How to generate the optimal index combination for the index recommendation of opengauss under multiple targets?

Question

The index recommendation of the AI module in the opengauss document supports the introduction of the optimal index combination within the limit of the index space. However, the index recommendation code only seems to use the hill-climbing method. The hill-climbing method is a greedy algorithm. Each time, it only selects the one with the largest current profit and converges and local solutions. However, under the constraints of the two goals of index return and space combination, is the algorithm unable to find the optimal solution? How do we calculate the optimal solution in this case?

Answer 1

Monte Carlo Tree Search, which can effectively solve some problems with huge exploration space, can balance exploration and utilization, and find effective solutions.