Metadata connection (broker -1) receive timeout causes test hangs

[thomhurst]

Summary

Integration tests intermittently hang when the metadata/bootstrap connection (broker -1) times out with Receive timeout after 30000ms. This prevents the producer from resolving partition leaders for newly created topics, causing tests to hang until the 5-minute test timeout.

Symptoms

• [Error] Dekaf.Networking.KafkaConnection: Receive timeout after 30000ms on broker -1 - marking connection as failed
• [Information] Dekaf.Networking.ConnectionPool: All connections closed appearing during test setup
• Tests that create multi-partition topics and produce to them hang indefinitely
• Affected tests: MultiPartition_SeekOnSpecificPartition_WorksCorrectly, MultiPartition_DifferentKeys_DistributeAcrossPartitions, others in the Messaging category

Context

Observed in CI runs on fix/flaky-tests-round2 branch after the producer batch lifecycle bugs were fixed. The "No response for partition" topic mismatch is now resolved — this is a separate issue.

Possible causes

• Bootstrap connection not being re-established after failure
• Connection pool disposal from a parallel test affecting metadata lookups
• Kafka container under load from parallel tests causing slow metadata responses
• Missing retry/reconnect logic for broker -1 metadata connections

CI references

• Run 23491170091 (commit 720e76d): 1/51 failed — MultiPartition_SeekOnSpecificPartition_WorksCorrectly
• Run 23490015213 (commit 0030eca): 1/51 failed — MultiPartition_DifferentKeys_DistributeAcrossPartitions