Tuesday, 19 June 2012

EPM 11.1.2.2 – Compact deployment

I thought I would put together a write up on compact deployment in 11.1.2.2 as judging by some of the posts on the OTN forums lately there is a bit of confusion around this area.

I did write up a few posts in the past on compact deployment in 11.1.2.1 which can be read here, here and here, luckily the process in 11.1.2.2 has been much more simplified.

Now I must stress that all the information contained in this blog is just my interpretation and there may be some inaccuracies, as always if you feel there is then please get in touch and share your views.

So what is a compact deployment, well basically all it means is that instead of each web application being deployed to its own managed server they are combined and deployed as one, the number of web applications deployed to one managed server depends on what is selected when you run the EPM configurator which I will go through shortly.

So why have a compact deployment, well I think one of the main reasons the compact deployment came about was the big swing in the amount of resources required when entering the world of 11.1.2.x. In previous versions it was easy for say a consultant to run the full suite of Hyperion products on a laptop or run a POC, demo or training on an average spec machine, when 11.1.2.x came along this all changed as WebLogic had to be deployed and with the number of web applications and the amount of memory consumed this stopped many from being able to run what they could in previous versions.

Probably down to the amount of complaints and pressure then Oracle came up with the compact deployment method.

So what are the main advantages of a compact deployment
  • Reduced overall memory requirement.
  • Reduced start up time.
  • Easier to manage as less services to start
  • Combined web application log (individual logs are available though)
  • Mixed mode of compact deployment web applications and standard deployment
There are disadvantages as well though
  • If the JVM crashes then all web applications are taken down.
  • As all web applications share one JVM then overall performance may not be as good.
  • If there is an issue with one web application then this can stop the managed server from running.
  • Reduced logging information.
The advantages definitely out way the disadvantages if you are using EPM for development, training, POCs etc  Oracle do say that compact deployment is now supported in production, I am not going to get into a debate of whether that is a good idea but personally I have reservations about using it in a production environment.

Oracle has also helped out those who want to quickly deploy Essbase, Planning or HFM to a one server machine which is known as a rapid deployment and uses the compact deployment method.

The rapid deployment documentation contains a step by step guide to deploying on Windows 2008 R2, the documentation even goes into detail on the OS configuration and steps for installing Oracle as the database.

The docs are available at the following locations for
Ok, so let’s go through the steps of a compact deployment but please note this is not a step by step configuration guide and does not contain screenshots of every panel.


After the required products have been installed the EPM configurator is started and all the Web Applications which are going to be deployed to one managed server can be selected, if there are web applications you don’t want to be part of the compact deployment and you want them to run under their own managed server you can deploy them later.

Please note that by default all the available Web Applications will be selected for deployment so if you don’t want to deploy all the web applications to one managed server only select the ones that you do.

If it is the first time the web applications are being deployed I make sure that the configure database is selected so the datasources are configured correctly in WebLogic.

I am going to be deploying the following to one managed server : Foundation (Workspace, Shared services), EPMA, Calculation Manager, Financial Reports, Web Analysis, Provider Services, EAS and Planning.

It is important to understand which web applications you actually want to deploy to one managed server first as even though the memory footprint is reduced from a standard deployment the more web apps you select the higher the totally memory consumed will be.

There may be web applications that you want to deploy but are not going to use often so it is possible you wouldn’t include them in the compact deployment and deploy them to their own managed server and start when required though this could depend on the resources available.


If the configure database option is selected for multiple products then you will only be able to configure to one database, if individual database/schemas are required then it is possible to select the configure database for one product, configure and then move on to the next product to configure the database until all have been configured and then deploy the web applications.


As this is the first deployment the web applications will need to be deployed to a new domain, most of the time the default configuration can be kept and just a password provided.


Now for the most important panel in a compact deployment and I think this is where the biggest change is with 11.1.2.2 deployments, “Deploy the web applications to a single managed server” is selected so a compact deployment is the default.

You will notice that all web applications will be deployed to one managed server called EPMServer0 and all web applications will be running on one port which the default is 9000.



If you deselect “Deploy the web applications to a single manage server” then it is standard deployment where each web application will be deployed to its own managed server and running on a port which you are probably more accustomed to.


It is possible to change the default port of 9000 or SSL port 9443 if required.


I think it is worth going over the web server confiugration as well as I have noticed some confusion over it, Oracle HTTP Server comes as part of a separate download and it is possible to install without including it, now if you take this route the default web server will be the newly introduced Embedded WebLogic one.

The Embedded WebLogic HTTP Server uses a proxy servlet which is bound to Foundation services so whatever port Workspace/Shared Services is running on will define the port http server is running on, in a default compact deployment the port is 9000 so the http server will be running on port 9000.


If you have a compact deployment which includes Workspace/Shared Services using the default port of 9000 and using the embedded http server you would access over port 9000.

If you are using OHS as the http server then the default would be port 19000.

It can get confusing if Foundation was deployed to its own managed server and the embedded http server was chosen as then the default port for the http server would now be 28080.

It is worth noting that the embedded http server is not supported in production environments and I have also noticed rendering issues when using it with IE9 which don’t seem to exist if using Firefox, the Oracle http server is not a big overhead in terms of resources so it might be the best option anyway.


If connecting to say EAS and it is part of a compact deployment then to be able to connect successfully you would need to include the port, this could be either the direct port EAS is running on or via the http server port.


If the deployment to one managed server is on a windows server then a service called “Hyperion EPM Server – Web Application” will have been created.

It can also be started up by a script startEPMServer.bat/sh in <MIDDLEWARE_HOME>\user_projects\<instancename>\bin


If you start up the WebLogic admin server then you will see EPMServer0 has been deployed and is running on port 9000.


You will also be able to view all the web applications that have been deployed as part of the compact deployment.



The managed server can be monitored just like any other using Enterprise Manager which is installed and configured by default in 11.1.2.2


The combined logs are available in the services directory for windows and the starter directory for *nix.


The individual logs are available in <MIDDLEWARE_HOME>\user_projects\domains\EPMSystem\servers\EPMServer0\logs

I have noticed there is not as much information available as when the web applications are deployed to their own managed server.

So how does the memory consumption compare between a standard and a compact deployment.


With the ten web applications I deployed to one managed server this stabilised after start up at around 1.2GB


The equivalent memory when deployed to individual managed servers averaged at around 6GB so you can see a massive difference.

Remember there is the overhead of other services to consider which depends on the products installed e.g. RAF Agent, EPMA, Essbase, RMI

The maximum JVM size for a compact deployment is set to 2701mb which can be increased if required either in the registry under

HKEY_LOCAL_MACHINE\SOFTWARE\Hyperion Solutions\EPMServer0\HyS9EPMServer and JVMOptionXX -Xmx2701m

or by editing

<MIDDWARE_HOME>\user_projects\<instancename>\bin\deploymentScripts\setCustomParamsEPMServer.bat/sh
and updating the set JAVA_OPTIONS line.

So how about a comparison on start-up times between compact and standard.

Please note this is not scientific and is based on the web applications I deployed and obviously depends on the hardware being used.

One managed server averaged at 3 ½ - 4 minutes.

The equivalent individual managed servers if started one by one and waiting until each one had fully started up before starting the next one was about 30 minutes, if all the individual services were started at once which is not always advisable then it took around 7 minutes.


What I did notice was that if you didn’t start the RAF agent first then the above errors appeared in the error log which caused the overall start-up time to increase.

This was not just applicable to the compact deployment as it happens if RAF web is deployed to its own managed server and the RAF agent is not started first, It doesn’t cause any problems if it is started up after it just slows down the start-up time, I am sure this did not occur in previous versions of 11.1.2

Anyway my testing of start-up times was done with the RAF agent started first.

It is possible to scale out the compact deployment to additional machines if required and to do this you first need to install the same set of web applications on the additional machine.


When you start up the EPM configurator an option will become available called “Scale out compact server on this machine


This would then deploy exactly the same web applications as in the compact deployment on the first machine.


If you then look in WebLogic admin console you notice that EPMServer1 has been created and is part of the EPMServer cluster.

You also need to run the Web Server configuration again as this will then add the scaled out deployment to the http server configuration.

####<Jun 17, 2012 11:06:03 AM BST> <Critical> <WebLogicServer> <EPM11CLUSTER> <EPMServer1> <Main Thread> <<WLS Kernel>> <> <> <1339409163293> <BEA-000386> <Server subsystem failed. Reason: java.lang.AssertionError: java.lang.reflect.InvocationTargetException

Make sure the Weblogic admin server is running before starting up the EPM Server on the scaled out server as otherwise it will not start and you will be hit with the above error, the admin server does not need to be running for subsequent start-ups.

If you want to add additional web applications to the compact server


In the configurator select “Deploy to Application Server” for web app you want to add to the compact deployment.


You should see the additional web application(s) available to be deployed to the managed server.

Make sure you also configure the “Foundation” - “Web Server” component so that newly deployed applications is also configured with the http server.

If you have already scaled out the deployment then this is what the documentation says when adding an additional web application

“If you add additional Web applications to the single managed server and you have multiple machines configured with the single managed server, for each additional machine (other than the WebLogic Administration Server machine), do the following:
1.    Install the additional applications.
2.    Restart the single managed server.”


I think this is missing a step because if you only follow those steps then the web applications on the scaled out server will not be added to the Shared Services registry and will not be added to the http server configuration.


So the registry will only contain web application information for the EPMServer0 and the http configuration will not add in the scaled out server even if you run the configure web server component again.


By selecting “Scale out compact server on this machine” again this should configure the additional web application.


Running the registry report displays the web application has been added against EPMServer1 and after running “Configure Web Server” the additional configuration for the scaled out server has been added.

Now on to the final piece and that is removing a web application from a single managed server, there may be a number of reasons why you want to do this such as if you want a web app to run in its own managed server instead of being part of the single managed server.

The documentation states

"Use the WebLogic Administration Console to remove any Web applications from the single managed server, and restart the single managed server on all machines.

If you uninstall any product from a machine that is part of the single managed server on that machine, the entire single managed server on the machine is removed."


It would be so much simpler if the configurator just provided the ability to deselect web applications from the “deploy to application server” panel.

Let me go through the first option “WebLogic Administration Console to remove any Web applications from the single managed server” and try and remove the planning web application.



The planning deployment is deleted from within the admin the console though the problem with this method is that from an EPM configuration perspective planning will still exist.


Restarting the web application and logging into workspace planning will still exist even though it has been deleted.


If you try to configure again the web application deployment still believe Planning is part of the compact deployment.

So what about the other statement in the documentation

“If you uninstall any product from a machine that is part of the single managed server on that machine, the entire single managed server on the machine is removed”


The uninstaller was run from <MIDDLEWARE_HOME>\EPMSystem11R1\uninstall\uninstall.cmd/sh and the planning web application removed.


Maybe I am misreading the documentation but the single managed server is not removed.

Anyway the process I finally went through to remove a web application from a compact deployment was
  • Uninstall the product on each server the single managed server has been deployed to.
  • Go to into the WebLogic admin console and delete all the deployments related to the web application.
  • “Configure Web Server” again in the EPM configurator to update the http server configuration.
If you wanted to then deploy the removed product to its own managed server you could reinstall the product, select “Deploy to Application Server” in the configurator just for that web application and in the deploy to application server panel deselect “Deploy the web applications to a single managed server”.

10 comments:

Unknown said...

that's a very interesting post. Do you think it would be possibile to deploy the entire suite for testing purposes in a "4gb ram quad core ssd" laptop?

John Goodwin said...

Depending on the products deployed I would not personally recommend anything less than 8GB.

Giacomo Lacava said...

I believe the docs are confused about removing apps. I have been told that the procedure should be:
- uninstall the product
- *redeploy* the whole compact.

This would force the configurator to take notice of the product disappearance.

I haven't tried it myself, tbh. As it is, I personally wouldn't recommend a compact in any production environment and not even in test, it's a very bad idea in most respects and little more than a "demo-drive mode".

Peg Rusnak Hill said...

Thank you for your post. It was very helpful.

Question. We see the benefits of a compact deployment in a production environment. Is there value in a dev/ test environment? Are there pros / cons with deploying Dev and test with the individual deployments to keep the logging files intact, vs. deploying compact in all 3 environments?

Thanks!

Alex said...

Hi John,

Thanks for sharing the information.

How feasible is it take a compact deployment and reconfigure to standard deployment without uninstalling the EPM product suite?
Is this even a feasible option?

Regards,
Alex

John Goodwin said...

Hi Alex,

You can certainly deploy products to their own managed server after being deployed to a single managed server, the problem is that the configurator does not remove it from the single managed server.
It is possible to remove it from the Weblogic admin console though the best option might be uninstall/install the product again.

Anonymous said...

John, Thanks for the pasting. I succesfully installed EPM 11122 (Essbase, Planning, Reporting and Analysis).
I'm able to connect to Administration Server using machine name: 9000.

But having issues connecting Essbase Server. When I confgured, I left the default settings "Essbase Cluster -1" and Agent Port 1523. (no SSL selected).

What part number should I be using for the Essbase Server to be connected in EAS?

John Goodwin said...

The default port for Essbase is 1423, you shouldnt need to add the Essbase port in EAS if it is running on the default port.
If you are having problems connecting to Essbase then check it is actually running and check the logs

Tej Ravi said...

Hello John , you said the JVM can be increased from 2701. What is the maximum it can be increased to? Thanks

John Goodwin said...

64bit then in theory the maximum is only limited by the memory available on the machine.