Aiming　to　the　new　features　of　data　processing　model　on　Spark,　A　fine-grained　task　monitoring　system　for　Spark,　called　FTMS,　is　proposed　in　this　paper.　Compared　to　the　original　monitoring　system　on　Spark,　FTMS　can　provide　phase-level　fine-grained　task　status　information　and　make　some　statistics　on　the　raw　task　information　to　help　users　decide　the　potential　performance　bottleneck　in　the　spark　application　execution.　Three　key　issues　of　FTMS,　including　the　fine-grained　task　information　model,　the　monitoring　information　collection　mechanism　and　the　information　statistic　mechanism,　are　described　in　detail　in　this　paper.　The　performance　evaluation　result　shows　that　FTMS　can　provide　the　more　complete　information　of　the　running　task　with　the　low　performance　overhead.　With　FTMS,　the　application　turnaround　time　is　only　increased　by　4.1%　by　average.