Apicurio Blog
Beyond OpenShift: Making Apicurio Registry Work with Any OAuth Provider via Dex ↗
Connecting identity providers through an OIDC bridge, and the practical differences between authentication, identity and authorization.
Writing
Full articles on distributed systems, data contracts and open-source infrastructure, connected to the code and tests behind them.
Browse Distributed Deep Dives ↗Elsewhere / Apicurio Blog
Apicurio Blog
Connecting identity providers through an OIDC bridge, and the practical differences between authentication, identity and authorization.
Apicurio Blog
Following schema-related requests through an event-driven application with distributed tracing.
Apicurio Blog
The project's move into the CNCF Sandbox and the work behind its open-source stewardship.
On this site
Seven real CDC events exercise rebuilding, ordered replay, deletion, alias cutover and a write after the switch.
Distinguish waiting on an author from waiting on a maintainer, using Apicurio's published lifecycle and a real review.
What mixed-batch rejection, direct broker access and Registry outages mean for a Kafka producer's failure handling.
Real token and authorization checks across Keycloak, Kafka and Registry, including the negative control that exposed disabled authentication.
The public deferral, renewed application and onboarding records behind Apicurio Registry's Sandbox journey.
A connected Quarkus workflow separates endpoint discovery, Agent Card publication and runtime behavior.
Follow an event through capture, consumption, an Iceberg snapshot and a query instead of equating connector health with progress.
A measured Spark experiment shows why removing processing delay can leave a source-admission bottleneck.
Use model disagreement to guide investigation without confusing agreement, confidence or throughput with accuracy.
The same deterministic function through official A2A and MCP SDKs: compare discovery, errors, progress and cancellation.
Pair a real Avro compatibility rejection with a units-normalization test at the inference-input boundary.
Version selection, cache refresh and a real Ollama evaluation that blocked every candidate instead of weakening the acceptance criteria.
How Avro named types, registry references and version selection interact—and which boundaries need separate tests.