Extending OMERO.server

Overview

Despite all the effort put into building OMERO, it will never satisfy the requirements of every group. Where we have seen it useful to do so, we have created extension points which can be used by third-party developers to extend, improve, and adapt OMERO. We outline most of these options below as well as some of their trade-offs. We are also always interested to hear other possible extension points. Please contact the ome-devel mailing list with any such suggestions.

Existing extension points

To get a feeling for what type of extension points are available, you might want to take a look at the following pages. Many of them will point you back to this page for packaging and deploying your new code.

Extending the Model

The OME Data Model and its OMERO representation, the OME-Remote Objects, intentionally draw lines between what metadata can be supported and what cannot. Though we are always examining new fields for inclusion, it is not possible to represent everyone’s model within OME.

Structured annotations

The primary extension point for including external data are the Working with annotations (SAs). SAs are designed as email-like attachments which can be associated with various core metadata types. In general, they should link to information outside of the OME model, i.e. information which OMERO clients and servers do not understand. URLs can point to external data sources, or XML in a non-OME namespace can be attached.

The primary drawbacks are that the attachments are opaque and cannot be used in a fine-grain manner.

Code generation

Since it is prohibitive to model full objects with the SAs, one alternative is to add types directly to the generated code. By adding a file named *.ome.xml to src/main/resources/mappings and running a full-build, it is possible to have new objects generated in all OMERO.blitz languages. Supported fields include:

  • boolean

  • string

  • long

  • double

  • timestamp

  • links to any other ome.model.* object, including enumerations

For example:

<types>
  <!-- "named" and "described" are short-cuts to generate the fields "name" and "description" -->
  <type id="ome.model.myextensions.Example" named="true" described="true">
    <required name="valueA" type="boolean"/>  <!-- This is NONNULL -->
    <optional name="valueB" type="long"/>     <!-- This is nullable -->
    <onemany  name="images" type="ome.model.core.Image"/> <!-- A set of images -->
  </type>
</types>

Collections of primitive values like <onemany name="values" type="long"/> are not supported. Please see the existing mapping files for more examples of what can be done.

The primary drawback of code-generating your own types is isolation and maintenance. Firstly, your installation becomes isolated from the rest of the OME ecosystem. New types are not understood by other servers and clients, and cannot be exported or shared. Secondly, you will need to maintain your own server and client builds of the system, since the provided binary builds would not have your new types.

Measurement results

For storing large quantities of only partially structured data, such as tabular/CSV data with no pre-defined columns, neither the SAs nor the code-generation extensions are ideal. SAs cannot easily be aggregated, and code-generation would generate too many types. This is particularly clear in the storage and management of HCS analysis results.

To solve this problem, we provide the OMERO.tables API for storing tabular data indexed via Roi, Well, or Image id.

Services

Traditionally, services were added via Java interfaces in the src/main/java/ome/api package. The creation of such “core” services is described under How To create a service. However, with the introduction of OMERO.blitz, it is also possible to write blitz-only services which are defined by a slice definition under src/main/slice/omero.

A core service is required when server internal code should also make use of the interface. Since this is very rarely the case for third-party developers wanting to extend OMERO, only the creation of blitz services will be discussed here.

Add a slice definition

The easiest possible service definition in slice is:

module example {
  interface NewService {
    void doSomething();
  };
};

This should be added to any existing or a new *.ice file under the src/main/slice/omero directory. After the next ant build, stubs will be created for all the OMERO.blitz languages, i.e. OMERO Java language bindings, OMERO Python language bindings, and OMERO C++ language bindings.

Note

Once you have gotten your code working, it is most re-usable if you can put it all in a single directory under tools/. These components also have their resources/*.ice files turned into code, and they can produce their own artifacts which you can distribute without modifying the main code base.

Warning: exceptions

You will need to think carefully about what exceptions to handle. Ice (especially OMERO C++ language bindings) does not handle exceptions well that are not strictly defined. In general, if you would like to add your own exception type, feel free to do so, but either 1) subclass omero::ServerError or 2) add to the appropriate throws clauses. And regardless, if you are accessing any internal OMERO API, add omero::ServerError to your throws clause.

See Exception handling for more information.

Java implementation using _Disp

To implement your service, create a class subclassing “example._NewServiceDisp” class which was code-generated. In this example, the class would be named “NewServiceI” by convention. If this service needs to make use of any of the internal API, it should do so via dependency injection. For example, to use IQuery add either:

void setLocalQuery(LocalQuery query) {
    this.query = query;
}

or

NewServiceI(LocalQuery query) {
    this.query = query;
}

The next step “Java Configuration” will take care of how those objects get injected.

Java implementation using _Tie

Rather than subclassing the _Disp object, it is also possible to implement the _Tie interface for your new service. This allows wrapping and testing your implementation more easily at the cost of a little indirection. You can see how such an object is configured in blitz-servantDefinitions.

Java configuration

Configuration in the Java servers takes place via Spring. One or more files matching a pattern like ome/services/blitz-*.xml should be added to your application.

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE beans PUBLIC "-//SPRING//DTD BEAN//EN" "http://www.springframework.org/dtd/spring-beans.dtd">
<beans>

  <bean class="NewServiceI">
    <description>
    This is a simple bean definition in Spring. The description is not necessary.
    </description>
    <constructor-arg ref="internal-ome.api.IQuery"/>
  </bean>

</beans>

The three patterns which are available are:

  • ome/services/blitz-*.xml - highest-level objects which have access to all the other defined objects.

  • ome/services/services-*.xml - internal server objects which do not have access to blitz-*.xml objects.

  • ome/services/db-*.xml - base connection and security objects. These will be included in background java process like the index and pixeldata handlers.

    Note

    Password Provider and similar should be included at this level.

See src/main/resources/ome/services and src/main/resources/ome/services for all the available objects.

Java deployment

Finally, these resources should all be added to OMERO_DIST/lib/server/extensions.jar:

  • the code generated classes

  • your NewServiceI.class file and any related classes

  • your ome/service/blitz-*.xml file (or other XML)

Non-service beans

In addition to writing your own services, the instructions above can be used to package any Spring-bean into the OMERO server. For example:

//
// MyLoginAttemptListener.java
//
import ome.services.messages.LoginAttemptMessage;

import org.springframework.context.ApplicationListener;

/**
 * Trivial listener for login attempts.
 */

public class MyLoginAttemptListener implements
        ApplicationListener<LoginAttemptMessage> {

    public void onApplicationEvent(LoginAttemptMessage lam) {
        if (lam.success != null && !lam.success) {
            // Do something
        }
    }

}
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE beans PUBLIC "-//SPRING//DTD BEAN//EN" "http://www.springframework.org/dtd/spring-beans.dtd">
<!--
//
// ome/services/blitz-myLoginListener.xml
//
-->
<beans>
  <bean class="myLoginAttemptListener" class="MyLoginAttemptListener">
    <description>
    This listener will be added to the Spring runtime and listen for all LoginAttemptMessages.
    </description>
  </bean>

</beans>

Putting MyLoginAttemptListener.class and ome/services/blitz-myLoginListener.xml into lib/server/extensions.jar is enough to activate your code:

~/example $ ls -1
MyLoginListener.class
MyLoginListener.java
lib
...
~/example $ jar cvf lib/server/extensions.jar MyLoginListener.class ome/services/blitz-myLoginListener.xml
added manifest
adding: MyLoginListener.class(in = 0) (out= 0)(stored 0%)
adding: ome/services/blitz-myLoginListener.xml(in = 0) (out= 0)(stored 0%)

Servers

With the OMERO.grid infrastructure, it is possible to have your own processes managed by the OMERO infrastructure. For example, at some sites, NGINX is started to host OMERO.web framework. Better integration is possible however, if your server also uses the Ice remoting framework.

One way or the other, to have your server started, monitored, and eventually shutdown by OMERO.grid, you will need to add it to the “application descriptor” for your site. When using:

omero admin start

the application descriptor used is etc/grid/default.xml. The <application> element contains various <node>s. Each node is a single daemon process that can start and stop other processes. Inside the nodes, you can either directly add a <server> element, or in order to reuse your description, you can use a <server-instance> which must refer to a <server-template>.

To clarify with an example, if you have a simple application which should watch for newly created Images and send you an email: mail_on_import.py, you could add this in either of the following ways:

Server element

<node name="my-emailer-node">  <!-- this could also be an existing node, but it must be unique -->
  <server id="my-emailer-server" exe="/home/josh/mail_on_import.py" activation="always">
    <env>${PYTHONPATH}</env>
    <!-- The adapter name must also be unique -->
    <adapter name="MyAdapter" register-process="true" endpoints="tcp"/>
  </server>
</node>

Server-template and server-instance elements

<server-template id="emailer-template">  <!-- must also be unique -->
  <property name="user"/>
  <server id="emailer-server-${user}" exe="/home/${user}/mail_on_import.py" activation="always">
    <env>${PYTHONPATH}</env>
    <adapter name="MyAdapter" register-process="true" endpoints="tcp"/>
  </server>
</server-template>

<node name="our-emailer-node">
  <server-instance id="emailer-template" user="ann">
  <server-instance id="emailer-template" user="ann">
</node>