I ran the repro script in
SERVER-48395 while sending mongod a SIGUSR2 every 10 seconds to collect stack traces. This resulted in network errors and failed operations at the application
accompanied by messages from the mongo shell like
The mongod logs tell a similar story:
The blue markers on the timeline show the points at which SIGUSR2 was received. These are accompanied by some number of connections ended (red curve) and a smaller number of connections accepted (blue curve), resulting in a net decrease of connections each time (green curve)
I wonder if we might not be re-trying network operations when they return the EINTR that would result from SIGUSR2.