I have a class with dependencies which I want to hot deploy without restarting the dependencies. The class has an interface but there's only one concrete implementation.
Initially I created a single bundle with exported the interface and registered the implementation using activator and implementation classes which were not exported. However, if I update the bundle, bundles which use the registered service get restarted after the update when PackageAdmin#refreshPackages is called (this is automatic when using fileinstall).
I have fixed this by creating a separate api bundle.
Is this the best way to achieve this?
Would you ever have a bundle which exports its api and includes the implementation in the same bundle. As far as I can see any give bundle would either export all its classes or no classes. What am I missing?
It is definitely a best practice to separate API bundles from their implementations in OSGi. If you do this, then any bundle that uses the API only needs to import classes from the API bundle, which can allow you to change implementations at runtime without restarting your dependent bundles.
Ideally your implementation bundle would implement the interface and export implementation as a service on the API provided interface. This allows the client bundles to utilize the service without referencing the implementation bundle.
In Apache Sling we do both: major APIs are in their own bundles, but for smaller things like extensions or optional components we often provide the default implementation in the same bundle as the API.
In the latter case, you can still allow for those default services to be replaceable, for example using service ranking values when you want to override them.
A bundle does not have to export all its classes, our bundles which include default services export just the API packages (and the default implementations are in different packages, obviously).
Unless there is a hard requirement to be able to replace implementation at runtime, without restarting client bundles, I would personally advocate keeping the explicit dependency link between API and implementation (either by including impl classes in the API bundle, or by having the API bundle import implementation packages from the impl. bundle).
The pb with the patterns suggested above is that they break the dependency chain. The benefits of dependency management go far beyond simple API compatibility, they also include ensuring predictable, consistent runtime behavior, as well as compatibility with the deployment ecosystem, and all of those require managing the implementation dependencies.
Related
I've been an AEM developer for almost a year now. I know AEM uses 'Declarative Services component framework' to manage life cycle of OSGi components.
Consider a scenario when i would export a package from a bundle and import that package from another bundle, i could create objects of classes in first bundle inside second bundle as well. it's a import-export contract in this case.
my question is when i should be using component framework to manage the lifecycle of my objects and when to handle it myself by creating them when required.
In an ideal design, you would NOT in fact be able to create objects from the exported package; because that package would contain only interfaces. This makes it a "pure" contract (API) export. If there are classes in there that you can directly instantiate, then they are implementation classes.
In general it is far better to export only pure APIs and to keep implementation classes hidden. There are two main reasons:
Implementation classes tend to have downstream dependencies. If you depend directly from implementation class to implementation class then you get a very large and fragile dependency graph... and eventually that graph will contain a cycle. In fact it's almost inevitable that it will. At that point, your application is not modular because you cannot deploy or change any part of it independently.
Pure interfaces can be analysed for compatibility between versions. As a consumer or a provider of an API, you know exactly which versions of the API you can support because the API does not contain executable code. However if you have a dependency onto an implementation class, then you never really know when they break compatibility because the breakage could happen deep down in executable code that you can't easily analyse.
If your objects are services then there's no question, they have to be OSGi components.
For other things, my first choice is OSGi components, unless they're trivial objects like data holders or something similar.
If an object requires configuration or refers to OSGi services then it's also clearly an OSGi component.
In general, it's best IMO to think in services and define your package exports as the minimum that allows other bundles to use a bundle's services. Unless a bundle is clearly a reusable library like commons-io (to take a simple example).
We run Restlet 2.1 in an OSGi environment (Equinox) as bundle (ie. not as library within a bundle). The problem is that the Restlet Engine does not detect helpers (like converters) that are provided by Restlet extensions. Specifically, the EngineClassLoader#getResources() call does not return any result. The extensions are also deployed as OSGi bundles in the target platform.
Is automatic converter registration actually supposed to work within OSGi environments?
In fact, Restlet supports such feature thanks to a dedicated activator (see the Activator class in the package org.restlet.engine.internal).
This activator introspects bundles to find out the following things:
servers corresponding to registered servers
servers corresponding to registered clients
authenticators corresponding to registered clients
converters
Be aware that to use this feature, we must use the OSGi edition of Restlet since it's the only that has the MANIFEST file of the org.restlet bundle with the activator class specified. Otherwise you don't have to care about the bundle loading order...
Hope it helps you.
Thierry
Unless the Restlet-bundle explicitly imports the packages that contain the extensions (and I doubt it does, and it shouldn't), it wouldn't be able to load them, because bundles have isolated class-spaces.
A possible solution would be to provide the extensions as fragments attached to the Restlet-bundle. Thus, if you make it use the bundle-classloader (the documentation says this can be done by setting the Engines classloader), it would be able to load classes from the fragments.
Indeed it doesn't quite work for OSGi, as it depends on the ability to see the entire class space.
The way to do this in OSGi would be to use the service registry for the extensions, but that only works for OSGi aware libraries.
There is some help on the way: In the recently released OSGi 5 (Service Loader Mediator) there will be support to 'bridge' META-INF/services (I don't know if Restlet uses those, though) onto OSGi services, so 'legacy' libraries should work well within OSGi.
There is an implementation in Apache Aries called Spi-Fly. I looked at it briefly a while back. It might do the trick for you, it might not.
My application obtains class names from a properties file. Classes represented by these class names could reside in certain OSGI bundles unknown at advance so in order to instantiate them I firstly have to find which bundle these classes belong to. I'm thinking about getting all installed bundles from BundleContext#getBundles, this means I have to obtain reference to BundleContext in AbstractUIPlugin#start. But I'm not sure if holding reference to BundleContext is the right thing to do since it's should be used only in the start method. So I need advice from OSGI experts here about the alternatives to get list of bundles.
Any help would be greatly appreciated.
Regards,
Setya
This is not really how OSGi is intended. If a bundle has something to add to the 'global' context, you should register a service. So each bundle that has something to share can do that in its own start method.
Some other component (DS, ServiceTracker, Blueprint, something like that) can then listen to these events, and act accordingly.
This is really important, if you start manually searching through all bundles you completely lose the advantages of OSGi.
Like written before you should try to use services to achieve what you want. I guess you have an Interface with one or more implementations that should be installable at runtime. So if you control the bundles that implement the interface then simply let them install their implementation as a service by using an Activator or a blueprint context. You can use service properties to describe your implementation.
The bundles that need the implementation can then lookup the services using a service tracker or a service reference in blueprint.
If that is not possible then you can use the bundle context to obtain the running bundles and instantiate the classes but this is not how OSGi should work. You will run into classloading problems as the bundle that tries to instantiate the classes will not have direct access to them.
Your bundle gets control at start up through the bundle activator, or better, through DS. At that time it can register services with the services registry so others can find/use them.
The route your planning to go (properties that name classes) is evil since you undoubtedly will run in class loading hell. Modularity is about hiding your implementation details, the name of your implementation classes are such details.
Exposing implementation classes in properties files is really bad practice and it looses the advantage of modularity. It does not matter if another class refers to your implementation class or a property file, the problem is that the impl. class is exposed.
Unfortunately this model has become so prevalent in our industry that many developers think it is normal :-(
OSGi allows you share instances typed by interfaces in a way that allows the implementation class to only be known inside the module.
As an OSGi newbie, I'm trying to wrap my head around the boundaries of the OSGi runtime. My app, which is not build on OSGi, i.e. it's not running in an OSGi container, starts an OSGi container into which we deploy OSGi bundles at run time. Some of these bundles register services. Later, in our non-OSGi code, we obtain those services and use them.
I'm having trouble wrapping my feeble mind around the OSGi boundaries here. To be specific, when I obtain a service and invoke one of its methods, can I assume that all of the subsequent execution is executing within the OSGi container (Felix)? In other words, are dependencies in that code resolved via the OSGi modularity mechanisms? Or did I lose that OSGi management because I am using the service from non-OSGi code?
If my question seems founded in obvious mistaken assumptions about OSGi, please feel free to point them out.
Chad, to more effectively answer your question, I'd like to know a few things:
1) How exactly are you getting the service reference from an external application?
2) Is the external application a stand-alone application, or is it inside of a different container? If so, there are ways to make that happen.
The question you pose is interesting. Lets put it into a context. Lets assume you are able to get a reference to an OSGi service from Felix by an external application. When you use this service, you will be interacting with it via an interface. In that interface in OSGi, you will have import statements referenced which will be used in the method signatures of the interface and also in any final attributes. These import statements will have thier matching dependant libraries defined in your pom.xml file.
In order to use the service by an external application, you will need to publish an API ".jar" file that will contain the interface, and will reference the interfaces' dependancies. Your external application will need to use that API, and will likely have it assembled into your .war, .jar, or .ear file's lib directory. Because of this, none of your external application's dependancies can conflict with your API dependancies.
As long as you can use the API, then you're right, none of the SPI's dependencies matter. You can use Spring 3.0.4.RELEASE in your external app and still use Spring 2.5.6.SNAPSHOT in your OSGi application. As long as the API doesn't have any dependancies that conflict with the external application, you should be ok. The trick here is that you need to put your interfaces into a minimal .jar file as your API, and then put your implementation details into an SPI. Your external ap will use the API, and inside OSGI, you will use both the API and SPI.
Please let me know if this helps.
If you can get the service, then the dependencies are satisfied by definition, because bundles cannot provide services unless their dependencies are satisfied. Executing services on the outside doesn't really change anything.
Do the packages that contain OSGI services need to be exported in order to publish to the service registry? Or does publishing to the service registry get round the need to export the package?
Short answers; 1st question = no, 2nd question = no, there's no "workaround"
The principle of "Design by contract" comes into play. Basically we provide an interface for consumers to depend on and providers to implement - this decouples our consumer code for the implementation details of our implementors. See http://www.osgi.org/blog/2011/08/package-as-contract.html for a good overview and http://en.wikipedia.org/wiki/Design_by_contract.
Given the interface (the API we wish to use) and the implementation (the concrete implementation of the API interface) and the consumer (code wishing to use an implementation of the interface), with OSGi you have a number of options.
A bundle can provide both interface and implementation, but it must export the interface if this service is to be used outside of the bundle itself
Having a separate bundle for the interface (which it exports) and other bundle(s) for the implementation(s) - the implementing bundles must import the interface's package but need not reveal their own implementation.
The advantages of number 2 include allowing multiple providers (OSGi uses LDAP filters that can be used to specify metadata matching on the registered services' properties), and leveraging the dynamic binding nature of OSGi (i.e. you can change implementation provider at runtime).
One thing to bear in mind, is that the consumer and provider must use the same classes as defined by the package version and exporting bundle (in other words the bundle classloader that loads the API classes must be the same for both consumer and provider).
The types of the service should be in exported packages. Services are typed and build upon the type sharing of the module layer. So you need to make sure that the provider of the service and the consumers of the service can all share the same type. So some bundle must export the package containing the type so that the provider and the consumers can import the package. This allows the framework to verify type safety of the service.