Approximate nearest neighbors search (ANNS) combined with attribute filtering is crucial for practical vector search tasks involving vector similarity and multi-attribute constraints. Existing methods focus only on discrete or continuous attributes, ignoring their joint constraints. Meanwhile, existing solutions for continuous attribute filtering directly combined with discrete attribute methods can lead to poor segment balance, which affects construction and query efficiency. To address these issues, we propose MAHI (multi-attribute hybrid index), an index framework that efficiently supports ANNS under hybrid attribute constraints. We use prefix trees to partition the original dataset based on discrete attributes. Within each generated block, we combine the KMeans algorithm with two different indexing methods to mitigate the data distribution unevenness caused by discrete attribute partitioning, thereby restoring the structure of continuous attributes. Experiments on large-scale datasets demonstrate that MAHI significantly outperforms existing state-of-the-art baseline algorithms in terms of recall, throughput, and indexing efficiency.