fix: Make test_kbucket_split_behavior more reliable

[acul71]

Problem

The test test_kbucket_split_behavior in tests/core/kad_dht/test_unit_routing_table.py was failing intermittently in CI/CD, as reported in discussion #1023 and issue #1032.

Root Cause

The mock_host fixture in the TestRoutingTable class didn't properly mock host.new_stream(). When a bucket is full, the KBucket.add_peer() method tries to ping the oldest peer using _ping_peer(), which calls host.new_stream(). Without proper mocking, the behavior was unpredictable, causing intermittent test failures.

Solution

Mock host.new_stream() in the TestRoutingTable.mock_host fixture to raise an exception, simulating unresponsive peers. This makes the test deterministic and matches real production behavior where new_stream() fails for unreachable peers.

Changes

• Added host.new_stream = AsyncMock(side_effect=Exception("Mock stream error")) to the mock_host fixture
• This ensures ping operations fail consistently, allowing peers to be added when buckets are full
• Matches real production behavior where unresponsive peers are replaced

Testing

• ✅ Test passes consistently (verified with multiple runs)
• ✅ All other tests in the file still pass (16/16)
• ✅ No linting errors

Related

• Fixes Fix flaky test: test_kbucket_split_behavior in kad_dht routing table #1032
• Related to discussion CI/CD issues: unrelated to PRs but propping up consistently #1023

[acul71]

Hi @seetadev and @sumanjeet0012,

I've investigated the flaky test issue you mentioned in discussion #1023 regarding test_kbucket_split_behavior failing intermittently in CI/CD.

Root Cause

The test was flaky because the mock_host fixture in TestRoutingTable didn't properly mock host.new_stream(). When a bucket is full, the code tries to ping the oldest peer, and without proper mocking, the behavior was unpredictable.

Solution

I've created a fix that mocks new_stream() to raise an exception, simulating unresponsive peers. This:

• Makes the test deterministic and reliable
• Matches real production behavior where new_stream() fails for unreachable peers
• Allows the routing table to correctly replace unresponsive peers (as per Kademlia protocol)

PR Created

I've opened PR #1033 with the fix. The test now passes consistently, and all other tests remain unaffected.

Could you please review when you have a chance? The fix is minimal and focused on making the test more reliable.

Thanks!

[seetadev]

Excellent work, @acul71 👏

This is a clean, well-reasoned fix to one of the more persistent CI/CD issues we’ve been seeing with test_kbucket_split_behavior. Your analysis of the root cause — particularly the insight that mock_host.new_stream() wasn’t properly mocked and was introducing nondeterministic behavior — is spot on.

The solution of explicitly mocking new_stream() to raise an exception makes perfect sense. It brings the test behavior closer to real-world scenarios, where peers can be unreachable and ping attempts fail. This approach not only stabilizes the test but also improves the accuracy of how the routing table logic is validated against actual Kademlia behavior.

It’s great to see that you verified consistency across multiple runs and ensured that all other tests remained unaffected.

Thanks for taking the time to dig into the underlying issue and provide a production-aligned fix. 🚀

Looks ready for merge.