$ the-wire · showcase
SPRING-AI FIXES AZURE AUTHENTICATION MESS, KAFKA CONSOLIDATES BUILD
By RepoJournal · Filed · About Spring · Composed from the cited sources · methodology
Spring AI patched a critical Azure OpenAI credential bug that was forcing unnecessary example.com requests through corporate proxies, while Spring Kafka overhauled its dependency management to centralize version control.
The Azure OpenAI authentication layer was acquiring bearer tokens through a live HTTPS request to example.com on every call [1], a wasteful hop that fails behind TLS-inspecting proxies requiring custom certificates. That fix ships alongside a broader Azure OpenAI polish that extracts refresh margins into named constants and cleans up token supplier logic [2]. On the foundry front, Spring AI is now properly propagating Microsoft Foundry decisions to the Azure SDK by setting the correct URL path mode [3], fixing a bug where enterprise proxy base-URLs would miss the Azure deployment path and api-version parameters. Spring Kafka meanwhile introduced gradle/libs.versions.toml to centralize dependency and plugin versions [4], eliminating manual interpolation of kafka-clients and kafka-server-common test-classifier coordinates. On the stability side, Soby Chacko landed three CI flakiness fixes across Kafka: PartitionResolverTests now awaits the send future before the latch to prevent producer metadata fetch time from consuming processing time [5], DefaultErrorHandlerNoSeeksBatchListenerTests stops only container1 instead of the entire registry to avoid a load-induced race [6], and transactional offset lag is now corrected when marker batches arrive separately [7]. Spring Integration tightened its CI environment handling to use OSS credentials for snapshot pulls and commercial Artifactory credentials only when needed [8], while also fixing a race in FileTests where empty files published before content write would cause the FileSplitter to emit only markers [9].
Action items
- → Upgrade to Spring AI with the example.com authentication fix if you use Azure OpenAI behind a TLS proxy spring-projects/spring-ai [immediate]
- → Review Spring Kafka's new gradle/libs.versions.toml if you extend or customize dependency builds spring-projects/spring-kafka [plan]
- → Cherry-pick Spring Kafka CI flakiness fixes to 4.1.x and 4.0.x branches spring-projects/spring-kafka [plan]
References
- [1] Avoid example.com request in Azure OpenAI authentication ↗ spring-projects/spring-ai
- [2] Polish Azure OpenAI bearer token supplier ↗ spring-projects/spring-ai
- [3] Propagate Microsoft Foundry to Azure URL path mode ↗ spring-projects/spring-ai
- [4] Use Gradle version catalog for dependencies ↗ spring-projects/spring-kafka
- [5] Fix CI flakiness in PartitionResolverTests ↗ spring-projects/spring-kafka
- [6] Fix CI flakiness in DefaultErrorHandlerNoSeeksBatchListenerTests ↗ spring-projects/spring-kafka
- [7] Fix tx offset lag when marker batch is separate (#4553) ↗ spring-projects/spring-kafka
- [8] Fix ci.yml for proper environment ↗ spring-projects/spring-integration
- [9] Fix race in the `FileTests.testFileSplitterFlow()` ↗ spring-projects/spring-integration
