From this GitHub PR

We started seeing a sudden spike in CPU usage once we switched to a certain mongodb+srv / sharding combination.

It turns out there is a bug in srv_polling.js that tries to access the wrong variable name (due to a capitalization error), which meant the rescan interval was undefined i.e. 0.

In our production system, this means our server performs millions of DNS requests per hour to our mongodb srv records (many thousand per second), which is costly and breaks server monitoring in various ways.