. That part is rather trivial and subject to change later. It's better to focus on the application than the communications.
You can also go off the menu, so to speak, and create your own Java/PHP wrapper around the OS-level d3tcl command, which is a script/wrapper around the d3 executable. This allows you to open a connection yourself and pass in commands.
Whatever option you select, you need to consider that opening and closing a DBMS connection is a slow process. You do not want to script a login around every data request. You do want to open a connection and keep that open persistently, while your client code accesses and releases that persistent connection as required. This is why we like mv.NET and FlashCONNECT. With MVSP and other mechanisms you need to create your own persistence model. You'll also need to manage a pool of connection resources - what happens when you get 10 simultaneous queries, or just 1 short one after one long one? You don't want queries to back up, you don't want to reject or timeout connections, and you don't want to fire up a connection for every client. You do want the proper number of DBMS sessions waiting for inbound connections. mv.NET and FlashCONNECT do this for you, the others do not.
Personally I'd shy away from FlashCONNECT. I was there for its initial development and testing and for years of end-user implementations. It's not as widely used as the other options and is more a tool for those who aren't familiar with other options. If you're talking about Java then you're probably not inclined to use FlashCONNECT. That said, if you have developers who are not familiar with anything outside D3 then FlashCONNECT is a decent server-side tool for them while someone else is focusing on the client-side with other technologies. Everyone should use their best skillset.
Finally, (already?) if someone is not familiar with external technologies, and more intimate with D3, then other options exist like DesignBAIS and Viságe, mostly removing the burden of communications and allowing developers to work on the client-side features and back-end rules in BASIC.
I discuss all of these topics plus mobile and telephony on my blog.
HTH
Systems which require communication between disparate applications is quite commmon.
What is the normal architecture for such applications? Am I right in thinking that web services are the usual tool (if so, what reasons why?).
Also, what other considerations are involved?
Thanks
There is no normal architecture and web services is certainly one way to link them up anda popular one. But the communications technllogy choice is a trivial part of this problem.
Who is communicating what, when and why, is the head scratcher, and of course we can't answer that in anything but the most general of terms.
You need to start with the type of interactions between the applications.
Just retrieving data from another aplication
Requesting another application to perform some function, no feedback of any kind required
Transactional. Do something with this and return a response (synchronous and asynchronous).
Are these applications yours? If so you should head towards SOA. ie you take the processing out of the application, make it a service and then request it. All your other apps then can also request it.
However you go there is no easy solution, because these applications were not designed to work togther.....
Akka Typed actors are created in two parts using JDK proxies, whereby the proxy is a product of a specified interface, and the implementation forms the backing managed instance. However this means of construction prevents a TypedActor from implementing multiple Types (interfaces).
I thought I had read someplace that Akka 2.0 was going to change this. Does anyone have any thoughts on this, or how to workaround? FYI, I am using Akka in pure Java, not Scala at this stage
Typed Actors in pre-2.0 are implemented using aspect weaving, and are thus not JDK proxies.
Typed Actors in 2.x are based on JDK proxies and you can essentially use as many interfaces as is supprted by the JDK: Supercharging
There is an official opinion that Typed actors are not the best (see When_to_use_Typed_Actors). If it is possible try to use Untyped actors with typed messages.
We have been using the messages of the kind:
class Contact<T>
class Signal<T>(contact:Contact<T>, data:T)
The instance of contact is easy to check on equality. (if-elseif-elseif) Usually a map of contact-handlers is sufficient to handle all inputs.
The idea of strictly typed signals is further developed in SynapseGrid library. It defines Builder to associate typed handlers with typed contacts.
I'm trying to write up a tool that requires knowledge of the state of other machines in a cluster (local LAN). This is for a network failover/high availability system similar to VRRP and corosync/openais, but I wish to contain more information (such as near real-time speed/performance characteristics) so devices can make more intelligent choices. This means using a protocol more complicated than a predetermine weight-based mechanism: by allowing all clustered machines to see the state of each other, they can communally agree on which is the most suitable to be the master device.
From my searches, I haven't found any (C, C++ or JavaME) libraries that offer a distributed state mechanism. Ideally, I'm looking for something that broadcasts/multicasts each individual machines state periodically so participating machines can build up a global state table and all can see who the master should be. State in this case is arbitrary key/value pairs.
I'd rather not re-invent any wheels so am curious to know if anyone here can point me in the right direction?
If I were you I'd investigate memcached (memcached.org) or one of the nosql variants.
It sounds like Apache ZooKeeper might be a good match. It's distributed, hierarchical key-value store. To quote their Overview page:
ZooKeeper was designed to store coordination data: status information, configuration, location information, etc.
Here's an example of a simple Leader Election recipie, although it would require adaptation to determine a leader by some weighted criterion.
I'm not sure if there is any application for your purpose or not.
But I know that you can write a simple program with MPI library and broadcast any information that you want.
all client's can send their state to root node, and the root node then broadcast the message.
functions that you need for this are:
MPI_Bcast
MPI_Send
MPI_Recv
there is lots of tutorial on C++/MPI on net, just google it!
Even though I've been in Java SE for quite some time now, I started EE & web w/ Java only about a month ago, so pardon if the question seems a bit noobish...
So here's the situation: I'm trying to write a JS based multi-player game with real-time interaction (let's say chess in this example, though it really doesn't matter what particular game it is, could be tennis or w/ever). The clients would interact with the server through JS calls, sending the move etc. Now, while I could just receive the move from one client & pass it straight on to the other player, not maintaining the game state on the server would mean putting a huge sign out saying "user JS scripts welcome" (and that's out of experience -- "hacked" a crapload of that kind myself). This brings me to my problem -- how do I share a stateful object between several sessions? One idea that came to mind was a singleton storing a Hashmap of stateful beans & then each session could retrieve the bean by it's hash, but I've no idea how right that is (and it seems rather complex for a fairly common thing like that). Tieing it to application scope seems overkill as well...
P.S. I do understand that the object would need concurrency managing etc, I just can't seem to put my finger on how to get it shared...
EDIT: I'm sorry I didn't mention it before -- using Glassfish, EE6.
You have a business process scenario which is defined according to Seam framework documentation as follows
The business process spans multiple interactions with multiple users, so this state is shared between multiple users, but in a well-defined manner. The current task determines the current business process instance, and the lifecycle of the business process is defined externally using a process definition language, so there are no special annotations for business process demarcation.
Here you can see a Seam business process management Tutorial
Notice Seam uses JBoss BPM behind the scenes to handle its business process context. If you just want to use plain JBoss BPM capabilities, you can see here how to integrate with JBoss
See also JBoss BPM User guide
Solved. Shared it via ServletContext, which I initially thought wouldn't work 'cause FacesServlet is a separate one, thought it has smthn like a different container either.