[Bug] Topic candidate selection races with concurrent cache eviction

[Aias00]

Runtime platform environment

All platforms; reproduced by a deterministic unit test on the current develop branch.

RocketMQ version

develop at e3458616d207ee636b1762f0f8dcf788a590d59d.

Describe the Bug

Both the Java producer and Proxy availability detectors select a cached topic by first checking whether a concurrent map is empty and then creating a new iterator and immediately calling next().

If the final cache entry is evicted between the two operations, next() throws NoSuchElementException. The client path is DefaultMQProducerImpl.pickTopic; the Proxy has the same pattern in TopicRouteService.pickTopic.

Steps to Reproduce

1. Populate the producer topic cache with one entry.
2. Simulate eviction immediately after isEmpty() returns false.
3. Invoke pickTopic through the availability detector helper.
4. Observe NoSuchElementException.

Expected Behavior

A cache that becomes empty during candidate selection returns an empty Optional, causing the detector to report no candidate rather than throwing.

Actual Behavior

Candidate selection throws before the detector can return a normal unavailable result.

Proposed Fix

Use one iterator and check hasNext() before calling next() in both Client and Proxy implementations. Add a regression test for the Client implementation.

[RockteMQ-AI]

Issue Evaluation

Category: bug | Status: Needs Verification

Thank you for reporting this issue. To help verify and fix it, please provide:

• Version/tag where this occurs
• Steps to reproduce
• Expected vs actual behavior
• Relevant logs or stack traces

---

Automated evaluation by github-manager-bot

[Aias00]

Verified on develop at e3458616d207ee636b1762f0f8dcf788a590d59d (before the proposed fix).

Steps to reproduce

1. Populate the topic cache with exactly one entry.
2. Remove that entry after the original isEmpty() check but before the subsequent iterator().next() call.
3. Invoke candidate-topic selection. The new deterministic regression tests model this with a ConcurrentHashMap that clears itself during isEmpty().

Expected behavior

Candidate selection should return an empty result when no entry is available, or a weakly-consistent candidate already observed by the iterator. It must not throw.

Actual behavior

The previous check-then-iterate sequence calls iterator().next() on an empty key set and throws java.util.NoSuchElementException. This can occur while a route/cache entry is concurrently evicted.

Evidence

Before the fix, DefaultMQProducerImplTest#testPickTopicToleratesConcurrentCacheEviction fails deterministically with InvocationTargetException caused by NoSuchElementException from DefaultMQProducerImpl.pickTopic. The same sequence existed in TopicRouteService.pickTopic.

PR #10893 replaces both sequences with a single iterator plus hasNext() check and adds directed tests for the Java producer and Proxy route service. Local verification with JDK 8 passed:

• DefaultMQProducerImplTest: 35 tests
• ClusterTopicRouteServiceTest + LocalTopicRouteServiceTest: 7 tests