<p>With the rapid development of knowledge graphs, there is a growing need for efficient storage solutions that enable fast querying and online updates. Mature relational storage struggles with highly selective queries, while graph storage that addresses this challenge suffers from inefficient data insertion and typically has storage constraints. Consequently, existing dual storage structures combining both storage show promise but require further improved query performance and lack online update support. We present EKGStore, a novel knowledge graph storage framework that addresses these challenges from three aspects. First, we develop an adaptive query processor that optimizes query decomposition across dual storage structures. Second, we propose a load-aware tuning algorithm based on multi-armed bandit to realize the periodic dynamic tuning of dual storage structures. Third, we implement an eventual consistency dual-write strategy that supports online updates, and ensures correct queries through a buffering mechanism. Experimental results show that for query-only workloads, EKGStore improves query performance by up to 39% compared to the existing dual storage structures, and it still maintains good query performance with a large number of updates.</p>