The integration tests had timing issues causing intermittent failures on CI:
- TestNATSEventBus_HighThroughput: Added subscriber readiness synchronization using a barrier event before bulk publishing. This ensures the NATS subscription is fully established before events are sent rapidly. Extended timeout from 30s to 60s for CI environments.
- TestNATSEventBus_EventOrdering: Added readiness barrier event to synchronize subscriber setup before publishing ordered events. Extended timeout from 10s to 15s to account for CI timing variations.
- TestNATSEventBus_ConcurrentPublishSubscribe: Added readiness synchronization before concurrent publishers start. Extended timeout from 10s to 30s to handle the increased load under CI constraints.
Root causes:
- Subscriber channels were not fully ready to receive when bulk publishing started, causing message loss
- CI runners (especially ARM64) have different timing characteristics than local development
- Insufficient timeouts for high-volume event collection under shared CI resources
The fixes use a barrier pattern: publish a ready signal, wait to receive it, then proceed with the test. This is more reliable than fixed sleep durations.
Closes#57
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
