An offline machine was breaking the graphs for my healthy machines
I ran into an annoying problem with my monitoring setup: when the local inference machine was off, the graphs for the rest of the lab started getting gaps too. The machines being graphed were fine.