By Version

Context Propagation in Quarkus

Traditional blocking code uses ThreadLocal variables to store contextual objects in order to avoid passing them as parameters everywhere. Many Quarkus extensions require those contextual objects to operate properly: RESTEasy Reactive, ArC and Transaction for example.

If you write reactive/async code, you have to cut your work into a pipeline of code blocks that get executed "later", and in practice after the method you defined them in have returned. As such, try/finally blocks as well as ThreadLocal variables stop working, because your reactive code gets executed in another thread, after the caller ran its finally block.

SmallRye Context Propagation an implementation of MicroProfile Context Propagation was made to make those Quarkus extensions work properly in reactive/async settings. It works by capturing those contextual values that used to be in thread-locals, and restoring them when your code is called.

完整源码

We recommend that you follow the instructions in the next sections and create the application step by step. However, you can go right to the completed example.

Clone the Git repository: git clone https://github.com/quarkusio/quarkus-quickstarts.git, or download an archive.

The solution is located in the context-propagation-quickstart directory.

Setting it up

If you are using Mutiny (the quarkus-mutiny extension), you just need to add the quarkus-smallrye-context-propagation extension to enable context propagation.

In other words, add the following dependencies to your build file:

pom.xml

<!-- RESTEasy Reactive extension if not already included -->
<dependency>
    <groupId>io.quarkus</groupId>
    <artifactId>quarkus-resteasy-reactive</artifactId>
</dependency>
<!-- Context Propagation extension -->
<dependency>
    <groupId>io.quarkus</groupId>
    <artifactId>quarkus-smallrye-context-propagation</artifactId>
</dependency>

build.gradle

// RESTEasy Reactive extension if not already included
implementation("io.quarkus:quarkus-resteasy-reactive")
// Context Propagation extension
implementation("io.quarkus:quarkus-smallrye-context-propagation")

With this, you will get context propagation for ArC, RESTEasy Reactive and transactions, if you are using them.

Usage example with Mutiny

Mutiny

This section uses Mutiny reactive types. If you are not familiar with Mutiny, check Mutiny - an intuitive reactive programming library.

Let’s write a REST endpoint that reads the next 3 items from a Kafka topic, stores them in a database using Hibernate ORM with Panache (all in the same transaction) before returning them to the client, you can do it like this:

    // Get the prices stream
    @Inject
    @Channel("prices") Publisher<Double> prices;

    @Transactional
    @GET
    @Path("/prices")
    @RestStreamElementType(MediaType.TEXT_PLAIN)
    public Publisher<Double> prices() {
        // get the next three prices from the price stream
        return Multi.createFrom().publisher(prices)
                .select().first(3)
                // The items are received from the event loop, so cannot use Hibernate ORM (classic)
                // Switch to a worker thread, the transaction will be propagated
                .emitOn(Infrastructure.getDefaultExecutor())
                .map(price -> {
                    // store each price before we send them
                    Price priceEntity = new Price();
                    priceEntity.value = price;
                    // here we are all in the same transaction
                    // thanks to context propagation
                    priceEntity.persist();
                    return price;
                    // the transaction is committed once the stream completes
                });
    }

Notice that thanks to Mutiny support for context propagation, this works out of the box. The 3 items are persisted using the same transaction and this transaction is committed when the stream completes.

Usage example for `CompletionStage`

If you are using CompletionStage you need manual context propagation. You can do that by injecting a ThreadContext or ManagedExecutor that will propagate every context. For example, here we use the Vert.x Web Client to get the list of Star Wars people, then store them in the database using Hibernate ORM with Panache (all in the same transaction) before returning them to the client as JSON using Jackson or JSON-B:

    @Inject ThreadContext threadContext;
    @Inject ManagedExecutor managedExecutor;
    @Inject Vertx vertx;

    @Transactional
    @GET
    @Path("/people")
    public CompletionStage<List<Person>> people() throws SystemException {
        // Create a REST client to the Star Wars API
        WebClient client = WebClient.create(vertx,
                         new WebClientOptions()
                          .setDefaultHost("swapi.dev")
                          .setDefaultPort(443)
                          .setSsl(true));
        // get the list of Star Wars people, with context capture
        return threadContext.withContextCapture(client.get("/api/people/").send())
                .thenApplyAsync(response -> {
                    JsonObject json = response.bodyAsJsonObject();
                    List<Person> persons = new ArrayList<>(json.getInteger("count"));
                    // Store them in the DB
                    // Note that we're still in the same transaction as the outer method
                    for (Object element : json.getJsonArray("results")) {
                        Person person = new Person();
                        person.name = ((JsonObject) element).getString("name");
                        person.persist();
                        persons.add(person);
                    }
                    return persons;
                }, managedExecutor);
    }

Using ThreadContext or ManagedExecutor you can wrap most useful functional types and CompletionStage in order to get context propagated.

The injected ManagedExecutor uses the Quarkus thread pool.

Overriding which contexts are propagated

By default, all available contexts are propagated. However, you can override this behaviour in several ways.

Using configuration

The following configuration properties allow you to specify the default sets of propagated contexts:

Configuration Key

Description

Default Value

mp.context.ThreadContext.propagated

The comma-separated set of propagated contexts

Remaining (all non-explicitly list contexts)

mp.context.ThreadContext.cleared

The comma-separated set of cleared contexts

None (no context), unless neither the propagated nor cleared sets contain Remaining, in which case the default is Remaining (all non-explicitly listed contexts)

mp.context.ThreadContext.unchanged

The comma-separated set of unchanged contexts

None (no context)

The following contexts are available in Quarkus either out of the box, or depending on whether you include their extensions:

Context Name

Name Constant

Description

None

ThreadContext.NONE

Can be used to specify an empty set of contexts, but setting the value to empty works too

Remaining

ThreadContext.ALL_REMAINING

All the contexts that are not explicitly listed in other sets

Transaction

ThreadContext.TRANSACTION

The JTA transaction context

CDI

ThreadContext.CDI

The CDI (ArC) context

Servlet

N/A

The servlet context

Jakarta REST

N/A

The RESTEasy Reactive or RESTEasy Classic context

Application

ThreadContext.APPLICATION

The current ThreadContextClassLoader

Overriding the propagated contexts using annotations

In order for automatic context propagation, such as Mutiny uses, to be overridden in specific methods, you can use the @CurrentThreadContext annotation:

    // Get the prices stream
    @Inject
    @Channel("prices") Publisher<Double> prices;

    @GET
    @Path("/prices")
    @RestStreamElementType(MediaType.TEXT_PLAIN)
    // Get rid of all context propagation, since we don't need it here
    @CurrentThreadContext(propagated = {}, unchanged = ThreadContext.ALL_REMAINING)
    public Publisher<Double> prices() {
        // get the next three prices from the price stream
        return Multi.createFrom().publisher(prices)
                .select().first(3);
    }

Overriding the propagated contexts using CDI injection

You can also inject a custom-built ThreadContext using the @ThreadContextConfig annotation on your injection point:

    // Get the prices stream
    @Inject
    @Channel("prices") Publisher<Double> prices;
    // Get a ThreadContext that doesn't propagate context
    @Inject
    @ThreadContextConfig(unchanged = ThreadContext.ALL_REMAINING)
    SmallRyeThreadContext threadContext;

    @GET
    @Path("/prices")
    @RestStreamElementType(MediaType.TEXT_PLAIN)
    public Publisher<Double> prices() {
        // Get rid of all context propagation, since we don't need it here
        try(CleanAutoCloseable ac = SmallRyeThreadContext.withThreadContext(threadContext)){
            // get the next three prices from the price stream
            return Multi.createFrom().publisher(prices)
                    .select().first(3);
        }
    }

Likewise, there is a similar way to inject a configured instance of ManagedExecutor using the @ManagedExecutorConfig annotation:

    // Custom ManagedExecutor with different async limit, queue and no propagation
    @Inject
    @ManagedExecutorConfig(maxAsync = 2, maxQueued = 3, cleared = ThreadContext.ALL_REMAINING)
    ManagedExecutor configuredCustomExecutor;

If you need to inject the same ManagedExecutor or ThreadContext into several places and share its capacity, you can name the instance with @NamedInstance annotation. @NamedInstance is a CDI qualifier and all injections of the same type and name will therefore share the same underlying instance. If you also need to customize your instance, you can do so using @ManagedExecutorConfig/ThreadContextConfig annotation on one of its injection points:

    // Custom configured ManagedExecutor with name
    @Inject
    @ManagedExecutorConfig(maxAsync = 2, maxQueued = 3, cleared = ThreadContext.ALL_REMAINING)
    @NamedInstance("myExecutor")
    ManagedExecutor sharedConfiguredExecutor;

    // Since this executor has the same name, it will be the same instance as above
    @Inject
    @NamedInstance("myExecutor")
    ManagedExecutor sameExecutor;

    // Custom ThreadContext with a name
    @Inject
    @ThreadContextConfig(unchanged = ThreadContext.ALL_REMAINING)
    @NamedInstance("myContext")
    ThreadContext sharedConfiguredThreadContext;

    // Given equal value of @NamedInstance, this ThreadContext will be the same as the above one
    @Inject
    @NamedInstance("myContext")
    ThreadContext sameContext;

Context Propagation for CDI

In terms of CDI, @RequestScoped, @ApplicationScoped and @Singleton beans get propagated and are available in other threads. @Dependent beans as well as any custom scoped beans cannot be automatically propagated via CDI Context Propagation.

@ApplicationScoped and @Singleton beans are always active scopes and as such are easy to deal with - context propagation tasks can work with those beans so long as the CDI container is running. However, @RequestScoped beans are a different story. They are only active for a short period of time which can be bound either to HTTP request or some other request/task when manually activated/deactivated. In this case user must be aware that once the original thread gets to an end of a request, it will terminate the context, calling @PreDestroy on those beans and then clearing them from the context. Subsequent attempts to access those beans from other threads can result in unexpected behaviour. It is therefore recommended to make sure all tasks using request scoped beans via context propagation are performed in such a manner that they don’t outlive the original request duration.

Due to the above described behavior, it is recommended to avoid using @PreDestroy on @RequestScoped beans when working with Context Propagation in CDI.

完整源码
Setting it up
Usage example with Mutiny
Usage example for CompletionStage
Overriding which contexts are propagated
Context Propagation for CDI