[2748109647]

Summary

• Add global and per-actor reentrancy configuration (enabled and optional maxStackDepth) to HTTP actor discovery and gRPC registration.
• Propagate Dapr-Reentrancy-Id through Reactor context for asynchronous calls, and use a scoped ThreadLocal while invoking synchronous actor methods so nested .block() calls retain the reentrancy ID. The previous value is restored when the invocation scope exits.
• Add configuration, serialization, propagation, concurrency, and HTTP/gRPC A → B → A coverage for both asynchronous and synchronous actor methods, plus usage documentation.

Validation

• mvn -B -pl sdk-actors test passed: 233 tests, 0 failures or errors.
• mvn -B -pl sdk-actors -am checkstyle:check passed with 0 violations.
• Compiled all sdk-tests integration test sources, including ActorReentrancyIT.
• The container-backed ActorReentrancyIT was not run locally because Docker is unavailable in this environment.

Fixes #1838

[javier-aliaga]

@2748109647 thanks for the PR. I don't want to ship reentrancy half done, and right now the sync path doesn't work. An actor method that calls another actor with .block() never sees the Reactor context, so the call deadlocks. Our own DemoActorImpl example writes methods that way.

Could you cover the sync path in this PR as well? One option is to also keep the id in a ThreadLocal around the method invocation. It needs a sync A -> B -> A test next to the async one.

[2748109647]

Thanks for the suggestion. I added synchronous .block() A → B → A coverage alongside the asynchronous cases for both HTTP and gRPC hosts.

The nested call can re-enter the same actor instance while its outer invocation is still active. I changed the actor lifecycle guard from a boolean to an active-call counter, preserve the outer invocation's cached state during nested calls, and decrement the counter on completion (including error/cancel paths). The existing scoped ThreadLocal reentrancy ID is exercised by the synchronous tests.

While running the end-to-end test, I also found that the test interface's Mono<String> methods need @ActorMethod(returns = String.class); without it the proxy uses the void-returning overload, so the test cannot observe the result. I added that annotation to the test API.

Validation:

• mvn -B -ntp -pl sdk-actors test — 233 passed, 0 failures/errors; Checkstyle 0 violations; SpotBugs 0 findings.
• mvn -B -ntp -Pintegration-tests -pl sdk-tests -Dit.test=ActorReentrancyIT -Dfailsafe.failIfNoSpecifiedTests=false -DskipITs=false -Dspotbugs.skip=true -Dmaven.javadoc.skip=true -Dmaven.source.skip=true test-compile failsafe:integration-test failsafe:verify — 4 passed, 0 failures/errors, using local Docker/Testcontainers.
• mvn -B -ntp -Pintegration-tests -pl sdk-tests checkstyle:check — 0 violations.
• git diff --check — passed.

On Windows I temporarily adjusted the existing local AppRun shell quoting so the gRPC app with an empty argument could start; I reverted that test-harness workaround and did not include it in the PR.