A provisioned Amazon Redshift cluster with five ra3.4xlarge nodes shows one node frequently at over 90% CPU while the other four are under 15% during normal operations. The cluster uses key distribution. The engineer wants to keep the same node count but balance load across nodes. What change will achieve this?

The correct answer is to change the distribution key to the table column that has the largest dimension. An uneven CPU utilization across nodes in Amazon Redshift, especially with key distribution, indicates data skew. The distribution key determines how data rows are distributed across compute nodes. If the chosen distribution key has highly skewed values (e.g., many rows share the same key value), those rows will all reside on the same node, leading to that node being overworked while others are underutilized. Changing the distribution key to a column with high cardinality (many unique values) and an even distribution will spread the data more evenly across all nodes, thus balancing the workload and CPU utilization. Changing the sort key or primary key primarily impacts query performance for filtering and joining, not data distribution across nodes. Upgrading node size increases overall capacity but doesn't resolve the underlying data skew issue; the larger node would still be overworked while others remain underutilized.