Singletons and clusters

Hi,
Does anyone have any idea on what to do with singletons on a cluster ?
Right now we have 3 singleton, main business objects that handle core
method calls.
If we decide to cluster , how do we deal with singletons across clusters
-Sam

Does anyone have any idea of the performance implications of moving singleton
objects to RMI objects for clustering ?
We have an Environment Object that is encapsulated in a Servlet so that PageObjects
can be loaded at startup. I could easily make the Environment Object an RMI object
but I would be creating extra network traffic from each cluster participant.
thanks
Joe Ryan
Don Ferguson wrote:
Rickard Öberg wrote:
Hey
Bytecode wrote:
My point exactly.
A singleton replicated ?? Isnt that a conradiction in terms ??
What state are these different instances going to be in ?? Excatly identical
at all points of time ??If I understood Don right it's the stub that will be replicated. Not
the RMI-object. So all stubs in the clustered JNDI-tree will point to
the same RMI-object.
/RickardYes, that's the idea.
-Don

Similar Messages

Singleton and non singleton in webdynpro java

Hi All,
where we use singleton and non singleton in webdynpro application.please send me with any example (scenario).
Regards
Anand

Hello Anand,
I think that your question is for u201CWeb Dynpro Javau201D forum..
Please use the correct forum for your question.
In general:: http://wiki.sdn.sap.com/wiki/display/WDJava/SingletonandNon-Singleton+Nodes
There is example at : http://wiki.sdn.sap.com/wiki/display/WDJava/DynamicUIGeneration
u201CWelcome to Web Dynpro Java! u201C Wiki : http://wiki.sdn.sap.com/wiki/display/WDJava/WelcometoWebDynproJava%21
Thank you and best regards, Natalia Khlopina

Session Failover and Clustering

          Let's say that we have two WebServers (NES) with the weblogic plugin (say WS1 and WS2) and a cluster with two WebLogicCommerce AppServers (say AS1 and AS2). Let's assume that each WebServer and AppServer runs on its own machine (total: 4 machines). Now, let's assume that the WebServer "obj.conf" files (on both WS1 and WS2) are setup so that they point to the servers in the cluster (WebLogicCluster="AS1:7601,AS2:7601").
          When a new request comes in to one of the WebServers (say WS1), the plugin will route it to one of the AppServers using Round-Robin (say AS1). A session will now be initiated in AS1 and it sends a response back to the client.
          Question 1: How does the other proxy in WS2 know that all future requests for this client need to be forwarded to AS1?
          Question 2: For failover, does the cluster automatically replicate the session state existing in AS1 onto AS2 before sending the response (does AS2 automatically become the secondary)?
          Now let's assume that AS1 crashes/dies. When the next request from the client comes to WS1 or WS2, they will forward it to AS1 (assuming that WS2 knows about the client session in AS1) . Since AS1 has crashed, will the client eventually get a timeout error message?
          Question 3: To ensure that the session failover happens so that AS2 gets the request instead (becomes the primary), do we need to setup a WebLogic Proxy Server? If so, why can't the plugins for NES provide the failover themselves?
          Thank you very much for your help!
          Giri


          Thank you very much for your responses. It has been very helpful and I am clear on the session/clustering stuff. I have new questions on EJB and clustering which I will post as a separate thread.
          Giri
          "Jason Rosenberg" <[email protected]> wrote:
          >And also, if the browser has cookies disabled, it is important for
          >the app server to embed the WebLogicSession info via url rewriting,
          >otherwise the proxy or NES will not be able to route the session
          >properly.
          >
          >So, in all http responses, be sure to pass the url string through
          >response.encodeURL(). This will do the right thing depending
          >on whether cookies are enabled or not.
          >
          >I've just only recently figured this out. Haven't actually tried it
          >all out yet, so forgive me if it is not quite this simple, but this
          >seems to be the gist of it...
          >
          >Jason
          >
          >
          >"Justin James" <[email protected]> wrote in message news:[email protected]...
          >>
          >> Giri,
          >>
          >> I'm not a weblogic representative, but I tried to replicate this proxing service inside a load balancing switch(BigIP) and I
          >discovered a few things. The weblogic server sets a cookie (WebLogicSession)that the webserver plugin uses to manage the proxying.
          >The cookie (found in the HTTP header information) contains encoded information about the primary and secondary application servers
          >that the session is bound too. Any web server can read the cookie to determine how to dispatch the request to the primary server.
          >If the primary server does not respond, the request is forwarded to the secondary server by the plugin. Regardless of cluster size,
          >the session is replicated to only one other server.
          >>
          >> <[email protected]> wrote:
          >> >Giri Alwar wrote:
          >> >
          >> >> I need a couple of clarifications. First with regard to Question 1, I understand that plugins provide load balancing and
          >failover but what I really was asking is how the plugin in WS2 knows that a session for the client has been initiated in AS1 as a
          >result of WS1 sending the initial request to AS1. If WS2 gets a future request from the client, it needs to know this to send the
          >request to AS1. Does the plugin talk to the cluster to find out if there is a primary and who it is?
          >> >>
          >> >
          >> >> I should have clarified that my other questions pertain to in-memory replication. If I do not persist the session in a database
          >then does the client get an error message (timeout) when AS1 goes down (assuming we use NES with the WebLogic plugin)?
          >> >
          >> >Plugins' know how to route requests based on cookies. If it can't reach the primary server it will automatically try secondary.
          >In your case it doesn't matter if it reaches to proxy 1 or proxy 2, it is still the same.
          >> >
          >> >- Prasad
          >> >
          >> >> To prevent this error message and achieve failover, do I need to use WebLogic as the proxy server? If so, why isn't the NES
          >plugin doing this?
          >> >
          >> >> Thanks.
          >> >> Giri
          >> >>
          >> >> Prasad Peddada <[email protected]> wrote:
          >> >> >Giri Alwar wrote:
          >> >> >
          >> >> >> Let's say that we have two WebServers (NES) with the weblogic plugin (say WS1 and WS2) and a cluster with two
          >WebLogicCommerce AppServers (say AS1 and AS2). Let's assume that each WebServer and AppServer runs on its own machine (total: 4
          >machines). Now, let's assume that the WebServer "obj.conf" files (on both WS1 and WS2) are setup so that they point to the servers
          >in the cluster (WebLogicCluster="AS1:7601,AS2:7601").
          >> >> >>
          >> >> >> When a new request comes in to one of the WebServers (say WS1), the plugin will route it to one of the AppServers using
          >Round-Robin (say AS1). A session will now be initiated in AS1 and it sends a response back to the client.
          >> >> >>
          >> >> >> Question 1: How does the other proxy in WS2 know that all future requests for this client need to be forwarded to AS1?
          >> >> >
          >> >> > Plugin takes care of load balancing and failover, it is all transparent to the client.
          >> >> >
          >> >> >
          >> >> >> Question 2: For failover, does the cluster automatically replicate the session state existing in AS1 onto AS2 before sending
          >the response (does AS2 automatically become the secondary)?
          >> >> >
          >> >> > If you have only two yes it is automatically your secondary. Yes, replication is synchronous.
          >> >> >
          >> >> >>
          >> >> >> Now let's assume that AS1 crashes/dies. When the next request from the client comes to WS1 or WS2, they will forward it to
          >AS1 (assuming that WS2 knows about the client session in AS1) . Since AS1 has crashed, will the client eventually get a timeout
          >error message?
          >> >> >
          >> >> > If you are using some kind of persistence then you will be able to retrieve the session information and since the server
          >is not available the request will automatically failover.
          >> >> >
          >> >> >> Question 3: To ensure that the session failover happens so that AS2 gets the request instead (becomes the primary), do we
          >need to setup a WebLogic Proxy Server? If so, why can't the plugins for NES provide the failover themselves?
          >> >> >
          >> >> > No, you need only one. Either NES or weblogic proxy.
          >> >> >
          >> >> >
          >> >> >> Thank you very much for your help!
          >> >> >> Giri
          >> >> >
          >> >> >- Prasad
          >> >> >
          >> >
          >>
          >
          >

When not using EJBs can I make BD a Singleton and cache facade instances?

Hi,
In an application which does not use EJBs can I make BD(Business Delegate) a singleton?
I was very sure about doing this but when I tried Google on the same subject the answers were'nt supportive of this but that was in the context of applications which used EJBs. And also item 4 in Effective Java isnt very supportive of caching Objects at the drop of a hat.
When not using EJBs would it be an unnecessary thing to make BD a singleton and cahce Facade instances in a BD and DAO instances in a Facade? I am planning to use a array based blocking bounded buffer for the purposes of caching. Or would it be better to make both BD and a facade as SIngletons and just cache DAOs in a Facade?
Any suggestion would be of good help to me.
Thanks a lot.

Not sure I understand all your design, but you seem
to describe an architecture where requests are queued
and handled serially.Sorry if I messed up while explaining it. No, it will not be handled serially. Since the BD is a singleton multiple threads can pass messages to it simulteanously, a bit like an object of the Action class in Struts. Since I dont see having any synchronized methods in a BD requests will be handled simulteanously.
The impact on throughput of handling requests
serially (as opposed to parallelizing them) probably
outweights by far the cost of instantiating one more
object per request...Yes, I understand that but as I explained above the reqests wont be handled serially.
To be more clear, I am thinking of using any one of these two things:
1) BD(Singleton)-->Facade(Singleton, caches DAOs in a thread safe data structure)
2)1) BD(Singleton, caches Facade instances in a thread safe data structure)-->Facade(caches DAOs in a thread safe data structure).
the thread safe data structure I am planning to have is a array based bounded buffer which blocks using wait and notify mechanism.
Thank you for the reply.

Rac and clustering setup

Hi all,
I want to know abt RAC and clustering.Can any one describe me abt these two things.
I got two instances running in 2 AIX machines of size 25 gb each. im zero in this. can any one help me regarding this
first of all i do no what is RAC and Clustering
Regards
Elamaran

2 instances (or more) will indeed be reading and writing to the same database files. When one instance fails, the other one takes over. Therefore availability is ensured.
look here fore info on RAC:
http://download-uk.oracle.com/docs/cd/B19306_01/rac.102/b14197/toc.htm

Configuring Servers and Clusters

Hi All,
Whenever I am creatin HFM application,it is asking server,I have tried to "Configuring Servers and Clusters" ,but I am unable to coniger,please assist me.
Regards,
Prabhu

For your second question I think it is E. Coherence Web Edition
Because Coherence provides the ability to scale your application and also offers high availability by replicating the session objects to coherence nodes/caches.
Thanks,
Vijaya

Singletons and multithreading question

Hi, I had a class created as a singleton as it was only used in one place. However since Ive now multithreaded the part of the program that calls the singleton code so it could be called twice at the same time.
If the method being called initializes all its variables within the method, is that method safe to be called in a mutli-threaded env or not ?
If it in my situation all variables are declared within the method except a precompiled javax.xml.XPathExpression class, the same one is used to parse an Input Source, the input source itself is created within the method. I
think I am having concurrency problems , (SAXParser Exceptions) and I think it is because of this XPathExpression class but cannot clearly understand why.
What is the correct solution, I dont want to initilize the XPathExpression within the method because the method is called many times always for the same expression, dropping the singleton and creating new instances of the class on every call would be even worse. I require multithreading to improve concurrency is there a safe Multi Singleton pattern that i should be using ?

There are two issues:
a) is the singleton object safe for multithreaded use
b) how do I safely create a singleton in a multi-threaded application
The answer to (a) depends on the class. It sounds from your description that you are using an object (the XPathExpression) that might not be thread-safe. I can't tell you if it is or not. Assuming it is not you either need to synchronize this method so that all threads will go through it (and use the expression) one at a time - which may defeat the purpose of multi-threading - or else you need a way to use multiple instances of these expressions in different threads. As I'm not familiar with these classes I can't tell you what you can and can't do concurrently.
There are a number of answers to (b). Construction via static initialization is the simplest and safest. If you want lazy initialization and the class that holds the reference to the singleton must be loaded even if the singleton isn't needed, then lookup the initialize-on-demand-holder-class idiom (see Josh Bloch's "Effective Java" item 48)

Multiple call managers and clusters sharing one application server

Hi
I need to program my application such that it can be shared by multiple call managers and clusters. That is, I can only have one application server, and as a result I need to tell my application from which call manager / cluster this "request" is coming from.
My question is how do we do that? do we have access to some parameters / variables, similar to "#DEVICENAME#", which can be used to pass the call manager identity to the application?
TIA

Use multiple instances of your application with different URIs (assuming the instances will not have to talk to each other), then, when configuring the service on each cluster, simply reference the specific URI of the application instance for that cluster. eg: Using Tomcat to host the PhoneMessenger application for Cluster1 and Cluster2, Deploy the same application twice: PhoneMessengerForCluster1 and PhoneMessengerForCluster2. The service URL on Cluster1 might be: http://app_server/PhoneMessengerForCluster1/servlet/PhoneMessenger , and for Cluster2 it could be http://app_server/PhoneMessengerForCluster2/servlet/PhoneMessenger

What's the differences between a singleton and a class with static methods

Hi everybody.
My question is in the subject. Perhaps "differences" is not the good word. The positive and negative points ?
Imagine you have to write a connection pool, sure you gonna choose a singleton but why ?
Thank you very much.

A class is a class. Java doesn't have (and I wish it
did) a static outer class. Any class can extend
another class or implement an interface.A Class object cannot extend or implement anything - it is not under programmer control in any way. You can create a class which implements interfaces, but calling static methods is completely unrelated. Interfaces only affect instance methods, not class ones. I think all of your comparison to C++ is actually confusing you more. In Java, there is a real class object at runtime, as opposed to C++.
YATArchivist makes a good point about being able to
serialize, altho I've not met that desire in practice.
Maybe a concrete example would help.
mattbunch makes another point that I don't understand.
A class can implement an interface whether it sticks
its data in statics or in a dobject, so I guess I
still don't get that one.See my comment above. Static methods are free from all contractual obligations.
I prefer instance singletons to singleton classes because they are more flexible to change. For instance I created a thread pool singleton which worked by passing the instance around, but later I needed two thread pools so I made a slight modification to the class and poof! no more singleton and 99% of my code compiled cleanly against it.
When possible, I prefer to hand the instance off from object to object rather than have everything call Singleton.instance() since it makes changes like I mentioned earlier more feasible.

Singleton & EJB & Clustering

Hello:
although I have read information about this question in
http://java.sun.com/developer/technicalArticles/Programming/singletons/ ,
I continue having doubts.
I need a class (A) called by stateless session beans to get next value from one previous value.
This A class only performs queries from database ( using iBatis framework ) to get the next value, but doesn't store any value or state.
My doubt is if I can have this class as a Singleton class , considering that it works in a clustered enviroment.
As this A class is called by many session beans, I think that it's a good idea has a unique instance. But I have doubts about performance with this and the access to the resources ( only queries) .
Can I use, in this case, Singleton pattern in clustered enviroment ?
Thanks !

The only thing is that this sounds like a class which is being used as a generator for primary keys.
The concern is that you have two requests of the generator at the same time and, if the generator function is poorly coded, you then either have two non-unique primary keys, or somebody throws an exception.
I would look and see if your database backend has some sort of sequence/auto-inc/generator functionality and hook into that (no point re-inventing the wheel)
On the other hand, if you are trying to maintain some sort of database portability, you might be quicker having the inc larger than 1.
e.g.
on server#1: nextval = curval - (curval mod 3) + 3
on server#2: nextval = curval - (curval mod 3) + 4
on server#3: nextval = curval - (curval mod 3) + 5
[Note the above strategy is not the most efficient, but will guarantee that the three servers _never_ produce duplicate IDs... how you would code some such is more fun. It has the advantage that each server can lag behind in updates of the most recent value of curval without causing major issues]
Other people may know a better way of solving the problem.
If you are inventing your own wheel, i'd try testing it for duplicate values under heavy load as a critical test before release

Shared singleton across clusters?

          I have a singleton class that I would like to share across
          servers in a cluster. Simply storing it in the session object
          doesn't seem like the solution because user a and user b, each with their own
          copy of this object, could quickly get out of synch. Is there a way for me to
          bind this object to the jndi tree for the entire cluster so that everyone uses
          the same
          object? I know I could keep things in sync by storing the info
          in the database, but the whole reason for this object's existence is to avoid
          unnecessary trips to the database. Thanks!


Hi,
          Is the Coherence Cache ur talking abt useful for the fast reads ???
          We have situation with our application, where we need a distributed cache
          which handles fast reads .. and rite now we r looking for a product or
          implementation which does the same...And this cache is dynamic to some
          extend that the applications can change this cache rarely ..but application
          access this cache very very often.
          Right now without thinking abt cluster with a Single Server we can do this
          easily by using JNDI.. but distributing JNDI across cluster in weblogic have
          there own problems.. as we need to find a way to replicate the tree when
          there is a change to this cache.
          Right now thinking abt wrapping this cache with a Stateful Session Bean and
          I dont have a clue how weblogic will replicate the changes and also abt the
          fail over situations... Do u think is this a good approach or is there any
          other alternative?
          Hope to hear back from u guys..
          Regards and TIA
          Amar Pratap
          "Cameron Purdy" <[email protected]> wrote in message
          news:[email protected]...
          > > I do not think it is a good idea to bind data objects in JNDI
          >
          > It does work for certain cases on Weblogic, but replicated objects can
          > collide in Weblogic's JNDI implementation and the approach is not
          portable.
          >
          > > Moreover singleton objects are for
          > > a single VM and not for a cluster.
          >
          > It is a common requirement for applications that have to scale up to a
          > cluster. These applications typically use a singleton when they aren't
          > clustered, and it can provide immense performance improvements ... but
          when
          > clustering there are relatively few choices for caching across a cluster.
          > Besides our Coherence product, which is a clustered cache, there are
          several
          > options:
          >
          > 1) read-only ejb with JMS invalidation (Seppuku pattern that Dima
          designed)
          > 2) JNDI (may suffer from collision problems)
          > 3) RMI object (single point of failure and requires network traffic)
          >
          > > I would suggest you consider the database
          >
          > That's definitely the choice for transactional data (i.e. you don't want
          to
          > lose it no matter what), but for a cache, you are typically caching the
          data
          > that is in the database ... because the database is high-latency and
          > increasingly expensive as you scale.
          >
          > Applications that rely solely on the database and do not cache suffer
          > terribly when scaling.
          >
          > Peace,
          >
          > --
          > Cameron Purdy
          > Tangosol Inc.
          > Tangosol Coherence: Clustered Coherent Cache for J2EE
          > Information at http://www.tangosol.com/
          >
          >
          > "Rajesh Mirchandani" <[email protected]> wrote in message
          > news:[email protected]...
          > > I do not think it is a good idea to bind data objects in JNDI. They will
          > not be
          > > available to the other servers in the cluster. Moreover singleton
          objects
          > are for
          > > a single VM and not for a cluster. I would suggest you consider the
          > database
          > > approach.
          > >
          > > There are a lot of discussions in this newsgroup. Search for previous
          > postings.
          > >
          > > sdb wrote:
          > >
          > > > I have a singleton class that I would like to share across
          > > > servers in a cluster. Simply storing it in the session object
          > > > doesn't seem like the solution because user a and user b, each with
          > their own
          > > > copy of this object, could quickly get out of synch. Is there a way
          for
          > me to
          > > > bind this object to the jndi tree for the entire cluster so that
          > everyone uses
          > > > the same
          > > > object? I know I could keep things in sync by storing the info
          > > > in the database, but the whole reason for this object's existence is
          to
          > avoid
          > > > unnecessary trips to the database. Thanks!
          > >
          > > --
          > > Rajesh Mirchandani
          > > Developer Relations Engineer
          > > BEA Support
          > >
          > >
          >
          >

Apex Listener and Clustering - WebLogic Server

Has anybody worked with the apex listener on a clustered WebLogic environment with a load balancer? We are having lots of connection issues, and I am wondering if this could be part of the problem.
Thanks,
Michelle

Someone Kindly try to give me reply for my issue .
Thanks& Regards,
Nitesh pareek.

WSS security and clustering in SOA Suite 11g

We are calling one composite from another as a reference. Both are on the same clustered domain.
We are using AIA WSDLs so there is a default WSS username password token as policy.
While on the single server domain, we do not need to pass the credentials since request was going on the same Weblogic server and it was running fine.
But when we moved to the clustered domain, we changed the endpoints in composites to an external hardware load balancer and now it needs the sercurity credentials because the request is coming via the load balancer running on the different machine.
I have two queries:
1. Is there a way we can skip external load balancer while calling composites within the same clustered domain and still be able to route the requests to all managed servers instead of only one? (may be there is a way we can provide multuple URLs in the composite somehow or any other setting on cluster which would enable us to send all the requests to a common hostname which will route it to multiple managed servers in round robin)
2. If above is not possible then we will have to use the external load balancer, in which case I need to set username password tokens while calling different composites. I need help in configuring this. It would be nice if anyone can provide me a link to a sample may be.
Thanks

In 11g, Oracle WSM Agent has been integrated into Oracle WebLogic Server. Please refer below link for more details -
http://docs.oracle.com/cd/E15523_01/web.1111/b32511/intro.htm
Regards,
Anuj

Configuartion data and clusters

When coding for test rigs, it is normally required to save a large amount of configuration parameters that will be used during the test sequence. And these have to be saved in the hard disk.
The only option seems to be to save these parameters in clusters and write to disk. I am talking of something like 40 to 60 values - single precision, boolean, bytes and even strings all mixed.
And when the test is running, these clusters are read from disk, unbundled and used in various functions for comparison, limit values and the like.
Obviously this involves intensive wiring to bundle the clusters and unbundle them. Are there any other options that are more elegant ?
Thanks
Raghunathan
PS : I had posted the same query today morning but somehow it had vanished ! Hence posting again.
Raghunathan
LV2012 to Automate Hydraulic Test rigs.

I think there are various ways to go about that. It is not completely clear to me what is bugging you the most, but it seems like you think it takes too much code to manage configuration data. First of all, I would like to point out the following: when using datalog or configuration file VI's, you do not need to cast any data: these VI's will do that automatically for you. Datalog files seem to be the most powerful because you can write and read complete complex clusters to file and read complete clusters as well.
here I got distracted by somebody and got a bunch of beers in lovely downtown Amsterdam....
Back behind my computer I notice your question got answered already. I would like to add a suggestion to the one VI solution:
Use two enums: one for read or write and one for all the "properies"/configurations. This will only work if you don't have to many different data types though. It is the cleanest solution I can think of, loading all settings into an uninitialized shift register (at startup) and calling that vi anywhere in your code where you need to read or write values. If you have many diffrent datatypes, you may consider embedding this vi inside new vi's that just do more specific tasks.
Or maybe I just had to much to drink..
Aart-Jan

Configuration data and clusters

While coding in LV for product test rig applications there is this need to save configuration data to disk and retrieve it while running the test. A typical scenario to test product "AA" will be :
1. There are 10 different variations of the basic product AA_1 to AA_10.
2. Each variant has 40 odd parameters that can vary.
3. User enters the values for these variants and will have to be saved on disk.
4. When a particular model is tested, all 40 parameters for it have to be read from disk and used in the various tests.
In such events the only option seems to be :
(*) Collect all the user entered parameters into clusters. If there are totally 8 tests for each model, then have 8 clusters and bundle parameters pertaining to each test into that cluster.
(*) Write the clusters to disk in a bit stream using "Write File.VI"
(*) When running the test, retrieve the parameters from disk and load them into the relevant clusters back. Unbundle by name and use each of the parameters as required. And if the same parameter is required to be used in many functions in the same test, then assign it to a local variable (my beloved ;-)) and use it.
I would like experienced LV professionals to comment on this method.
Handling the clusters and wiring them for large data storing and retrieval calls for heavy wiring. Is there no way out ?
Thanks
Raghunathan
Raghunathan
LV2012 to Automate Hydraulic Test rigs.

Hi Raghunathan,
Since you have to deal with so many elements of data, it looks like clusters will be your best method. It will simplify your file I/O and will also help you keep the data organized. I'm not sure why you must write the data to disk, but if it is necessary, clusters will be helpful. It's true that there will be many wires when you bundle and unbundle those clusters, but that will happen no matter how you deal with large numbers of parameters. It will probably make your code more readable if you write subVIs to take the clusters as input, and then break apart the clusters within the subVIs.
I hope this is helpful!
Megan B.
National Instruments

Singletons and clusters

Similar Messages

Maybe you are looking for