We noticed issues traffic was send to dead secondaries >60s after they died. I think the issue here is - 

There is no timeout set in the heartbeat loop for op.Execute. So it defaults to OperationTimeout set at the topology. If OperationTimeout is set to 60s, then it will take about 70-80s case for the heartbeat to finish with heatbeatTimeout and heartbeatInterval set to 10s.

We can use context with heartbeat timeout for the IsMaster command to finish the heartbeat faster.