Select count(*) on spider table gives wrong result

I have a cluster with a master node running spider engine + 3 data nodes. I created a table (test_table)sharded on the three data nodes, with one (or more) secondary index. Running a select count(*) where ... from test_table on the spider node I have a result different from the sum of the same query executed directly on the three data nodes. Removing the secondary index from the table the query gives the right result.