com.netflix.server.context

Annotation Type BiDirectional

@Retention(value=RUNTIME)
@Target(value=TYPE)
public @interface BiDirectional

Any contexts added to ContextsContainer.addContext(String, Object, ContextSerializer) is unidirectional i.e. it always flows down the call chain, eg: In the below call graph

        Service A ---> Service B ---> Service C
                                |
                                 ---> Service D

 ------------------ Context X Flows through the call graph ----------------->
 

Any changes to Context "X" (which gets added by Service A) in Service B will only be available to Service C & D but not service A.
This interface addresses the use-case where the update in context X must be available to Service A.
By annotating their class with this annotation, the context implementers can guarantee that the context changes, flow bi-directionally in the call graph i.e. any code invoked after the change will get the updated context.
These updates will be effective iff, the following steps are taken:

• Whenever the context gets updated, update the same with ContextsContainer by calling ContextsContainer.addContext(String, Object, ContextSerializer). Implementation may batch these updates if such a batch can be made reliably.
• Before returning a response from the service, get the updated bidirectional contexts via ContextsContainer.getModifiedBidirectionalContexts() and add it as response headers.
• On receiving a response, read & update bidirectional contexts by calling ContextsContainer.consumeBidirectionalContextsFromResponse(ContextKeySupplier)

NIWS infrastructure does the last two steps transparently. If you (or your dependencies) are not using NIWS you may have to do either one or both of these yourselves based on whether you are using client side, server side or none of the NIWS components.

Mergeable contexts

The infrastructure always over-writes the existing context when writing back a bi-directional context to the current thread. In the above example, when service B returns, it reads back the bi-directional context & update the context in Service A, at this point the default behavior is to over-write the existing context. However, in some cases, it may be desirable to merge the two contexts. If so, is the case with your context, you must implement the interface MergeableContext to indicate that the old context must be merged into the new context.
For such a context, the infrastructure will call MergeableContext.merge(MergeableContext) on the old context with the new context as the argument.

New Contexts added in Child service

Any bi-directional contexts added fresh in a child service also flows back in the response.