During a recent outage of production, connections to an un-healthy mongo server became really slow but continued to work. This tied up gunicorn workers and resulted in an LMS/CMS outage. If possible we should lower the current response timeout for responses from mongo.
More details about the incident can be found here: https://openedx.atlassian.net/wiki/display/EdxOps/2016-04-12+Mongo+Instance+Became+Unhealthy
Determine the current timeout for mongo responses
If reasonable, reduce it further so that gunicorn workers recover faster.