Airbnb has a system with hundreds of servers (thousands, probably). They need to understand whether the servers are running OK, or are having strange behaviours, before these become catastrophic.

To track this, the servers take samples every 5 seconds of different measures: CPU used, memory, power consumption, free disk, network throughput, etc etc.

This data is sent from all servers to a system that stores all these samples to be able to alert the operators of the system, who can then take measure to remediate potential issues.

The article talks about how the changed this system from using a custom-made software stack, to using open standards and more shared code.

Let me know if it wasn’t clear enough, I can explain further