This browser is no longer supported.
Upgrade to Microsoft Edge to take advantage of the latest features, security updates, and technical support.
Your A2A discovery registry becomes unavailable for 30 seconds during a network partition. Agent requests that depend on registry lookups fail. Which design best improves registry resilience?
Implement retry logic in each agent client that retries registry calls every 5 seconds for up to 2 minutes.
Cache a point-in-time snapshot of the registry in each routing service with a Time-to-Live (TTL), and fall back to the snapshot when the live registry is unreachable.
Move all agent endpoint URLs to environment variables in each consuming agent so agents can discover each other without the registry.
Two agents concurrently read version 5 of a shared research document and both submit updates—agent A adds equity analysis, agent B adds risk metrics. Both attempt to write as version 6 with their respective additions. What happens with optimistic concurrency control, and what should the system do?
Both writes succeed, and the last write wins based on timestamp ordering.
The second write fails with an ETag mismatch; the losing agent reads the updated document (now containing agent A's equity analysis), applies its own risk metrics on top, and retries as version 7.
The second write fails; the losing agent logs an error and aborts its update to prevent a conflict cascade.
Your market analysis agent returns a bullish outlook on a stock, while the risk agent returns a high-risk warning for the same stock citing similar data. A simple priority-based resolution gives the analysis agent's output precedence. What additional step should the system include for Contoso Capital's regulatory compliance requirements?
Silently discard the risk agent's conflicting output since the higher-priority agent's result was selected.
Log the full conflict record—both agent outputs, which agents produced them, which resolution pattern was applied, the final result, and a timestamp—before proceeding with the resolved output.
Escalate all conflicts involving risk warnings to a human analyst before proceeding, regardless of the priority resolution outcome.
You must answer all questions before checking your work.
Was this page helpful?
Need help with this topic?
Want to try using Ask Learn to clarify or guide you through this topic?