We noticed when adding sharding availability tests that in Atlas mongos connection pools for a host have a max limit of 250, but this limit is separate for connections to hosts in the config server and is raised to the integer maximum. This seemingly was done just with dedicated config servers in mind, but now that embedded config servers are on by default in Atlas clusters, this also applies to them. This was probably unintentional and should be changed to be consistent (ie use 250 for config shards too), but this will substantially change the profile of operations spike style overload against config shards, which means all of our 1 shard results may change (a single shard cluster in Atlas is one config shard by default since 8.0) and our 3 shard results (Atlas keeps a config shard until reaching 4 shards).

We should evaluate how big of a change this would be, since it could potentially require re-tuning all of our workloads.