Using the simple test app below, in addition to the attached jMeter script, the driver experiences significant performance degradation especially when initially scaling the connection pool.
The logic in pool.js has evolved many times and we've been constantly struggling with the driver to mantain an acceptable level of production traffic. The driver will fall apart when significant traffic spikes are seen (NewRelic graphs available) and this needs to get addressed.
Steps to reproduce:
1. Build and run node.js sample code below
2. Open jMeter and open 256 or 512 threads to localhost:3000
3. Execute test
4. Open POSTman or similar and execute a query to the server, observe a significant delay which gradually reduces once the box has reached load
5. In comparison, build .12 of the mongodb-core driver does NOT exhibit this extreme slowdown and is significantly faster