Skip to main content

Multitenancy With the Thread Context

What Is a Thread Context?

The SAP Cloud SDK for Java provides a so-called ThreadContext. It serves as thread-safe storage for potentially sensitive information. Specifically, the following three objects are stored:

  • The current Tenant
  • The current Principal (User)
  • The JSON Web Token (JWT)

This information is used throughout the SAP Cloud SDK to provide features like tenant and principal isolation, JWT verification and authorization against other systems and services. To ensure different tenants and users are properly isolated in an application, this information is always limited to the thread it was created on unless it is explicitly passed on by the application (see Propagating the Thread Context).

info

Multi-tenancy describes the access of different, technically separated user groups to the same instance of an application. In the terms of XSUAA service, these user groups are called Tenants, while in terms of Identity Authentication Service (IAS) they are called Zones.

The SAP Cloud SDK for Java uses the term Tenant to refer to both XSUAA Tenants and IAS Zones. This implies, that the tenantId exposed in the Tenant interface either returns the tenantId or zoneId, depending on the context you are currently running in.

How Is a Thread Context Created?

The SAP Cloud SDK provides a RequestFilter that will listen to incoming HTTP requests. This filter will extract all provided HTTP headers from the incoming request and store them in the ThreadContext. Additionally, if the Authorization header contains a JWT from the AppRouter, the filter will:

  • Verify this token
  • Store it in the ThreadContext and
  • Pull the Tenant and Principal information from it
Integration with CAP

In case you are using CAP (with the cds-integration-cloud-sdk dependency) the Tenant, Principal, and Headers will instead be derived from the CAP Request Context. That also means that the ThreadContext will not be used when accessing this information.

Running Asynchronous Operations

As the name suggests, the ThreadContext is bound to a thread, more specifically to the one it was created in. In consequence, other threads, i.e. asynchronous operations, cannot access the stored information if it hasn't been explicitly propagated.

The SAP Cloud SDK offers a dedicated API to execute asynchronous operations while preserving the ThreadContext.

Given any operation:

Callable myTask = () -> doStuff();
// or
Runnable myTask = () -> doStuff();

The operation can be executed asynchronously via:

Future runningTask = ThreadContextExecutors.submit(myTask);

Operations that are submitted this way will be executed by a single instance of ThreadContextExecutorService. By default, this instance is based on a CachedThreadPool, which manages the concurrency internally.

Spring Integration

You can conveniently integrate this functionality to work with Springs @Async annotation.

The configuration below will ensure all methods annotated with @Async will have the ThreadContext available:

@EnableAsync
@Configuration
public class AsynchronousConfiguration implements AsyncConfigurer {

@Override
public Executor getAsyncExecutor() {
return ThreadContextExecutors.getExecutor();
}
}

You can read more about the @Async functionality here.

Security Context

The Spring SecurityContext can be propagated to @Async calls. Replace the above executor with this one:

new DelegatingSecurityContextExecutor(ThreadContextExecutors.getExecutor());

And add this dependency:

<dependency>
<groupId>org.springframework.security</groupId>
<artifactId>spring-security-core</artifactId>
</dependency>

Passing on Other ThreadLocals

Other libraries and frameworks may also rely on ThreadLocal variables to hold contextual data. For example, the CAP RequestContext is also bound to the current Thread.

To not lose this state when creating a new Thread they can be passed on by implementing a ThreadContextDecorator. A typical implementation for this purpose should look like this:

@Override
public <T> Callable<T> decorateCallable( @Nonnull final Callable<T> callable ) {
Object valueToPass = YourThreadLocal.get();

return () -> {
Object initial = YourThreadLocal.get();
YourThreadLocal.set(valueToPass);
try {
return callable.call();
}
finally {
YourThreadLocal.set(initial);
}
};
}

Such a custom ThreadContextDecorator can be registered via DefaultThreadContextDecoratorChain.registerDefaultDecorator.

tip

The SAP Cloud SDK already comes with a default decorator that passes on the SecurityContext of com.sap.cloud.security:java-api. If the CAP integration is used, also the RequestContext is passed on (requires CDS version 1.28 or higher).

Configuring the Executor

The ThreadContextExecutors class leverages a single ThreadContextExecutorService instance that can be configured.

You can create a custom ThreadContextExecutorService, for example to use a different thread pool, via:

ThreadContextExecutorService executor = DefaultThreadContextExecutorService
.of(Executors.newFixedThreadPool(3));

// use it directly:
executor.submit(myTask);

// or set it to be used by the static ThreadContextExecutors API:
ThreadContextExecutors.setExecutor(executor);
ThreadContextExecutors.submit(myTask);

Modifying the ThreadContext

Caution

Modifying the ThreadContext as depicted below is currently supported only when using the default Facade (e.g. DefaultTenantFacade) implementations. This is an issue especially when using the CAP integration dependency cds-integration-cloud-sdk. To still manipulate the request context, please refer to the CAP documentation.

You may want to run an asynchronous operation in a completely new or custom context. By default, the executor will transfer any existing context to the new thread.

To run something in a completely new, empty context, use:

ThreadContextExecutors.submit(myTask, new DefaultThreadContext());

By contrast, to pass on the current context, but modifying the tenant (and only the tenant), use:

Tenant myTenant = new DefaultTenant("foo");
Callable myTaskWithTenant = () -> TenantAccessor.executeWithTenant(myTask, myTenant);

ThreadContextExecutors.submit(myTaskWithTenant);

To avoid multiple wrappings, in particular when changing multiple values, you can also build a custom executor:

Tenant myTenant = new DefaultTenant("foo");
Principal myPrincipal = new DefaultPrincipal("bar", Collections.emptySet(), Collections.emptyMap());

ThreadContextExecutor customExecutor = ThreadContextExecutor.fromNewContext()
.withListeners(
new TenantThreadContextListener(myTenant),
new PrincipalThreadContextListener(myPrincipal)
);

ThreadContextExecutors.submit(myTask, customExecutor);

How Can the Thread Context Be Used?

Accessing Information

The Thread context can be accessed via the static ThreadContextAccessor.

For the frequently needed HTTP Headers, Tenant, Principal, and JWT there are also dedicated accessors:

Storing Information

The ThreadContext allows for some manipulation by the application. However, oftentimes it is more convenient to leverage the executeWith...() functionality offered by the dedicated accessors.

Consider a scenario where some part of the code should run on behalf of a specific tenant. In that case you can override the current tenant explicitly:

TenantAccessor.executeWithTenant(customTenant, () -> doStuff());
CAP Integration

The above does not apply for accessors like TenantAccessor, PrincipalAccessor, RequestHeaderAccessor when using the CAP framework (with the cds-integration-cloud-sdk dependency). When using CAP the Tenant, Principal, and Headers are derived from the RequestContext. Please refer to this section on how to override existing values in the CAP context.

caution

Be aware, that the executeWith methods shown above only replaces the given property, but does not update properties derived from it.

Example: You have a special AuthToken, that you propagate with AuthTokenAccessor.executeWithAuthToken. Inside the given code block only the AuthToken will be replaced, while e.g. the Tenant is the same as in the original context.

If you want to start a fresh context based on a given AuthToken, for example accessing information of the provider tenant while in a subscriber context, have a look at this code:


Tenant retrieveProviderTenant()
{
// retrieves an access token from the provider context
AuthToken providerXsuaaAccessToken = new ScpCfAuthTokenFacade().tryGetXsuaaServiceToken().get();
ThreadContextExecutor
// create a new, empty context
.fromNewContext()
// add the provider token into the new context
.withListeners(new AuthTokenThreadContextListener(providerXsuaaAccessToken))
// return the actual provider tenant
.execute(TenantAccessor::getCurrentTenant);
}

Manipulating HTTP Headers

The RequestHeaderAccessor#getHeaderContainer() method provides convenient access to the HTTP headers of the current incoming request. It returns an instance of RequestHeaderContainer, which is, by API contract, an immutable container that allows case insensitive access to HTTP header values. Although the RequestHeaderContainer cannot be altered once created, there are scenarios where manipulating HTTP headers is required. In such cases, a new instance of RequestHeaderContainer can be created in a few different ways.

Create a new RequestHeaderContainer From Scratch

A common way to represent HTTP headers is to use Map<String, String> for 1:1 relationships between header names and values or even Map<String, Collection<String>> for 1:n relationships. To make the transition from either of those representations to the SAP Cloud SDK's RequestHeaderContainer as easy as possible, the DefaultRequestHeaderContainer offers corresponding factory methods:

  • fromSingleValueMap(Map<String, String>) and
  • fromMultiValueMap(Map<String, ? extends Iterable<String>>)

The latter one also enables convenient interoperability with the Spring Headers:

@RequestMapping( method = RequestMethod.GET )
public ResponseEntity<ExampleResponse> getExample( @RequestHeader final HttpHeaders headers)
{
final RequestHeaderContainer container = DefaultRequestHeaderContainer.fromMultiValueMap(headers);

RequestHeaderAccessor.executeWithHeaderContainer(container, () -> {
Tenant tenant = TenantAccessor.getCurrentTenant();
});
}

Besides these two factory methods, the DefaultRequestHeaderContainer also offers the possibility to create an instance of RequestHeaderContainer.Builder through the static builder() method. An example for how the returned Builder can be used is shown in the chapter below.

Additionally, to make customizing the current HTTP headers even easier, the RequestHeaderAccessor comes with an overload of the executeWithHeaderContainer method that accepts a Map<String, String>.

Updating an Existing RequestHeaderContainer

As pointed out above, the RequestHeaderContainer is an immutable structure. Therefore, updating an already existing instance is impossible.

However, in cases where you would like to, for example, just add a new custom header to the set of existing headers, the RequestHeaderContainer offers the toBuilder method. As the name suggests, this method can be used to retrieve an instance of RequestHeaderContainer.Builder. In contrast to the RequestHeaderContainer, the Builder can be mutated as much as needed. Additionally, the toBuilder method will make sure that the returned Builder instance is already pre-filled with all HTTP headers that are also present in the instance of RequestHeaderContainer.

To make things less theoretical, let's examine an example.

Example: Usage of the Builder

Assume your application received an HTTP request that contains the following headers

  • Authorization: Bearer DUMMY_JWT
  • Set-Cookie: cookie-1; cookie-2
  • Accept-Language: en-US
  • x-app-specific-header: customer-value

These values can be accessed as follows:

RequestHeaderContainer headers = RequestHeaderAccessor.getHeaderContainer();
headers.getHeaderValues("authorization"); // will return ["Bearer DUMMY_JWT"]
headers.getHeaderValues("set-cookie"); // will return ["cookie-1", "cookie-2"]
headers.getHeaderValues("accept-language"); // will return ["en-US"]
headers.getHeaderValues("x-app-specific-header"); // will return ["customer-value"]

Note how accessing the values for specific HTTP headers will work independent of the casing of the provided name.

Now let's say your use case requires that HTTP cookies shall not be leaked into further application execution. Additionally, you have to make sure that the x-app-specific-header contains an additional application provided value. Lastly, our application should always serve German customers and, therefore, you need to make sure the Accept-Language header is always adjusted accordingly.

Using the Builder API, fulfilling these requirements is straightforward:

RequestHeaderContainer updatedHeaders =
headers
.toBuilder()
.withoutHeader("set-cookie")
.withHeader("x-app-specific-header", "application-value")
.replaceHeader("accept-language", "de-DE")
.build();

Once again, the API guarantees that header names are treated case insensitively.

Finally, to make sure the updated headers are also taken into consideration, you have to overwrite the existing headers in our ThreadContext. This can be done using the following code:

RequestHeaderAccessor.executeWithHeaderContainer(updatedHeaders, () -> yourBusinessLogic());