Live video streaming demands high user Quality of Experience (QoE) and requires significant computing power and bandwidth for video encoding and transmission. The standard adaptive live streaming approach encodes the source video content at multiple predefined bitrates, allowing users to select the most appropriate bitrate for optimal rate adaptation. The current methods for encoding ladder design primarily focus on optimizing user QoE. However, the relationship between the encoding ladder and user QoE is complex due to the dynamic nature of live streaming, where the number of users fluctuates, and each user experiences varying and evolving bandwidth conditions. On the other hand, resource consumption in live streaming significantly impacts operational costs, as it requires substantial resources for both video encoding and delivery. To address this challenge, we propose QRALadder, a QoE and resource consumption-aware approach to encoding ladder design. Our method simultaneously optimizes user QoE while considering the impact of the encoding ladder on resource consumption in live streaming, and it can dynamically select the most suitable encoding ladder based on the current system state. We use the Dueling Double Deep Q Networks (Dueling DDQN) algorithm to learn the optimal encoding ladder, maximizing overall performance by analyzing user QoE feedback and resource consumption about system states. We conduct comparative experiments using various types of video content and real-world bandwidth traces to validate our approach. The results show that QRALadder can achieve higher overall performance compared to baseline methods.