Failing test: global partitioning + Separated mode

[jeremydmiller]

Summary

• Adds a failing integration test demonstrating that global partitioning combined with MultipleHandlerBehavior.Separated throws NoHandlerForEndpointException
• The test sets up two global partitions for the same message types, with two separate handlers per message type and a cascading message flow
• Related to Add global partitioning for cluster-wide GroupId concurrency control #2273 (global partitioning feature)

Root Cause Analysis

When Separated mode is active, all handlers are moved to sticky ByEndpoint chains bound to handler-type-specific local queues (e.g., local://partitionhandlerone/). Global partitioning routes messages to companion local queues (e.g., local://partition-a2/).

In HandlerGraph.HandlerFor(Type, Endpoint) (line 196), the fanout handler logic has this guard:

if (allLocal && endpoint is not LocalQueue)
{
    return getOrBuildFanoutHandler(messageType, chain);
}

throw new NoHandlerForEndpointException(messageType, endpoint.Uri);

The fanout handler only activates when the incoming endpoint is not a LocalQueue (i.e., an external transport endpoint). But global partition companion queues ARE LocalQueue instances, so the fanout path is skipped and the exception is thrown.

Error Output

NoHandlerForEndpointException: No handlers for message type PartitionedCommand 
at endpoint local://partition-a2/

Test plan

• New test correctly fails demonstrating the bug
• All 1160 existing CoreTests still pass

🤖 Generated with Claude Code