In its initial implementation, the formula for the cost of an indexed nested loop join ignored the number of components the index has. This means that the plan enumerator cannot rely on the cost model to have enough resolution to distinguish between two INLJ plans using different indexes yet. The plan enumerator must use some heuristics to determine which index is best for now. Once this ticket is done, the plan enumerator should be able to avoid relying on this heuristic and instead the cost model to determine which index is best.