A 3-minute live TV voting spike must handle high write load, show partial results while open, and produce an exact once final count while minimizing cost. Which architecture meets these requirements?

This architecture is optimal because Pub/Sub handles the high ingress write load, decoupling the voting application from downstream processing. Dataflow provides flexible, scalable processing to load data into Bigtable for low-latency, real-time partial results during the voting window. BigQuery is used for cost-effective, exact-once final counts and later analysis. Shutting down Bigtable after voting minimizes cost, as it's only needed for real-time results. Memorystore is an in-memory cache, not ideal for persistent, exact-once counting. Cloud SQL is a relational database, which struggles with the high write throughput of a voting spike and is less cost-effective for real-time analytics at scale compared to Bigtable. Using Cloud Functions to write directly to BigQuery might not handle the sustained high write volume efficiently and could lead to higher costs and potential rate limiting issues compared to a Dataflow pipeline.