The　existing　coarse-grained　monitoring　mechanism　in　Spark　has　much　difficulty　in　analyzing　and　locating　the　bottleneck　during　the　execution　of　tasks.　Aiming　to　solve　this　problem,　we　divide　the　execution　process　of　Spark　tasks　into　several　subphases,　and　propose　a　fine-grained　subphases-based　monitoring　mechanism.　The　fine-grained　monitoring　mechanism　can　help　users　to　understand　the　detail　execution　status　of　Spark　tasks　and　be　beneficial　to　analyze　and　locate　the　performance　bottleneck　in　Spark.