Replication Architecture Question

Hello,
I have a problem with identifying the architecture for my materialized view environment. Let me tell you what I want to achieve:
I have a production database that is the source (master site) of my data. Portions of that data gets replicated to materialized view sites on a daily basis. The materialized view sites get accessed through the internet. The problem is that after the refresh more or less extensive calculations have to be done. during that time the materialized view site is not ready to be accessed via internet, because some data tables are not yet filled (they get filled during the calculations). That means there is a downtime of the system.
In order to eliminate this downtime I want to have 2 instances of the materialized view site, which get replicated sequentially, e.g. mv1 gets updated every 48 hours as well as mv2, but there is a time shift of 24 hours between the refresh of each materialized view site.
The webapplication accesses either mv1 or mv2. (it accesses mv1 while mv2 is updated and vice versa).
This approach will eliminate the downtime of the system.
But now i have a little problem with understanding the materialized view concepts. If mv1 is refreshed do all the entries in the materialized view logs at the master site get emptied? If so then I have a problem when i want to refresh mv2, because then only the changes since refresh of mv1 are present in the materialized view logs at the master site. so i will loose all the changes made to the database that already got replicated to mv1.
Is there a way of achieving the scenario mentioned above? What I want to have are separate materialized view logs for each of the materialized view sites.
Does anyone know how can I do this or has an alternative approach that fits to the above requirements?
Any suggestions are welcomed.
best regards
Mirko

Hi Justin,
Thank you for your reply. If I understand you correctly, the oracle replication mechanism automatically handles the situation of various materialized view sites that refresh from one master site at different times. Easily spoken, that means that oracle automatically stores all changes on the master site in one materialized view log for each table, but keeps track, which entry was already sent to the individual materialized view site (just as if there were one materialized view log for each materialized view site). So I don't have to worry about anything, when i want to refresh more than one materialized view site from one master site? All Materialized view sites will get the correct data?
If so then it sounds good.
I have another short questions for clarification of the materialized view technique:
What happens when the materialized view site refreshs and queries are sent to the materialized view site at the same time? What data will be used to perform the query?
My assumption is that until the refresh group (all mvs are within one refresh group) completed the update, the user (that initiated the query) sees the old data. Is that correct? Does that user notice (performance related) that the materialized view site is currently refreshed?
Regarding the calculation after refresh: Yes I have to do it after the refresh was made, because i have to call java stored procedures, that do the calculation. so i can't do it while creating / refreshing the materialized views.
regards
mirko

Similar Messages

Oracle VM Server for SPARC - network multipathing architecture question

This is a general architecture question about how to best setup network multipathing
I am reading the "Oracle VM Server for SPARC 2.2 Administration Guide" but I can't find what I am looking for.
From reading the document is appears it is possible to:
(a) Configure IPMP in the Service Domain (pg. 155)
- This protects against link level failure but won't protect against the failure of an entire Service LDOM?
(b) Configure IPMP in the Guest Domain (pg. 154)
- This will protect against Service LDOM failure but moves the complexity to the Guest Domain
- This means the there are two (2) VNICs in the guest though?
In AIX, "Shared Ethernet Adapter (SEA) Failover" it presents a single NIC to the guest but can tolerate failure of a single VIOS (~Service LDOM) as well as link level failure in each VIO Server.
https://www.ibm.com/developerworks/mydeveloperworks/blogs/aixpert/entry/shared_ethernet_adapter_sea_failover_with_load_balancing198?lang=en
Is there not a way to do something similar in Oracle VM Server for SPARC that provides the following:
(1) Two (2) Service Domains
(2) Network Redundancy within the Service Domain
(3) Service Domain Redundancy
(4) Simplify the Guest Domain (ie single virtual NIC) with no IPMP in the Guest
Virtual Disk Multipathing appears to work as one would expect (at least according the the documentation, pg. 120). I don't need to setup mpxio in the guest. So I'm not sure why I would need to setup IPMP in the guest.
Edited by: 905243 on Aug 23, 2012 1:27 PM

Hi,
there's link-based and probe-based IPMP. We use link-based IPMP (in the primary domain and in the guest LDOMs).
For the guest LDOMs you have to set the phys-state linkprop on the vnets if you want to use link-based IPMP:
ldm set-vnet linkprop=phys-state vnetX ldom-name
If you want to use IPMP with vsw interfaces in the primary domain, you have to set the phys-state linkprop in the vswitch:
ldm set-vswitch linkprop=phys-state net-dev=<phys_iface_e.g._igb0> <vswitch-name>
Bye,
Alexander.

Architecture question, global VDI deployment

I have an architecture question regarding the use of VDI in a global organization.
We have a pilot VDI Core w/remote mysql setup with 2 hypervisor hosts. We want to bring up 2 more Hypervisor hosts (and VDI Secondaries) in another geographic location, where the local employees would need to connect desktops hosted from their physical location. What we don't want is to need to manage multiple VDI Cores. Ideally we would manage the entire VDI implementation from one pane of glass, having multiple Desktop Provider groups to represent the geographical locations.
Is it possible to just setup VDI Additional Secondaries in the remote locations? What are the pros and cons of that?
Thanks

Yes, simply bind individual interfaces for each domain on your web server,
one for each.
Ensure the appropriate web servers are listening on the appropriate
interfaces and it will work fine.
"Paul S." <[email protected]> wrote in message
news:407c68a1$[email protected]..
>
Hi,
We want to host several applications which will be accessed as:
www.oursite.com/app1 www.oursite.com/app2 (all using port 80 or 443)
Is it possible to have a separate Weblogic domain for each application,all listening
to ports 80 and 443?
Thanks,
Paul

Running MII on a Wintel virtual environment + hybrid architecture questions

Hi, I have two MII Technical Architecture questions (MII 12.0.4).
Question1: Does anyone know of MII limitations around running production MII in a Wintel virtualized environment (under VMware)?
Question 2: We're currently running MII centrally on Wintel but considering to move it to Solaris. Our current plan is to run centrally but in the future we may want to install local instances local instances of MII in some of our plants which require more horsepower. While we have a preference for Solaris UNIX based technologies in our main data center where our central MII instance will run, in our plants the preference seems to be for Wintel technologies. Does anybody know of any caveats, watch outs or else around running MII in a hybrid architecture with a Solarix Unix based head of the hybrid architecture and the legs being run on Wintel?
Thanks for your help
Michel

This is a great source for the ins/outs of SAP Virtualization: https://www.sdn.sap.com/irj/sdn/virtualization

Architectural question

Little architectural question: why is all the stuff that is needed to render a page put into the constructor of a backing bean? Why is there no beforeRender method, analogous to the afterRenderResponse method? That method can then be called if and only if a page has to be rendered. It seems to me that an awful lot of resources are waisted this way.
Reason I bring up this question is that I have to do a query in the constructor in a page backing bean. Every time the backing bean is created the query is executed, including when the page will not be rendered in the browser...

Little architectural question: why is all the stuff
that is needed to render a page put into the
constructor of a backing bean? Why is there no
beforeRender method, analogous to the
afterRenderResponse method? That method
can then be called if and only if a page has to be
rendered. It seems to me that an awful lot of
resources are waisted this way.There actually is such a method ... if you look at the FacesBean base class, there is a beforeRenderResponse() method that is called before the corresponding page is actually rendered.
>
Reason I bring up this question is that I have to do
a query in the constructor in a page backing bean.
Every time the backing bean is created the query is
executed, including when the page will not be
rendered in the browser...This is definitely a valid concern. In Creator releases prior to Update 6 of the Reef release, however, there were use cases when the beforeRenderResponse method would not actually get called (the most important one being when you navigated to a new page, which is a VERY common use case :-).
If you are using Update 6 or later, as a side effect of other bug fixes that were included, the beforeRenderResponse method is reliably called every time, so you can put your pre-rendering logic in this method instead of in the constructor. However, there is still a wrinkle to be aware of -- if you navigate from one page to another, the beforeRenderResponse of both the "from" and "to" pages will be executed. You will need to add some conditional logic to ensure that you only perform your setup work if this is the page that is actually going to be rendered (hint: call FacesContext.getCurrentInstance().getViewRoot().getViewId() to get the context relative path to the page that will actually be displayed).
One might argue, of course, that this is the sort of detail that an application should not need to worry about, and one would be absolutely correct. This usability issue will be dealt with in an upcoming Creator release.
Craig McClanahan

BPEL/ESB - Architecture question

Folks,
I would like to ask a simple architecture question;
We have to invoke a partner web services which are rpc/encoded from SOA suite 10.1.3.3. Here the role of SOA suite is simply to facilitate communication between an internal application and partner services. As a result SOA suite doesn't have any processing logic. The flow is simply:
1) Internal application invokes SOA suite service (wrapper around partner service) and result is processed.
2) SOA suite translates the incoming message and communicates with partner service and returns response to internal application.
Please note that at this point there is no plan to move all processing logic from internal application to SOA suite. Based on the above details I would like get some recommedation on what technology/solution from SOA suite is more efficient to facilate this communication.
Thanks in advance,
Ranjith

You can go through the design pattern called Channel Adapter.
Here is how you should design - Processing logic remains in the application.. however, you have to design and build a channel adapter as a BPEL process. The channel adapter does the transformation of your input into the web services specific format and invoke the endpoint. You need this channel adapter if your internal application doesn't have the capability to make webservice calls.
Hope this helps.

Architecture Question...brain teasing !

Hi,
I have a architecture question in grid control. So far Oracle Support hasnt been able to figure out.
I have two management servers M1 and M2.
two VIP's(Virtual IP's) V1 and V2
two Agents A1 and A2
the scenerio
M1 ----> M2
| |
V1 V2
| |
A1 A2
Repository at M1 is configured as Primary and sends archive logs to M2. On the failover, I have it setup to make M2 as primary repository and all works well !
Under normal conditions, A1 talks to M1 thru V1 and A2 talks to M2 thru V2. No problem so far !
If M1 dies, and V1 forwards A1 to M2 or
if M2 dies, V2 forwards A2 to M1
How woudl this work.
I think (havent tried it yet) but what if i configure the oms'es with same username and registration passwords and copy all the wallets from M1 to M2
and A1 to A2 and just change V1 to V2. Would this work ????
please advice!!

SLB is not an option for us here !
Can we just repoint all A1 to M2 using DNS CNAME change ??

Inheritance architecture question

Hello,
I've an architecture question.
We have different types of users in our system, normal users, company "users", and some others.
In theory they all extend the normal user. But I've read alot about performance issues using join based inheritance mapping.
How would you suggest to design this?
Expected are around 15k normal users, a few hundred company users, and even a few hundred of each other user type.
Inheritance mapping? Which type?
No inheritance and append all attributes to one class (and leave these not used by the user-type null)?
Other ways?
thanks
Dirk

sorry dude, but there is only one way you are going to answer your question: research it. And that means try it out. Create a simple prototype setup where you have your inheritance structure and generate 15k of user data in it - then see what the performance is like with some simple test cases. Your prototype could be promoted to be the basis of the end product if the results or satisfying. If you know what you are doing this should only be a couple of hours of work - very much worth your time because it is going to potentially save you many refactoring hours later on.
You may also want to experiment with different persistence providers by the way (Hibernate, Toplink, Eclipselink, etc.) - each have their own way to implement the same spec, it may well be that one is more optimal than the other for your specific problem domain.
Remember: you are looking for a solution where the performance is acceptable - don't waste your time trying to find the solution that has the BEST performance.

Enterprise Manager 11g Sybase Plugin architecture question

Hi,
have successfully installed and configured Grid 11g on RedHat Enterprise 5.5. Deployed and configured agents to solaris and linux environments..so far so good.
However, we're going to test the Sybase ASE plugin to monitor ASE with EM. My question is a simple one and I think I know the answer but I'd like to see what you guys think of this.
We'd like to go with a simple centralised agent rather than one agent/plugin per sybase machine, atleast for the tests. No doubt there may be pro's (first one clearly being one of a single point of failure - welll we can live with this for now) to this approach. My instinct is to install the oracle agent/plugin on a machine other than the grid machines itself, however the question arose - why not install the ASE plugin on the grid infrastructure machine agents themselves? Pros and cons?
The architecture we have currently : repository database configured to failover between 2 redhat boxes. 2 OMS running 1 on each of these boxes configured behind SLB using nfs based shared upload directory. One 'physical agent' running on each box. Simple for now. But I have the feeling , given that the Sybase servers will communicate or be interrogated via the sybase plugin directly to the grid infrastructure machines placing load etc on them , and in case of problems might interfere with the healthy running of the grid. Or am I being over cautious?
John
Edited by: user1746618 on 12-Jan-2011 09:01

well I have followed the common sense approach and avoided the potential problem by installing on a remote server and configuring the plugin on this.
Seems to be working fine and keeps the install base clean..

Three tier architecture questions

Hello,
My question is in regards to using Toplink in a three tier architecture situation. If I wish to send an object A which has a collection of Bs and B has a collection of C ( A nested object structure with two or more levels of indirection). Is the best solution to have the named query be part of a unit of work so that even if on the client side somebody unknowingly were to make the modification to one of the entity objects ( a POJO) the shared session cache would not be affected ?
This is assuming the client side HTTP layer and the RMI/EJB layer are on different JVMs.
Some of the other suggestions I have heard is to retrieve it from the shared session cache directly and if in case I need to modify one or more of the objects do a named query lookup on that object alone and then proceed to register that object in a unit of work and then commit the changes.
Also the indirection would have to be utilised before the data objects are sent to the Servlet layer I presume ?(That is if I do a a.getAllOfBObjects() on the servlet side I would get a nullpointer exception unless all of B were already instatiated on the server side). Also when the objects are sent back to the server do I do a registerObject on all the ones that have changed and then do a deepMergeClone() before the uow.commit() ?
Thanks,
Aswin.

Aswin,
If your client is remote to the EJB tier then all persistent entities are detached through serialization. In this architecture you do not need to worry about reading and modifying the shared instance as it never the one being changed on the client (due to serialization).
Yes, you do need to ensure that all required indirect relationships are instantiated on the server prior to returning them from the EJB call.
Yes, you do need to merge the changes of the detached instance when returned to the server. I would also recommend first doing a read for the entity being merged (by primary key) on the new UnitOfWork prior to the merge. This will handle the case where you are merging into a different node of the cluster then where you read as well as allowing you to check for the case where the entity no longer exists in the database (if the read returns null then the merge will result in an INSERT and this may not be desired).
Here is an example test case that does this:
    public void test() throws Exception {
        Employee detachedEmp = getDeatchedEmployee("Jill", "May");
        assertNotNull(detachedEmp);
        // Remove the first phone number
        PhoneNumber phone = detachedEmp.getPhoneNumber("Work");
        assertNotNull("Employee does not have a Work Phone Number",
                      detachedEmp.getPhoneNumber("Work"));
        detachedEmp.removePhoneNumber(phone);
        UnitOfWork uow = session.acquireUnitOfWork();
        Employee empWC = (Employee) uow.readObject(detachedEmp);
        if (empWC == null) { // Deleted
            throw new RuntimeException("Could not update deleted employee: " + detachedEmp);
        uow.deepMergeClone(detachedEmp);
        uow.commit();
     * Return a detached Employee found by provided first name and last name.
     * Its phone number relationship is instantiated.
    public Employee getDeatchedEmployee(String firstName, String lastName) {
        ReadObjectQuery roq = new ReadObjectQuery(Employee.class);
        ExpressionBuilder builder = roq.getExpressionBuilder();
        roq.setSelectionCriteria((builder.get("firstName").equal(firstName)).and(builder.get("lastName").equal(lastName)));
        Employee employee = (Employee)session.executeQuery(roq);
        employee.getPhoneNumbers().size();
        return (Employee)SerializationHelper.serialize(employee);
    }One other note: In these types of application optimistic locking is very important. You should also make sure that the locking field(s) are mapped into the object and not stored only in the TopLink cache. This will ensure the locking semantics are maintained across the detachment to the client and the merge back.
Doug

Architecture question...where to put the code

Newbie here, so please be gentle and explicit (no detail is
too much to give or insulting to me).
I'm hoping one of you architecture/design gurus can help me
with this. I am trying to use good principals of design and not
have code scattered all over the place and also use OO as much as
possible. Therefore I would appreciate very much some advice on
best practices/good design for the following situation.
On my main timeline I have a frame where I instantiate all my
objects. These objects refer to movieClips and textFields etc. that
are on a content frame on that timeline. I have all the
instantiation code in a function called initialize() which I call
from the content frame. All this works just fine. One of the
objects on the content frame is a movieClip which I allow the user
to go forward and backward in using some navigation controls.
Again, the object that manages all that is instantiated on the main
timeline in the initialize() function and works fine too. So here's
my question. I would like to add some interactive objects on some
of the frames of the movieClip I allow the user to navigate forward
and backward in (lets call it NavClip) . For example on frame 1 I
might have a button, on frame 2 and 3 nothing, on frame 4 maybe a
clip I allow the user to drag around etc. So I thought I would add
a layer to NavClip where I will have key frames and put the various
interactive assets on the appropriate key frames. So now I don't
know where to put the code that instantiates these objects (i.e.
the objects that know how to deal with the events and such for each
of these interactive assets). I tried putting the code on my main
timeline, but realized that I can't address the interactive assets
until the NavClip is on the frame that holds the particular asset.
I'm trying not to sprinkle code all over the place, so what do I
do? I thought I might be able to address the assets by just
providing a name for the asset and not a reference to the asset
itself, and then address the asset that way (i.e.
NavClip["interactive_mc"] instead of NavClip.interactive_mc), but
then I thought that's not good since I think there is no type
checking when you use the NavClip["interactive_mc"] form.
I hope I'm not being too dim a bulb on this and have missed
something really obvious. Thanks in advance to anyone who can help
me use a best practice.

1. First of all, the code should be:
var myDraggable:Draggable=new Draggable(myClip_mc);
myDraggable.initDrag();
Where initDrag() is defined in the Draggable class. When you
start coding functions on the timeline... that's asking for
problems.
>>Do I wind up with another object each time this
function is called
Well, no, but. That would totally depend on the code in the
(Draggable) class. Let's say you would have a private static var
counter (private static, so a class property instead of an instance
property) and you would increment that counter using a
setInterval(). The second time you enter the frame and create a new
Draggable object... the counter starts at the last value of the
'old' object. So, you don't get another object with your function
literal but you still end up with a faulty program. And the same
goes for listener objects that are not removed, tweens that are
running and so on.
The destroy() method in a custom class (=object, I can't
stress that enough...) needs to do the cleanup, removing anything
you don't need anymore.
2. if myDraggable != undefined
You shouldn't be using that, period. If you don't need the
asset anymore, delete it using the destroy() method. Again, if you
want to make sure only one instance of a custom object is alive,
use the Singleton design pattern. To elaborate on inheritance:
define the Draggable class (class Draggable extends MovieClip) and
connect it to the myClip_mc using the linkage identifier in the
library). In the Draggable class you can define a function unOnLoad
(an event fired when myClip_mc is removed using
myClip_mc.removeMovieClip()...) and do the cleanup there.
3. A destroy() method performs a cleanup of any assets we
don't need anymore to make sure we don't end up with all kinds of
stuff hanging around in the memory. When you extend the MovieClip
Class you can (additionally) use the onUnLoad event. And with the
code you posted, no it wouldn't delete the myClip_mc unless you
program it to do so.

DVD Replication/Duplication Questions

I didn't know where to post this so I figured since it had to do with video production I would post it under the FCP category. I am at the stage in my production of figuring out how to print my DVDs and I have a couple of questions, I am planning on sending in a master to a replication/duplication company to have 100 copies printed with cases.
Specifications:
- The film will be 30-40 minutes long
- It is shot fully Canon DSLRs
- Footage that is edited is in the format of 1280 x 720 prores 422
- Master wil be made in Adobe Encore
- 100 copies will be made
Questions:
1.) Is there any major difference between duplication and replication of DVD's? the information that I have gathered is that replication is for companies looking to produce thousands and thousands of DVD's where as duplication is for more short run projects that require a couple hundred copies. seeing that I am looking to produce 100 copies I figured this site would be a good way to go http://www.sfvideo.com/ it seems to have a good price and is one of the only sites ive found that lets you place an order under 300 copies.
2.) I have read a bit on printing a master copy of your film and have come to the conclusion that if you are going to replicate mass amounts of DVDs you have to burn to a digital linear tape but if you are doing short run projects you can just burn to a recordable DVD and send it in, is this correct? or do all masters need to be printed to a DLT? or does it just depend on which company you are going through and what they accept? I am going to use Adobe encore to print the DVD.
3.) What type of file should you bring your film into encore in? I am editing with 1280 x720 prores files.
Thanks,
Tyler
Re: Producing an Indie film

First off, you do realize that DVD is an SD medium? If you want hidef, you'll need to make blue rays.
As to your questions: Replicated dvd's will play in any player. Sometimes burned discs will have trouble in some players. Don't know the place you've mentioned.
Whatever you give a replicator, they will make a glass master from which to replicate. A DLT is really only necessary if you plan on paying extra for a copy protection scheme.
I've never used encore so cannot help there.
Anyone else?

EDW Architecture question

Hi experts.
I was searching for orientation in SAP help about EDW architecture details, but I didnt find anything.
Were developing a new BW (2004s version) project and the challenge is to have one physical layer with EDW and another layer just for reporting.
My question is about the communication between these two layers, as long as they are in two different machines. If my reporting layer infocubes are loaded by EDW layer DSO´s how can I create a transformation linking both objects? Is it possible? Or our concept is not right?
Thanks in advance.
TP

Physical layer with EDW: It will be the Modeling area and data will reside in your InfoProviders (DSO/Cube etc).
Rreporting: You will use Front-Ent which is BI Reporting Tools to create, edit, use reports in Bex tools like Analyzer, WAD, Report Designer. And data will be coming from your InfoProviders.
My question is about the communication between these two layers, as long as they are in two different machines.
Do you mean to keep two production server where one will be used just for backup and another one for all users? BTW users will be using reports only through BW Reporting Tools and they will be able to access through network.
If my reporting layer infocubes are loaded by EDW layer DSO´s how can I create a transformation linking both objects? Is it possible? Or our concept is not right?
Check these:
Modeling:
https://www.sdn.sap.com/irj/servlet/prt/portal/prtroot/docs/library/uuid/6ce7b0a4-0b01-0010-52ac-a6e813c35a84
https://www.sdn.sap.com/irj/sdn/docs?rid=/webcontent/uuid/93eaff2d-0a01-0010-d7b4-84ac0438dacc
Hope it helps..

Project architecture question

hello
iam trying to develop a mobile application using j2me to communicate with linux machine through wireless LAN .
my question is that iam confused about the project architecture.
ihave 2 ways:
1-develope desktop application working as aserver (on the linux machine) and the communication with mobile through streams.
OR
2-make the server is tomcat running my developed servlet and the communication through HTTP.
which way is prefered and why? or any other suggestions?
thanks in advanced.

HTTP is perfered because it is more widely supported.

General architectural question....

Say you have a Person object like this:
class Person {
private String name,address,phone;
public String getPhone() {
return phone;
public void setPhone(String phone) {
this.phone=phone;
}Now you have some "client" application which creates Persons and puts them in the cache:
Person joe = new Person();
cache.add("joe", joe);This application is maintaining a reference to joe. Perhaps it's even passing joe around to other classes. You want to make sure that when some class in this client app calls "joe.getPhone()" that it's getting a correct result, one that reflects changes made to the cached value by other JVMs. You could do this a few ways:
- By maintaining a local hashmap of keys to objects and having that local map kept up to date by a CQC. In that case you'd not pass around object references in your app, just object keys. This doesn't work well for applications which were already written assuming a single-JVM architecture since all your code is passing around objects and calling getter/setter methods on those objects.
- You could have the object implement MapListener and listen for updates to itself. Every time a MapEvent showed up it would have to repopulate its member variables with the values from the MapEvent. That's not great either because if your object has many member variables, you're doing a lot of unnecessary serializing/deserializing/repopulating everytime one variable is updated. Also if a CQC maintains a reference to a Person implementing MapListener, that person will never be garbage collected(right?).
- You could change your getter/setter methods to interact with the cached values instead of the local member variable values but that's not good either because in cases of storage disabled nodes you're hitting the network every time you do a getXXX() even if the value has not changed. Also doesn't coherence's usage of reflection depend on the presence of typical getXXX/setXXX methods which look like the one above?
There has got to be a better solution that I'm missing, right? I'd been running apps with storage disabled and CQCs keeping a local hashmap updated. Maybe it makes more sense to enable storage on each of the client app JVMs. I'd wanted to have many small storage disabled apps running on a single machine with just one or two storage enabled JVMs on it.
Thanks,
Andrew

Hi Andrew,
snidely_whiplash wrote:
Say you have a Person object like this:
class Person {
private String name,address,phone;
public String getPhone() {
return phone;
public void setPhone(String phone) {
this.phone=phone;
}Now you have some "client" application which creates Persons and puts them in the cache:
Person joe = new Person();
cache.add("joe", joe);This application is maintaining a reference to joe. Perhaps it's even passing joe around to other classes. You want to make sure that when some class in this client app calls "joe.getPhone()" That is not really safe, particularly it is not thread-safe, and you can't always expect it to work.
that it's getting a correct result, one that reflects changes made to the cached value by other JVMs. And this will never work with Coherence, because any changes arriving from other nodes will always be deserialized into a new object instance created by Coherence, never into existing ones.
You could do this a few ways:
- By maintaining a local hashmap of keys to objects and having that local map kept up to date by a CQC. In that case you'd not pass around object references in your app, just object keys. This doesn't work well for applications which were already written assuming a single-JVM architecture since all your code is passing around objects and calling getter/setter methods on those objects.
Nonetheless, passing on only keys and accessing the cache by key is the way it is intended to be used.
- You could have the object implement MapListener and listen for updates to itself. Every time a MapEvent showed up it would have to repopulate its member variables with the values from the MapEvent. That's not great either because if your object has many member variables, you're doing a lot of unnecessary serializing/deserializing/repopulating everytime one variable is updated. Also if a CQC maintains a reference to a Person implementing MapListener, that person will never be garbage collected(right?). Also Coherence will not know that that object is supposed to be the cached object, therefore whenever you do a CQC.get() with the same key, you would get a different object, and poof, you don't have single object per identity anymore.
>
- You could change your getter/setter methods to interact with the cached values instead of the local member variable values but that's not good either because in cases of storage disabled nodes you're hitting the network every time you do a getXXX() even if the value has not changed. Also doesn't coherence's usage of reflection depend on the presence of typical getXXX/setXXX methods which look like the one above?I don't really see what you refer to here.
>
There has got to be a better solution that I'm missing, right? I'd been running apps with storage disabled and CQCs keeping a local hashmap updated. Maybe it makes more sense to enable storage on each of the client app JVMs. I'd wanted to have many small storage disabled apps running on a single machine with just one or two storage enabled JVMs on it. I don't really see where you are trying to get to. Even in case of storage-enabled JVMs, majority of the data in a partitioned cache (where storage disabled has a meaning) will not be local, similarly to storage-disabled JVMs where all of the data is not local.
Altogether, Coherence was never intended and is not designed to be used as a pass-by-reference Map. You should expect your objects to be passed by value. Coherence only guarantees that data whicy you get from a cache.get or cache.entrySet or similar stuff is up-to-date (instead of stale data successfully overwritten with another entry) at the point in time of getting it, but you should not try to rely on certain cases where your objects are stored in Java object form.
If you try to reproduce such functionality like a single objects for each identity being kept up-to-date, you would have to write your own repository of data objects and keep them up-to-date by a non-lite listener.
Also, you would still have to
- either resolve race conditions related to putting such an object back into the cache (the object being written to the network is vulnerable to modifications by the listener writing incoming changes to the object),
- or copy state to your own object before putting an object to the cache, and then comes the problem of having multiple identities at that point for the object and it is problematic to ensure that the object that is kept is holding the correct state... all in all it exposes the problem of maintaining many synchronized instances of data without distributed locking (or having a very slow and not really reliable solution using distributed locking) and therefore having to reconcile changes from multiple directions (similar problems like the ones tackled by active-active push replication)
Because of the above things, this is probably a bad idea.
Best regards,
Robert

Replication Architecture Question

Similar Messages

Maybe you are looking for