Computational thinking is a way of thinking that helps people "think like a computer scientist" to solve practical problems. However, practicing computational thinking through programming is dependent on the problem solvers' metacognition. This study investigated students' metacognitive planning and problem-solving performance in programming through two quantitative studies. First, we analyzed the performance of metacognitive planning and of problem solving through the programming of 21 freshmen, and found that the metacognitive planning performance related to "problem description" and "program comprehension" was significantly correlated with problem-solving performance. Second, semi-scaffolding and full-scaffolding were designed based on the first study. Another 89 freshmen were randomly divided into three groups and were asked to write their programming plan with no-scaffolding, semi-scaffolding, or with full-scaffolding. ANCOVA revealed that the problem-solving performance of the no-scaffolding group was significantly weaker than that of the other two groups, but there was no significant difference between the semi-scaffolding and the full-scaffolding groups. The study indicated that semi-scaffolding had a similar effect to full-scaffolding on problem-solving performance. The study suggests that teachers should emphasize supporting students' "problem description" and "program comprehension" using semi-scaffolding. This scaffolding technique is sufficient and efficient for training students' computational thinking through problem solving in programming.