A parks kiosk stream hangs until users give up, and distributed traces show gateway timeouts on long foundation-model completions. What should the team prioritize?
Select an answer to reveal the explanation.
Short Explanation
When the kiosk stream hangs, you can use observability to compare long foundation-model completions with gateway timeouts. Align timeout and streaming settings so answers have time to reach visitors before the connection closes.
Full Explanation
The team should use observability, such as X-Ray traces, to compare gateway timeout settings with long foundation-model (FM) completion times, then align timeout and streaming configuration. The likely failure is a mismatch: the model takes longer to complete than the gateway allows, and traces expose that mismatch so engineers can tune timeouts or streaming rather than guess. Disabling tracing fails by concept because it removes the visibility needed to correlate hangs with gateway timeouts. Shortening only the kiosk screensaver while ignoring API timeout budgets fails because a cosmetic screen setting does not change how long the gateway waits for an FM response. Expanding kiosk capacity and preloading printed brochures while treating stalled streams as local demand problems also fails by concept: it does not check or correct the suspected gateway limits against long model completions. Exam caveat: when traces show gateway timeouts during long FM calls, prioritize timeout and streaming configuration—not disabling observability or substituting a local-demand response. Practical check: compare p95/p99 FM completion durations with API gateway and load-balancer idle timeouts, verify token streaming to the kiosk, adjust timeout budgets with safety margins, load-test long answers before festival weekends, and retain X-Ray or equivalent correlation IDs during peak use.