Boris,
Thanks, we’ll try the 2.12.0 driver and see if that helps.
If it helps anyone, after my last response on the thread we dug into AWS a bit more and “discovered” that all AWS DNS Resolvers have a hard limit at 1024 requests per second, and we were running 5-6 fairly request heavy services on a single instance. While our code is not reporting any DNS timeouts, it was happening across multiple services in the cluster when trying to reach Mongo. We are verifying this as the cause right now by updating connection strings to use the internal VPC IPs. If the problem goes away, we investigate a better/more resilient DNS solution (or instance level cache).
Documentation on AWS also notes that their Linux flavors do not provide any DNS caching by default.