While looking at FTDC for the insert workload from SERVER-31694, I noticed some interesting WT log statistics. Namely we are force writing the log slot almost as many times as we write a log record. MongoDB uses the in-memory mode for maximal consolidation and performance. There are a number of places where and reasons why we force out the current in-memory log records. We should investigate where these calls are coming from.
Most of these are effectively a no-op because in attempting to force out the log records we find we can skip it because there are no outstanding records to write.