Im building a real-time *(5 second latency) analytics server. During load testing with ~600 concurrent connections I've noticed severe discrepancies between the number of known test documents and the number of reported documents. That is, I know there should have been 600 "currently active" records, but on find()s mongo is reporting much less.
This seems to be caused by the "write and forget" nature of mongo and that the default syncdelay is set to 60 seconds. Is this correct?
I believe that I can fix the issue either by using safe writes, or using a syncdelay of 5 seconds. Either will surely degrade performance, but I'm curious as to how each might affect performance.