A Dataflow job aggregates and writes time-series metrics to Bigtable but writes are slow and many dashboard users need fresher data. Which two actions improve write latency and concurrency?

Increasing maxNumWorkers for the Dataflow job allows for more parallel processing of data, which can significantly improve write throughput and reduce latency, especially for aggregation tasks. Adding nodes to the Bigtable cluster directly increases its capacity for both storage and serving requests. This improves write performance by distributing the load across more Bigtable servers, allowing for higher ingestion rates and lower latency for individual writes. Running Dataflow in local execution mode is for testing and debugging on a single machine and would severely limit scalability and performance, making the problem worse. Flatten and CoGroupByKey are Dataflow transforms that restructure data. While they are useful for specific data processing needs, they do not directly address write latency or concurrency issues with the Bigtable sink or the Dataflow job's execution resources.