Tuesday, 11 August 2020

The magic of CorrelationSets

CorrelationSets in BPEL are as old as the road to Rome. I wrote about it before: 

Although it was in the BPEL project from the very beginning, when Oracle acquired it in 2004, you might not have dealt with it before. But maybe not even realized that you can use it in Oracle Integration Cloud, with structured processes.

In the first week of june I got to do a presentation about this subject, in a series of Virtual Meetups.

If you weren't able to attend, but would like to watch it then you're in luck, it got recorded by Phil Wilkins:

In my presentation I start with a simple demo based on a BPEL process. I have put the resulting code on GitHub: https://github.com/makker-nl/blog/tree/master/CorrelationDemo

Then I move on to a more complicated situation in OIC. I created an export for that project and placed it on GitHub too: https://github.com/makker-nl/blog/tree/master/CorrelationDemoOIC

This allows you to inspect it and try to recreate it yourself.

My sincere appologies for this late sharing.

Wednesday, 15 July 2020

Receive and send WSA Properties in BPEL 2.0

Last week I had the honour to present on CorrelationSets in a Virtual Meetup, which is a feature that relates to the WS-Addressing support of SOA Suite.

At my current customer, I had to rebuild a BPEL Process from 1.1 to 2.0, to be able to split it up using embedded and reusable  subprocesses.

One requitement is to receive the wsa-Action property and reply it back, concatenated with 'Response'.

Since it implements a WSDL with 3 operations, I need a Pick-OnMessage construction.

To receive properties you can open the activity, in my case the OnMessage:
In the source his looks like the following:
<onMessage partnerLink="MyService_WS" portType="ns1:myService" operation="myOperation"
          <bpelx:fromProperty name="wsa.action" variable="wsaAction"/>
With the wsaAction variable here is based on xsd:string.

However, this turns out not to work: the wsaAction variable stays empty.
This turns out to be a bug, that should have been solved since, but apparently still works as is. Read more about it in support document 1345071.1.

Solution is simple: just remove the wsa. prefix:
<onMessage partnerLink="MyService_WS" portType="ns1:myService" operation="myOperation"
          <bpelx:fromProperty name="action" variable="wsaAction"/>
For invoke, reply, receive and other activities it works the same.

As said, in my case I need to reply with a wsa.action that is a concatenation of the received action with 'Response'. This can be done using an expression:
Again, first choose wsa.action and then in the source remove the wsa. prefix:
<reply name="ReplyMyService" partnerLink="MyService_WS" portType="ns1:myService"
                 variable="MyService_OutputVariable" operation="myOperation">
              <bpelx:toProperty name="action">concat($wsaAction, 'Response')</bpelx:toProperty>
            <bpelx:property name="action" variable="WSAction"/>
Testing this in SoapUI or ReadyAPI will show:
<env:Envelope xmlns:env="http://schemas.xmlsoap.org/soap/envelope/" xmlns:wsa="http://www.w3.org/2005/08/addressing">

For more info on setting properties, see the docs.

Tuesday, 30 June 2020

A little bit of insight in SOA Suite future

A few weeks ago I was made aware of a few announcements, which I think makes sense and that I want to pass on to my followers, sauced with a bit of my own perspective.

Containerized SOA

Last year I had made myself familar with the Oracle Weblogic Kubernetes Operator. See for instance my Cheat Sheet Serie. I also had the honour to talk about it during the Tech Summit at OUK in december '19. Weblogic under Kubernetes is apparently the way to go for Weblogic. And with that, also the Fusion Middleware Stack. However, until now only 'plain' Weblogic is supported under Kubernetes, on all Cloud platforms, as well as on your own on-premises Kubernetes platform.

It was no surprise that SOA Suite would follow, and in March there an early acces for SOA Suite on Kubernetes was announced.

In the announcement it is stated that Oracle will provide Container images for SOA Suite including OSB, that are also certified for deployment on production Kubernetes environments. Also documentation, support files, deployment scripts and samples.

Later on other components will be certified. This is good news, because it will allow SOA Suite be run in co-existence with cloud native applications and be part of a more heterogenous application platform. To me this makes sense. It makes High Availability and Disaster Recovery easier, but although the application landscape will be diverse and heterogenous, this makes the maintenance, install, deploy and upgrade of FMW within that landscape more uniformly aligned with other application compents like web applications, possible microservices, etc.

Paid Market Place offering

Another announcment I got recently is about the release of  a "Paid" listing of Oracle SOA Suite for Oracle Cloud Infrastructure" on the Oracle Marketplace. There was already a Bring Your Own Licence offering, that you could bring in to use your universal cloud credits to host your SOA Suite instance in the cloud. You could purchase a separate license, but now you can also use the Universal Cloud Credits  to have a paid, licensed instance of SOA Suite in the cloud, without the need to purchase a license.

And so there are two new offerings in the market place:
  • Oracle SOA Suite on Oracle Cloud Infrastructure (PAID)
  • Oracle SOA Suite with B2B EDI Adapter on Oracle Cloud Infrastructure (PAID)
These offerings include:
  • SOA with Service Bus & B2B Cluster, with additional leverage of the B2B EDI Adapter.
  • MFT Cluster
  • BAM
This will provide better options for deploying SOA Suite on OCI, to:
  • Provision SOA instances using OCI
  • Manage instances using OCI
  • Scale up/down/in/out using OCI
  • Backup/restore using OCI.
Oracle's focus on delivering SOA Suite from the Market place. It is expected that current SOA Cloud Service customers will migrate to this offering. The Marktet Place SOA Suite will be enhanced and improved with new capabilities and functions, that not necessarily will be added to the SOA CS.
Probably this will give Oracle a better and more uniform way to improve and deliver new versions of SOA Suite. It also makes sense in relationship to the SOA Suite on Containers announcement.

For new customers the Marketplace is the way to get SOA Suite. Existing customers can use the BYOL offering, but might need to move to the new offering when contract renewal might be opportune.

What about Oracle Integration Cloud (OIC)?

This is still Oracle's prime offering for integrations and process modelling. You should first look at OIC for new projects. Only if you're an existing SOA Suite customer and/or have specific requirements that drive the choice to SOA Suite and related components, then you should consider the Marketplace SOA Suite offering.

This makes the choices a bit clearer, I think.

Friday, 19 June 2020

Use of correlation sets in SOA Suite

Years ago, I had plans to write a book about BPEL or at least a series of articles to be bundled as a BPEL Course. I stranded with only one Hello World article.

This year, I came up with the idea of doing something around Correlation Sets. Preparing a series of articles and a talk. And therefor, let's start with an article on Correlation Sets in BPEL. Maybe later on I could pick up those earlier plans again.

You may have read "BPEL", and tend to skip this article. But wait, if you use BPM Suite: the Oracle BPM Process Engine is the exact same thing as the BPEL Process engine! And if you use the Processes module of Oracle Integration Cloud: it can use Correlation Sets too. Surprise: again it uses the exact same Process Engine as Oracle SOA Suite BPEL and Oracle BPM Suite.

Why Correlation Sets?

Now, why Correlation Sets and what are those? You may be familiar with OSB or maybe Mulesoft, or other integration tools.
OSB is a stateless engine. What comes in is executed at once until it is done. So, services in OSB are inherently synchronous and short-lived. You may argue that you can do Asynch Services in OSB. But those are in fact "synchronous" one-way services. Fire & Forget, as you will. They are executed right away (hence the quoted synchronous) , until it is done. But the calling application does not expect a result (and thus asynchronous in the sense that the caller won't wait).

You could, and I have done it actually, create asynchronous request response services in OSB. Asynchronous Request Response services are actually two complementary one way fire & forget services. For such a WSDL both services are defined in different port types: one for the actual service consumer, and one callback service for the service provider. Using WS-Addressing header elements the calling service will provide a ReplyTo callback-endpoint and a MessageId to be provided by the responding service as an RelatesTo MessageId.

This RelatesTo MessageId serves as a correlation id, that maps to the initiating MessageId. WS- Addressing is a Webservice standard that describes the SOAP Header elements to use. As said, you can do this in OSB, OSB even has the WS-Addressing namespaces already defined. However, you have to code the determination and the setting of the MessageId and ReplyTo-Address yourself.

Because of the inherently stateless foundation of OSB the services are short-lived and thats why OSB is not suitable for long running processes. The Oracle SOASuite BPEL engine on the other hand, is designed to orchestrate Services (WebServices originally, but from 12c onwards REST Services as well) in a statefull way. This makes BPEL suitable for long running transactions as well. Because of that after the acquisition of Collaxa, the company who created the BPEL Engine, Oracle decided to replace it's own database product Oracle Workflow (OWF) with BPEL.  And SOA Suite and it's BPEL engine natively support WS-Addressing. Based upon an Async Request/Response WSDL it will make sure it adds the proper WS-Addressing elements and has a SOAP Endpoint to catch response messages. Based upon the RelatesTo message id  in the response it will correlate the incoming response with the proper BPEL process instance that waits for that message.

A BPEL process may run from a few seconds, to several minutes, days, months, or potentially even years. Although experience learned us that we wouldn't recommend BPEL services to run for longer than a few days. For real long running process you should choose BPM Suite or Oracle Integration Cloud/Process.

WS-Addressing helps in correlating response messages to requests that are sent out previously. But, it does not correlate Ad-Hoc messages. When a process runs for more than a few minuts, chances are that the information stored within the process is changed externally. A customer waiting for a some process may have relocated or even died. So you may need to interact with a running process. You want to be able to send a message with the changed info to the running process instance. And you want to be sure that the engine correlates the message to the correct instance. Correlation Sets help with these ad-hoc messages that may or may not be send at any time during the running of the process.

An example BPEL process

Let's make a simple customer processing process that reads an xml file and processes it and writes it back to an xml file.
My composite looks like:
It has two File Adapter definitions, an exposed service that polls on the /tmp/In folder for customer*.xml files. And a reference service that writes an xml file into the /tmp/Out folder as customer%SEQ%_%yyMMddHHmmss%.xml. I'm not going to explain how to setup the File adapters, that would be another course-chapter.

For both adapters I created the following XSD:
<?xml version="1.0" encoding="UTF-8" ?>
<xsd:schema xmlns:xsd="http://www.w3.org/2001/XMLSchema" xmlns:cmr="http://xmlns.darwin-it.nl/xsd/demo/Customer"
            targetNamespace="http://xmlns.darwin-it.nl/xsd/demo/Customer" elementFormDefault="qualified">
  <xsd:element name="customer" type="cmr:CustomerType">
      <xsd:documentation>A customer</xsd:documentation>
  <xsd:complexType name="CustomerType">
      <xsd:element name="id" maxOccurs="1" type="xsd:string"/>
      <xsd:element name="firstName" maxOccurs="1" type="xsd:string"/>
      <xsd:element name="lastName" maxOccurs="1" type="xsd:string"/>
      <xsd:element name="lastNamePrefixes" maxOccurs="1" type="xsd:string" minOccurs="0"/>
      <xsd:element name="gender" maxOccurs="1" type="xsd:string"/>
      <xsd:element name="streetName" maxOccurs="1" type="xsd:string"/>
      <xsd:element name="houseNumber" maxOccurs="1" type="xsd:string"/>
      <xsd:element name="country" maxOccurs="1" type="xsd:string"/>
(Just finishing this article, I encounter that I missed a city element. It does not matter for the story, but in the rest of the example I use the country field for city.)

The first iteration of the BPEL process just receives the file from the customerIn adapter, assigns it to the the input variable of the invoke of the customerOut adapter and invokes it:

Deploy it to the SOA Server and test it:

[oracle@darlin-ind In]$ ls ../TestFiles/
customer1.xml  customer2.xml
[oracle@darlin-ind In]$ cp ../TestFiles/customer1.xml .
[oracle@darlin-ind In]$ ls
[oracle@darlin-ind In]$ ls
[oracle@darlin-ind In]$ ls
[oracle@darlin-ind In]$ ls
[oracle@darlin-ind In]$ ls ../Out/
[oracle@darlin-ind In]$
The output customer hasn't changed and is just like the input:
[oracle@darlin-ind In]$ cat ../Out/customer2_200617125051.xml
<?xml version="1.0" encoding="UTF-8" ?><customer xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://xmlns.darwin-it.nl/xsd/demo/Customer ../Schemas/Customer.xsd" xmlns="http://xmlns.darwin-it.nl/xsd/demo/Customer">
  <streetName>Rue d'Oxygene</streetName>
[oracle@darlin-ind In]$

This process is now rather short-lived and doesn't do much except for moving the contents of the file. Now, let's say that this processing of the file takes quite some time and but during the processing the customer can have relocated, or died or has otherwise changed it's information.

I expanded my composite with a SOAP Service, based on a One-Way WSDL, that is based upon the same xsd:
And this is how I changed the BPEL:

In this example, after setting the customer to the customerOut variable, there is a long running  "customer processing" sequence, that takes "about" 5 minutes.

But in parallel it now also listens to the UpdateCustomer partnerlink using a Receive. This could be done in a loop to also receive more follow-up messages.

This might look like a bit unnecessarily complex, with the throw and catch combination. But the thing with the Flow activity is that it completes only when all the branches are completed. So, you need a means to "kill" the Receive_UpdateCustomer activity. Adding  a Throw activity does this nicely. Although the activity is colored red, this is not an actual Fault exception. I use it here as a flow-control activity. It just has a simple fault name, that I found the easiest to enter in the source:
<throw name="ThrowFinished" faultName="client:Finished"/>

This is because you can just use the client namespace reference. While in the designer you should provide a complete namespace URI:

Same counts for the Catch, after creating one, it's easier to add the namespace from the source:
        <catch faultName="client:Finished">
          <assign name="AssignInputCustomer">
              <to expressionLanguage="urn:oasis:names:tc:wsbpel:2.0:sublang:xpath1.0">$Invoke_WriteCustomerOut_Write_InputVariable.body</to>

Side-note: did you know that if you click on an activity or scope/sequence in the Designer and switch to the source, the cursor will be moved to the definition of the activity you selected? To me this often comes handy with larger BPELs.

By throwing the Finished exception the flow activity is left and by that all the unfinished branches are also closed and by that the Receive is quit too.

When you get a SOAP message in the bpel example above you would still wait for finishing the process branch. You probably also need to notify the customer processing branch that the data is changed. That can be done in the same way, by doing a throw of a custom exception.

How to define Correlation Sets

The example above won't work as is. Because, how does BPEL know to which process instance the message has to be sent? We need to create a Correlation set. And to do so we need to define how we can correlate the UpdateCustomer message to the customerIn message. Luckily there is a Customer.id field. For this example that will do. But keep in mind: you can have multiple processes running for a customer. So you should add something that will identify the instance.

You can add and edit correlation sets on the invoke, receive, and pick/onMessage activities. But also from the BPEL menu:

Then you can define a Correlation Set:

As you can see you can create multiple Correlation Sets, each with one or more properties. In the last window, create a property, then select it for the Correlation Set and click ok. Up to the first dialog.

You'll see that the Correlation Set isn't valid yet. What misses, what I didn't provide in the last dialog, are the property aliases. We need to map the properties to the messages.
I find it convenient to do that on the activities, since we also need to couple the correlation Sets to particular Invoke, Receive, and/or Pick/OnMessage activities. Let's begin with the first receive:

Select the Correlations tab, and add the Correlation Set. Since this is the activity that the Customer Id first appears in a message in the BPEL processe, we need to initiate the Correlation Set. This can also be done on an invoke, when calling a process that may cause multiple ad-hoc follow-up messages. So, set the Initiate property to yes.
Also, here you can have multiple correlation sets on an activity.

Then click the edit (pencil) button to edit the Correlation Set. And add a property alias:

To find the proper message type, I find it convenient to go through the partnerlink node, then select the proper WSDL. From that WSDL choose the proper message type. Now, you would think you could select the particular element. Unfortunately, it is slightly less user-friendly. After choosing the proper message type in the particular WSDL, click in the query field and type CTRL-Space. A balloon will pop up with the possible fields and when the field has a child element, then a follow-up balloon will pop-up. Doing so, finish your xpath, and click as many times on Ok to get all the dialogs closed properly.

Another side-node, the CTRL-Space way of working with balloons also works with the regular expression builder in with creating Assign-Copy-rules. Sometimes I get the balloons un-asked for, which I actually find annoying sometimes.

Do the same for the customer update Receive:
Here it is important to select No for the initate: we now adhere to the initiated Correlation Set.

Wrap this up, deploy the composite and test.

Test Correlations

As in the first version copy an xml file to the /tmp/In folder. This results in a following BPEL Flow:

The yellow-highlighted activities are now active. So, apparently it waits for a Receive and for the processing (Wait activity).

From the flow trace you can click on the compositename, besides the instance id, and then click on the Test button:
And enter new values for your customer:

In the bottom right corner you can click on the "Test Web Service" button, and from the resulting Response tab you can click on the launch flow trace.

You'll find that the Receive has been done, and the Assign after that as well. Now, only the Wait activity is active.

After processing the flow it throws a Finished exception, and finishes the BPEL Flow.
In this case the Receive was earlier than finisihing the Wait activity. So, in this flow the throw is unnecessary, but when the message wasn't received, then the throw is needed.

Looking in the /tmp/Out folder we see that the file is updates neatly from the Update:
[oracle@darlin-ind In]$ ls ../Out/
customer2_200617125051.xml  customer3_200619160921.xml
[oracle@darlin-ind In]$ cat ../Out/customer3_200619160921.xml
<?xml version="1.0" encoding="UTF-8" ?><ns1:customer xmlns:ns1="http://xmlns.darwin-it.nl/xsd/demo/Customer">
        </ns1:customer>[oracle@darlin-ind In]$

A bit of techie-candy

Where is all this beautifull stuff registered?
First of all, for the Correlation properties, you will find a new WSDL has appeared:

At the top of the source of the BPEL you'll find the following snippet:
      <bpelx:property name="propertiesFile">
  <import namespace="http://xmlns.oracle.com/pcbpel/adapter/file/CorrelationDemo/CorrelationDemo/customerIn"
          location="../WSDLs/customerIn.wsdl" importType="http://schemas.xmlsoap.org/wsdl/" ui:processWSDL="true"/>

Here you see a reference to the properties wsdl. Also an import of the customerIn.wsdl. Let's take a look in there:
<?xml version= '1.0' encoding= 'UTF-8' ?>
    <plt:partnerLinkType name="Read_plt">
        <plt:role name="Read_role">
            <plt:portType name="tns:Read_ptt"/>
    <vprop:propertyAlias propertyName="cor:customerId" xmlns:tns="http://xmlns.oracle.com/pcbpel/adapter/file/CorrelationDemo/CorrelationDemo/customerIn"
         messageType="tns:Read_msg" part="body">
    <vprop:propertyAlias propertyName="cor:customerId" xmlns:ns13="http://oracle.com/sca/soapservice/CorrelationDemo/CorrelationDemo/Customer"
         messageType="ns13:requestMessage" part="part1">
    <wsdl:import namespace="http://oracle.com/sca/soapservice/CorrelationDemo/CorrelationDemo/Customer"
    <wsdl:import namespace="http://xmlns.oracle.com/CorrelationDemo/CorrelationDemo/ProcessCustomer/correlationset"

Below the partnerLinkType you find the propertyAliases.
Especially with older, migrated processes, this might be a bit tricky, because you might get the property aliases not in the wsdl you want. Then you need to register the proper wsdl in the BPEL and move the property aliases to the other wsdl, together with the vprop namespace declaration.
When you move the WSDL to the MDS for reuse, move the property aliases to another wrapper WSDL. You shouldn't move the property aliases to the MDS with it. Because they belong to the process and shouldn't be shared, but also it makes it impossible for the designer to change. I'm not sure if it even would work. Probably it does, but you should not have that.

As I mentioned before, you can have multiple Correlation Sets in your BPEL (or BPMN) process and even on an activity. In complex interactions this may make perfectly sense. For instance when there is overlap. You  may have initiated one Correlation Set on an earlier Invoke or Receive, and use that to correlate to another message in a Receive. But that message may have another identifying field that can be used to correlate with other interactions. And so you may have a non-initiating Correlation Set on an activity that initiates another one. Maybe even based on different property-aliases on the same message.


Per CorrelationSet you can have multiple properties. They are concatenated on to a string. Don't use too many properties to make-up the correlation set. Perferably only one. And use short scalar elements for the properties. In the past the maximum length was around 1000 characters. I've no idea what it is now. But multiple properties and property aliases may make it error-prone. During the concatenation, a different formatting may occur. It is harder to check, validate if the correlation elements in the messages conform with eachother.

In the example above I used the customer id for the correlation property. This results in an initiated correlation set where the UpdateCustomer Receive is listening for. If you would initiate another process instance for the same customer, the process engine will find at the UpdateCustomer Receive that there already is a (same) Receive with the same Correlation Set. And will fail. The process engine identifies the particular activity in the process definition and the combination process-activity and Correlation Set is unique. A uniqueness violation at this point will result in a runtime fault.

It doesn't matter if the message is initiate before of after the Receive is activated. If you would be so fast to issue an UpdateCustomer request before the process instance has activated the Receive, then it will be stored in a table, and picked up when the Receive activity is reached.


This may be new to you and sound very sophisticated. Or, not of course, if you were already familiar with it. If this is new to you: it was already in the product when Oracle acquired it in 2004!
And not only that, you can use it in OIC Processes as well. Also for years. I wrote about that in 2016.

More on correlation sets, check out the docs.

Tuesday, 19 May 2020

Honey, I shrunk the database!

For my current assignment I need to get 3 SOA/B2B environments running. I'm going to try-out the multi-hop functionality, where the middle B2B environment should work as a dispatcher. The idea is that in Dev, Test and Pre Prod environments the dev environment can send messages to the remote trading partner's test environment, through the in-between-environment. To the remote trading partner, that in-between-environment should act as the local test environment, but then should be able to dispatch the message to the actual dev, test or pre-prod environment.

I envision  a solution, where this in-between B2B environment should act as this dispatching B2B hop. So I need to have 3 VMs running with their own database (although I could have them share one database), and Fusion Middleware domain.

The Vagrant project that I wrote about earlier this week, creates a database and then provisions all the FMW installations and a domain. That database is a 12cR1 database (that I could upgrade) that is installed default. In my setup it takes about 1.8GB of memory. My laptop is 16GB, so to have 2 VMs running on it, and let Windows have some memory too, I want to have a VM of at most 6,5 GB.
I need to run an AdminServer and a SOAServer, that I gave 1GB and GB respectively. And since they're not Docker containers, they both run an Oracle Linux 7 OS too.

So, one of the main steps is to downsize the database to "very small".

My starting point is an article I wrote years ago about shrinking an 11g database to XE proportions.
As described in that article I created an pfile as follows:
create pfile from spfile;

This creates an initorcl.ora in the $ORACLE_HOME/dbs folder.

I copied that file to initorcl.ora.small and editted it:
orcl.__oracle_base='/app/oracle'#ORACLE_BASE set from environment
*.dispatchers='(PROTOCOL=TCP) (SERVICE=orclXDB)'

The lines that I changed are copied with the original values commented out. So I downsized the db_cache_size, java_pool, large_pool and pga_aggregate_target. Also the sga_target, shared_io_pool(have it auto-managed) and shared_pool. I needed to set the sga_target to at least 350M, to get it started.
SOASuite needs at least  300 processes and open_cursors.

Now the script, checks if the database running. It is actually a copy of the startDB.sh script also in my Vagrant project.

If it is running, it shutdowns the database. It then creates a pfile for backup. If the database isn't running, it only creates the pfile.

Then it I copied that file to initorcl.ora.small and creates a spfile from it. And then it starts the database again.

SCRIPTPATH=$(dirname $0)
. $SCRIPTPATH/../../install_env.sh
. $SCRIPTPATH/db12c_env.sh
db_num=`ps -ef|grep pmon |grep -v grep |awk 'END{print NR}'`

if [ $db_num -gt 0 ]
  echo "Database Already RUNNING."
  $ORACLE_HOME/bin/sqlplus "/ as sysdba" <<EOF
shutdown immediate;
prompt create new initorcl.ora.
create pfile from spfile;
  # With use of a plugable database the following line needs to be added after the startup command
  # startup pluggable database pdborcl; 
  sleep 10
  echo "Database Services Successfully Stopped. "
  echo "Database Not yet RUNNING."
  $ORACLE_HOME/bin/sqlplus "/ as sysdba" <<EOF
prompt create new initorcl.ora.
create pfile from spfile;
  sleep 10
echo Copy initorcl.ora.small to $ORACLE_HOME/dbs/initorcl.ora, with backup to $ORACLE_HOME/dbs/initorcl.ora.org
mv $ORACLE_HOME/dbs/initorcl.ora $ORACLE_HOME/dbs/initorcl.ora.org
cp $SCRIPTPATH/initorcl.ora.small $ORACLE_HOME/dbs/initorcl.ora
echo "Starting Oracle Database again..."
$ORACLE_HOME/bin/sqlplus "/ as sysdba" <<EOF
create spfile from pfile;

The scripts can be found here.

Oh, by the way: I must state here that I'm not a DBA. I'm not sure if those settings make sense all together. (Should have someone review it). So you should not rely on them for a serious environment. Not even a development one. My motto is that a development environment is a developer's production environment. For me this is to be able to try something out. And to show the mechanism to you.

Friday, 15 May 2020

New FMW 12c Vagrant project


Several years ago I blogged about automatic creation of Fusion Middleware environments.
See for instance this article on installation, this one on the domain creation and these notes.

In between I wrote several articles on issues I got, start/stop scripts, etc.

Later I found out about Vagrant and since then I worked with that. And this I enhanced through the years, for instance, nowadays I use different provisioners to setup my environment.

Until this week I struggeled with a Oracle Linux 7 Update 7 box, as I wrote earlier this week.

For my current customer I needed to create a few B2B environments. So I got back to my vagrant projects and scripts and build a Vagrant project that can create a SOA/BPM/OSB+B2B environment.

You can find it on GitHub in my ol77_soa12c project, with the scripts in this folder.

You'll need to get a Oracle Linux 7U7 Vagrant base box yourself. I tried to create one based on the simple base box of Oracle, as I wrote earlier this year. But in the end I created a simple base install of OL7U7, with one disk, and a Server with GUI package, a vagrant user (with password vagrant). As you can read in earlier articles.

Also  you'll need to download the installer zips from edelivery.oracle.com.


What I did with my scripts in this revision, is that I split up the main method of the domain creation script:
def main():
    # Section 1: Base Domain + Admin Server
    # Section 2: Extend FMW Domain with templates
    # Section 3: Create Domain Datasources
    # Section 4: Create UnixMachines, Clusters and Managed Servers
    # Section 5: Add Servers to ServerGroups.
    print('Updating the domain.')
    print('Closing the domain.')
    # Section 6: Create boot properties files.
    # Checks
    print('\n7. Checks')
  except NameError, e:
    print 'Apparently properties not set.'
    print "Please check the property: ", sys.exc_info()[0], sys.exc_info()[1]
    apply(traceback.print_exception, sys.exc_info())

All the secions I moved to several sub-functions. I added an extra section for Checks and validations. One check I added is to list the server groups of the domain servers. But I may envision other validations later.

Policy Manager

Another thing is that in the method addFMWServersToGroups() I changed the script so that it complies to the topology suggestions from the Oracle Product management. Important aspect here is that for SOA, OSB and BAM it is important to determine if you want a domain with only one of these products, or that you create a combined domain. By default these products will have the Oracle Webservice Managment Policy Manager targetted in to the particular cluster or server. However, you should have only one PolicyManager per domain. So, if you want a combined domain with both SOA and OSB, then you would need to create a separate WSM_PM cluster. This is done using the wsmEnabledcproperty in the fmw.properties file. Based on this same property the server groups are added:
# Add a FMW server to the appropriate group depending on if a separate WSM PM Cluster is added.
def addFMWServerToGroups(server, svrGrpOnly, srvGrpComb):
  if wsmEnabled == 'true':
    print 'WSM Disabled: add server group(s) '+",".join(svrGrpOnly)+' to '+server
    setServerGroups(server, svrGrpOnly)
    print 'WSM Enabled: add server group(s) '+",".join(srvGrpComb)+' to '+server
    setServerGroups(server, srvGrpComb)  
# 5. Set Server Groups to the Domain Servers
def addFMWServersToGroups():
  print('\n5. Add Servers to ServerGroups')
  #print 'Add server groups '+adminSvrGrpDesc+ ' to '+adminServerName
  #setServerGroups(adminServerName, adminSvrGrp)     
  if osbEnabled == 'true':
    addFMWServerToGroups(osbSvr1, svrGrpOnly, srvGrpComb)
    if osbSvr2Enabled == 'true': 
      addFMWServerToGroups(osbSvr2, osbSvrOnlyGrp, osbSvrCombGrp)
  if soaEnabled == 'true':
    addFMWServerToGroups(soaSvr1, soaSvrOnlyGrp, soaSvrCombGrp)
    if soaSvr2Enabled == 'true': 
      addFMWServerToGroups(soaSvr2, soaSvrOnlyGrp, soaSvrCombGrp)
  if bamEnabled == 'true':
    addFMWServerToGroups(bamSvr1, bamSvrOnlyGrp, bamSvrCombGrp)
    if bamSvr2Enabled == 'true':  
      addFMWServerToGroups(bamSvr2, bamSvrOnlyGrp, bamSvrCombGrp)
  if wsmEnabled == 'true':
    print 'Add server group(s) '+",".join(wsmSvrGrp)+' to '+wsmSvr1+' and possibly '+wsmSvr2
    setServerGroups(wsmSvr1, wsmSvrGrp)
    if wsmSvr2Enabled == 'true': 
      setServerGroups(wsmSvr2, wsmSvrGrp)
  if wcpEnabled == 'true':
    print 'Add server group(s) '+",".join(wcpSvrGrp)+' to '+wcpSvr1+' and possibly '+wcpSvr2
    setServerGroups(wcpSvr1, wcpSvrGrp)
    if wcpSvr2Enabled == 'true': 
      setServerGroups(wcpSvr2, wcpSvrGrp)
  print('Finished ServerGroups.')

The groups are declared at the top:
# ServerGroup definitions
# See also: https://blogs.oracle.com/soa/soa-suite-12c%3a-topology-suggestions

For SOA, OSB and BAM you see that there is a default or "combined" server group, and a "server only" group. If wsmEnabeld is false, then the combined group is used and then the Policy Manager is added to the managed server or cluster. If it is true then the "only"-group is used.

Other Remarks

An earlier project I did failed when creating the domain. Somehow I had to run it twice to get the domain created. Somehow this is solved.

In my scripts I still use the zips. But the scripts are quite easiliy adaptable for I'll do that in the near future hopefully. But my current customer still uses this version. So, I went from here.

The project also adapts the nodemanager properties, creates a nodemanager linux service, and copies start stop scripts. However, I missed the bit of setting the listener port and type (plain or SSL) of the nodemanager in the Machine definition. So starting the domain needs a little bit of tweaking.

And for my project I need at 3 environments. So I need to downsize the database and the managed servers so that I can run it in 6GB, and can have 2 VMs on my 16GB laptop.

And I need to add a bridged network adapter to the Vagrant project, so that I can have the environments connect to each other.

Wednesday, 13 May 2020

Vagrant Oracle Linux and the Vagrant user: use a password

Last year and earlier this year I have been struggling to create a new Vagrant box based on an installation of the Oracle Base box. I had some extra requirements, for instance having a GUI in my server to get to a proper desktop when it comes handy. I found in the end that it might be more convenient to create a base box by myself. I also tried using a ssh-key to have the vagrant user connect to the box to do the provisioning. But what I did, I get "Cannot allocate memory"-errors in any stage of the provisioning. For instance, when upgrading the guest additions:

Using a ssh-key is actually the recommended approach. Read my previous blog article on the matter for instructions on how to do it.

It struck me on why I couldn't have a Oracle Linux 7U7 box working as a base for new VMs. And why would I get these nasty memory allocation errors.
I upgraded from Vagrant 2.2.6  to 2.2.9, and VirtualBox from 6.1.4 to 6.1.6, but this wasn't quite related to these versions.

And just now I realized that the one thing I do differently with this box in stead of my OL7U5 box is the vagrant user ssh-key in stead of the password. So, I made sure that the vagrant user can logon using an ssh password. For instance by reviewing the file /etc/ssh/sshd_config and specifically the option PasswordAuthentication:
# To disable tunneled clear text passwords, change to no here!
PasswordAuthentication yes
#PermitEmptyPasswords no
#PasswordAuthentication no
Make sure it's set to yes.

Then re packed the box

d:\Projects\vagrant>vagrant package --base OL7U7 --output  d:\Projects\vagrant\boxes\ol77GUIv1.1.box
==> OL7U7: Exporting VM...
==> OL7U7: Compressing package to: d:/Projects/vagrant/boxes/ol77GUIv1.1.box

And removed the old box:
d:\Projects\vagrant\ol77_gui>vagrant box list
ol75        (virtualbox, 0)
ol77        (virtualbox, 0)
ol77GUIv1.0 (virtualbox, 0)
ol77GUIv1.1 (virtualbox, 0)

d:\Projects\vagrant\ol77_gui>vagrant box remove ol77GUIv1.0
Removing box 'ol77GUIv1.0' (v0) with provider 'virtualbox'...

d:\Projects\vagrant\ol77_gui>vagrant box list
ol75        (virtualbox, 0)
ol77        (virtualbox, 0)
ol77GUIv1.1 (virtualbox, 0)

I found that it might be usefull to check if there are vagrant processes currently running. Since I got an exception that Vagrant said that the box was locked:
d:\Projects\vagrant\ol77_gui>vagrant global-status
id       name   provider state  directory
There are no active Vagrant environments on this computer! Or,
you haven't destroyed and recreated Vagrant environments that were
started with an older version of Vagrant.

If your box is running it could say something like:
d:\Projects\vagrant\ol77_gui>vagrant  global-status
id       name   provider   state   directory
42cbd44  darwin virtualbox running d:/Projects/vagrant/ol77_gui

The above shows information about all known Vagrant environments
on this machine. This data is cached and may not be completely
up-to-date (use "vagrant global-status --prune" to prune invalid
entries). To interact with any of the machines, you can go to that
directory and run Vagrant, or you can use the ID directly with
Vagrant commands from any directory. For example:
"vagrant destroy 1a2b3c4d"

You could also do a prune of invalid entries:
d:\Projects\vagrant\ol77_gui>vagrant  global-status --prune
id       name   provider   state   directory
42cbd44  darwin virtualbox running d:/Projects/vagrant/ol77_gui

In the Vagrantfile I set the ssh username and password:
Vagrant.configure("2") do |config|
  # The most common configuration options are documented and commented below.
  # For a complete reference, please see the online documentation at
  # https://docs.vagrantup.com.

  # Every Vagrant development environment requires a box. You can search for
  # boxes at https://vagrantcloud.com/search.
  config.vm.box = BOX_NAME
  config.vm.define "darwin"
  config.vm.provider :virtualbox do |vb|
    vb.name = VM_NAME
    vb.gui = true
    vb.memory = VM_MEMORY
    vb.cpus = VM_CPUS
    # Set clipboard and drag&drop bidirectional
    vb.customize ["modifyvm", :id, "--clipboard-mode", "bidirectional"]
    vb.customize ["modifyvm", :id, "--draganddrop", "bidirectional"]

It is common to have the vagrant user's password be "vagrant". Lastly I "upped" my VM. And this all seemed to solve my memory allocation problems.

Apparently, we can't use the ssh-key to provision the box.