As of Corda 2.0, constructors of classes annotated with #CordaService are automatically invoked during node start. Corda Service original class implementations may, for example, initialise network connections in their constructors, which may not be desirable in unit tests.
My current workaround is to use a boolean flag (set from an environment variable) that marks whether to run or skip Corda Service initialisation code. Is there a better way to do this (e.g. inject a mock Corda Service class instance into MockNetwork.MockNode without using PowerMock and the like)?
Your workaround is currently the best way to handle this. See https://r3-cev.atlassian.net/browse/CORDA-989.
In your approach, you might define your service as follows:
#CordaService
class MyService(services: ServiceHub) : SingletonSerializeAsToken() {
private companion object {
val logger = loggerFor<MyService>()
val load = System.getenv("enable-my-service")
}
init {
if (load == "false") {
logger.warn("Not loading service")
} else {
logger.info("Loading service")
}
}
}
And then you can toggle your service off in tests using an environment variable.
Related
Is it possible to pass anonymous object implementations over Corda's RPC interface? - For example:
Workflow
#CordaSerializable
interface ExampleInterface {
val number: Int
}
#StartableByRPC
class SquareFlow(private val example: ExampleInterface) : FlowLogic<Int>() {
#Suspendable
override fun call(): Int = example.number * example.number
}
RPC Client
val value = object : ExampleInterface {
override val number: Int = 5
}
return rpc.startFlowDynamic(SquareFlow::class.java, value).returnValue.getOrThrow()
Exception
net.corda.client.rpc.RPCException: java.util.List<*> -> Unable to create an object serializer for type class com.example.client.ExampleService$square$value$1: No unique deserialization constructor can be identified
Either annotate a constructor for this type with #ConstructorForDeserialization, or provide a custom serializer for it
Whilst this is an example, what I actually want to do is pass object : TypeReference<SomeType>(){} over RPC.
As per Corda's docs anonymous objects are not supported by their AMQP
serialization which is used in the RPC client which you can read here. This is likely due to a public constructor being needed for serialization and deserialization I imagine.
Additionally I assume you are referencing Jackson's TypeReference class which uses generics which are only available at compile time. Due to how Corda's serialization works the generic won't be available at runtime so when the TypeReference is passed from the rpc application to the corda node the corda node will not know what the generic is. I haven't tried this myself but I believe some exception will be thrown whenever the flow attempts to suspend or possibly before in the RPC client.
You can probably pass a jackson JavaType class to a flow without a problem as I believe it is whitelisted as serializable in Corda by default, if not look at how to whitelist the class here using the SerializationWhitelist. You can obtain the JavaType by instantiating a TypeFactory and calling constructType(new TypeReference<SomeType> {}). then just pass the JavaType to your flows constructor.
In a SpringBoot application, I have the following configuration:
axon:
axonserver:
servers: "${AXON_SERVER:localhost}"
serializer:
general: jackson
messages: jackson
events: jackson
logging.level:
org.axonframework.modelling.saga: debug
Downsizing the scenario to bare minimum, the relevant portion of Saga class:
#Slf4j
#Saga
#ProcessingGroup("AuctionEventManager")
public class AuctionEventManagerSaga {
#Autowired
private transient EventScheduler eventScheduler;
private ScheduleToken scheduleToken;
private Instant auctionTimerStart;
#StartSaga
#SagaEventHandler(associationProperty = "auctionEventId")
protected void on(final AuctionEventScheduled event) {
this.auctionTimerStart = event.getTimerStart();
// Cancel any pre-existing previous job, since the scheduling thread might be lost upon a crash/restart of JVM.
if (this.scheduleToken != null) {
this.eventScheduler.cancelSchedule(this.scheduleToken);
}
this.scheduleToken = this.eventScheduler.schedule(
this.auctionTimerStart,
AuctionEventStarted.builder()
.auctionEventId(event.getAuctionEventId())
.build()
);
}
#EndSaga
#SagaEventHandler(associationProperty = "auctionEventId")
protected void on(final AuctionEventStarted event) {
log.info(
"[AuctionEventManagerSaga] Current state: {scheduleToken={}, auctionTimerStart={}}",
this.scheduleToken,
this.auctionTimerStart
);
}
}
In the final compiled class, we will end up having 4 properties: log (from #Slf4j), eventScheduler (transient, #Autowired), scheduleToken and auctionTimerStart.
For reference information, here is a sample of the general approach I've been using for both Command and Event classes:
#Value
#Builder
#JsonDeserialize(builder = AuctionEventStarted.AuctionEventStartedBuilder.class)
public class AuctionEventStarted {
AuctionEventId auctionEventId;
#JsonPOJOBuilder(withPrefix = "")
public static final class AuctionEventStartedBuilder {}
}
When executing the code, you get the following output:
2020-05-12 15:40:01.180 DEBUG 1 --- [mandProcessor-4] o.a.m.saga.repository.jpa.JpaSagaStore : Updating saga id c8aff7f7-d47f-4616-8a96-a40044cb7e3b as {}
As soon as the general serializer is changed to xstream, the content is serialized properly, but I face another issue during deserialization, since I have private static final class Builder classes using Lombok.
So is there a way for Axon to handle these scenarios:
1- Axon to safely manage Jackson to ignore #Autowired, transient and static properties from #Saga classes? I've attempted to manually define #JsonIgnore at non-state properties and it still didn't work.
2- Axon to safely configure XStream to ignore inner classes (mostly Builder classes implemented as private static final)?
Thanks in advance,
EDIT: I'm pursuing a resolution using my preferred serializer: JSON. I attempted to modify the saga class and extend JsonSerializer<AuctionEventManagerSaga>. For that I implemented the methods:
#Override
public Class<AuctionEventManagerSaga> handledType() {
return AuctionEventManagerSaga.class;
}
#Override
public void serialize(
final AuctionEventManagerSaga value,
final JsonGenerator gen,
final SerializerProvider serializers
) throws IOException {
gen.writeStartObject();
gen.writeObjectField("scheduleToken", value.eventScheduler);
gen.writeObjectField("auctionTimerStart", value.auctionTimerStart);
gen.writeEndObject();
}
Right now, I have something being serialized, but it has nothing to do with the properties I've defined:
2020-05-12 16:20:01.322 DEBUG 1 --- [mandProcessor-0] o.a.m.saga.repository.jpa.JpaSagaStore : Storing saga id c4b5d94c-7251-40a5-accf-332768b1cacd as {"delegatee":null,"unwrappingSerializer":false}
EDIT 2 Decided to add more insight into the issue I experience when I switch general to use XStream (even though it's somewhat unrelated to the main issue described in the title).
Here is the issue it complains to me:
2020-05-12 17:08:06.495 DEBUG 1 --- [ault-executor-0] o.a.a.c.command.AxonServerCommandBus : Received command response [message_identifier: "79631ffb-9a87-4224-bed3-a957730dced7"
error_code: "AXONIQ-4002"
error_message {
message: "No converter available\n---- Debugging information ----\nmessage : No converter available\ntype : jdk.internal.misc.InnocuousThread\nconverter : com.thoughtworks.xstream.converters.reflection.ReflectionConverter\nmessage[1] : Unable to make field private static final jdk.internal.misc.Unsafe jdk.internal.misc.InnocuousThread.UNSAFE accessible: module java.base does not \"opens jdk.internal.misc\" to unnamed module #7728643a\n-------------------------------"
location: "1#600b5b87a922"
details: "No converter available\n---- Debugging information ----\nmessage : No converter available\ntype : jdk.internal.misc.InnocuousThread\nconverter : com.thoughtworks.xstream.converters.reflection.ReflectionConverter\nmessage[1] : Unable to make field private static final jdk.internal.misc.Unsafe jdk.internal.misc.InnocuousThread.UNSAFE accessible: module java.base does not \"opens jdk.internal.misc\" to unnamed module #7728643a\n-------------------------------"
}
request_identifier: "2f7020b1-f655-4649-bbe0-d6f458b3c2f3"
]
2020-05-12 17:08:06.505 WARN 1 --- [ault-executor-0] o.a.c.gateway.DefaultCommandGateway : Command 'ACommandClassDispatchedFromSaga' resulted in org.axonframework.commandhandling.CommandExecutionException(No converter available
---- Debugging information ----
message : No converter available
type : jdk.internal.misc.InnocuousThread
converter : com.thoughtworks.xstream.converters.reflection.ReflectionConverter
message[1] : Unable to make field private static final jdk.internal.misc.Unsafe jdk.internal.misc.InnocuousThread.UNSAFE accessible: module java.base does not "opens jdk.internal.misc" to unnamed module #7728643a
-------------------------------)
Still no luck on resolving this...
I've worked on Axon systems where the only used Serializer implementation was the JacksonSerializer too. Mind you though, this is not what the Axon team recommends. For messages (i.e. commands, events and queries) it makes perfect sense to use JSON as the serialized format. But switching the general Serializer to jackson means you have to litter your domain logic (e.g. your Saga) with Jackson specifics "to make it work".
Regardless, backtracking to my successful use case of jackson-serialized-sagas. In this case we used the correct match of JSON annotations on the fields we desired to take into account (the actual state) and to ignore the one's we didn't want deserialized (with either transient or #JsonIgnore). Why both do not seem to work in your scenario is not entirely clear at this stage.
What I do recall is that the referenced project's team very clearly decided against Lombok due to "overall weirdnes" when it comes to de-/serialization. As a trial it thus might be worth to not use any Lombok annotations/logic in the Saga class and see if you can de-/serialize it correctly in such a state.
If it does work at that moment, I think you have found your culprit for diving in further search.
I know this isn't an exact answer, but I hope it helps you regardless!
Might be worthwhile to share the repository where this problems occurs in; might make the problem clearer for others too.
I was able to resolve the issue #2 when using XStream as general serializer.
One of the Sagas had an #Autowired dependency property that was not transient.
XStream was throwing some cryptic message, but we managed to track the problem and address it.
As for JSON support, we had no luck. We ended up switched everything to XStream for now, as the company only uses Java and it would be ok to decode the events using XStream.
Not the greatest solution, as we really wanted (and hoped) JSON would be supported properly out of the box. Mind you, this is in conjunction with using Lombok which caused for the nuisance in this case.
I have a service which I use both from a custom command and an HTML page. I want to prevent multiple executions of the the service in parallel. For the command there is the Lock component that does that. But is it possible to achieve the same thing for a controller method ?
The lock component doesn't work if the service is called from a controller:
$store = new FlockStore(sys_get_temp_dir());
$factory = new Factory($store);
$lock = $factory->createLock('MY_SERVICE');
I wanted to avoid calling the command from the controller (that's why I created a service) mainly because the service doesn't have the same output for the HTML page and the CLI.
Inject the lock Factory into your service directly instead of creating the lock in the command AND in the controller.
First you have to install Lock Component:
composer require symfony/lock
Then, for example, you can declare your service like this:
use Symfony\Component\Lock\Factory as LockFactory;
class MyService {
private $lock;
public function __construct(LockFactory $lockFactory) {
$this->lock = $lockFactory->createLock('LOCK_KEY');
}
public function doWork() {
$this->lock->acquire();
try {
// DO THINGS
} finally {
$this->lock->release();
}
}
}
I said:
The lock component doesn't work if the service is called from a controller:
Actually the issue I had was the Symfony built-in dev server which is single-threaded, so requests can't be executed in parallel, while the CLI PHP is multi-threaded. I couldn't run the script in parallel through the dev server, request were queued, service script was never locked.
The lock component is working the same whether it's called from a command or a controller.
Using the lock like this in the service works fine:
use Symfony\Component\Lock\Factory;
use Symfony\Component\Lock\Store\FlockStore;
$store = new FlockStore(sys_get_temp_dir());
$factory = new Factory($store);
$lock = $factory->createLock('LOCK_KEY');
if ($lock->acquire()) {
//some locked code
$lock->release();
}
I have the gRPC server code as below:
public void buildServer() {
List<BindableService> theServiceList = new ArrayList<BindableService>();
theServiceList.add(new CreateModuleContentService());
theServiceList.add(new RemoveModuleContentService());
ServerBuilder<?> sb = ServerBuilder.forPort(m_port);
for (BindableService aService : theServiceList) {
sb.addService(aService);
}
m_server = sb.build();
}
and client code as below:
public class JavaMainClass {
public static void main(String[] args) {
CreateModuleService createModuleService = new CreateModuleService();
ESDStandardResponse esdReponse = createModuleService.createAtomicBlock("8601934885970354030", "atm1");
RemoveModuleService moduleService = new RemoveModuleService();
moduleService.removeAtomicBlock("8601934885970354030", esdReponse.getId());
}
}
While I am running the client I am getting an exception as below:
Exception in thread "main" io.grpc.StatusRuntimeException: UNIMPLEMENTED: Method grpc.blocks.operations.ModuleContentServices/createAtomicBlock is unimplemented
at io.grpc.stub.ClientCalls.toStatusRuntimeException(ClientCalls.java:233)
at io.grpc.stub.ClientCalls.getUnchecked(ClientCalls.java:214)
at io.grpc.stub.ClientCalls.blockingUnaryCall(ClientCalls.java:139)
In the above server class, if I am commenting the line theServiceList.add(new RemoveModuleContentService()); then the CreateModuleContentService service is working fine, also without commenting all the services of RemoveModuleContentService class are working as expected, which means the problem is with the first service when another gets added.
Can someone please suggest how can I add two services to Server Builder.
A particular gRPC service can only be implemented once per server. Since the name of the gRPC service in the error message is ModuleContentServices, I'm assuming CreateModuleContentService and RemoveModuleContentService both extend ModuleContentServicesImplBase.
When you add the same service multiple times, the last one wins. The way the generated code works, every method of a service is registered even if you don't implement that particular method. Every service method defaults to a handler that simply returns "UNIMPLEMENTED: Method X is unimplemented". createAtomicBlock isn't implemented in RemoveModuleContentService, so it returns that error.
If you interact with the ServerServiceDefinition returned by bindService(), you can mix-and-match methods a bit more, but this is a more advanced API and is intended more for frameworks to use because it can become verbose to compose every application service individually.
long time ASP.Net interface developer being asked to learn WCF, looking for some education on more architecture related fronts - as its not my strong suit but I'm having to deal.
In our current ASMX world we adopted a model of creating ServiceManager static classes for our interaction with web services. We're starting to migrate to WCF, attempting to follow the same model. At first I was dealing with performance problems, but I've tweaked a bit and we're running smoothly now, but I'm questioning my tactics. Here's a simplified version (removed error handling, caching, object manipulation, etc.) of what we're doing:
public static class ContentManager
{
private static StoryManagerClient _clientProxy = null;
const string _contentServiceResourceCode = "StorySvc";
// FOR CACHING
const int _getStoriesTTL = 300;
private static Dictionary<string, GetStoriesCacheItem> _getStoriesCache = new Dictionary<string, GetStoriesCacheItem>();
private static ReaderWriterLockSlim _cacheLockStories = new ReaderWriterLockSlim();
public static Story[] GetStories(string categoryGuid)
{
// OMITTED - if category is cached and not expired, return from cache
// get endpoint address from FinderClient (ResourceManagement SVC)
UrlResource ur = FinderClient.GetUrlResource(_contentServiceResourceCode);
// Get proxy
StoryManagerClient svc = GetStoryServiceClient(ur.Url);
// create request params
GetStoriesRequest request = new GetStoriesRequest{}; // SIMPLIFIED
Manifest manifest = new Manifest{}; // SIMPLIFIED
// execute GetStories at WCF service
try
{
GetStoriesResponse response = svc.GetStories(manifest, request);
}
catch (Exception)
{
if (svc.State == CommunicationState.Faulted)
{
svc.Abort();
}
throw;
}
// OMITTED - do stuff with response, cache if needed
// return....
}
internal static StoryManagerClient GetStoryServiceClient(string endpointAddress)
{
if (_clientProxy == null)
_clientProxy = new StoryManagerClient(GetServiceBinding(_contentServiceResourceCode), new EndpointAddress(endpointAddress));
return _clientProxy;
}
public static Binding GetServiceBinding(string bindingSettingName)
{
// uses Finder service to load a binding object - our alternative to definition in web.config
}
public static void PreloadContentServiceClient()
{
// get finder location
UrlResource ur = FinderClient.GetUrlResource(_contentServiceResourceCode);
// preload proxy
GetStoryServiceClient(ur.Url);
}
}
We're running smoothly now with round-trip calls completing in the 100ms range. Creating the PreloadContentServiceClient() method and adding to our global.asax got that "first call" performance down to that same level. And you might want to know we're using the DataContractSerializer, and the "Add Service Reference" method.
I've done a lot of reading on static classes, singletons, shared data contract assemblies, how to use the ChannelFactory pattern and a whole bunch of other things that I could do to our usage model...admittedly, some of its gone over my head. And, like I said, we seem to be running smoothly. I know I'm not seeing the big picture, though. Can someone tell me what I've ended up here with regards to channel pooling, proxy failures, etc. and why I should head down the ChannelFactory path? My gut says to just do it, but my head can't comprehend why...
Thanks!
ChannelFactory is typically used when you aren't using Add Service Reference - you have the contract via a shared assembly not generated via a WSDL. Add Service Reference uses ClientBase which is essentially creating the WCF channel for you behind the scenes.
When you are dealing with REST-ful services, WebChannelFactory provides a service-client like interface based off the shared assembly contract. You can't use Add Service Reference if your service only supports a REST-ful endpoint binding.
The only difference to you is preference - do you need full access the channel for custom behaviors, bindings, etc. or does Add Service Reference + SOAP supply you with enough of an interface for your needs.