There are a lot of aspect that could go wrong. you can check the profile, look under common metrics for each plan_node_id to determine what went wrong. some common issues are network latency, slow lock (if there's another query on the table), too many tablets