Mongos routers will, under certain circumstances, restart interrupted queries when an election occurs in a shard.

When the election was triggered by an event (such as OOM) caused by workload, the workload causing the failure may be restarted, causing cascading failures through the members of a shard.