Agreement-based resource management is becoming the foundation of grid system. Since the grid resource management encompasses a wide range of different scenarios, the decision making for service level agreement is a complicated procedure. Traditional resource management can not meet the new requirements. In this paper, we analyze and categorize the situations of resource availabilities and requirements. Then, we modeling each category with fuzzy constraints and propose an algorithm for incremental decision making. We also propose a resource reallocation mechanism for reducing agreement violation. Its efficiency is shown by experiment results.